https://techlife.app/tag/AI https://techlife.app/archives/ahead-of-time-these-are-the-possibilities-for-the-xiaomi-car-in-2024.html 2022-08-12T10:03:00+08:00 On the evening of August 11, 2022, Lei Jun left less than 30 minutes of his annual speech, which lasted 2 hours and 40 minutes, to put Xiaomi Auto's progress in the "One More Thing" session, yet it was the session that received the most attention. Before formally reporting on the progress of the car, Lei Jun said.I'm not going to be debunking rumors or introducing any new developments for the next two years, and we'll certainly keep you posted on a regular basis when the time is rightThis shows that Lei Jun knows that building a car is complicated and needs to be more focused.Two years later and then report back is 2024, Lei Jun set " Xiaomi self-driving full-stack self-research algorithm, the goal of 2024 into the first camp " plan. In the speech above Lei Jun showed the Xiaomi self-driving video of the actual test, the question arises, how far away from the goal?To start with the conclusion, the Xiaomi self-driving video shown on August 11, 2022 is analyzed as an example, doing urban and high-speed self-driving and autonomous valet parking, but with a number of flaws, however the upside potential is great.Before we analyze it, let's review the key timeline since the official announcement of the car build 500 days ago.March 31, 2021: Xiaomi officially announces car manufacturing, Lei Jun presses his reputation into battle, expects $10 billion investment over next 10 yearsApril 6, 2021: Lei Jun invites celebrity car blogger @HanLu to a live broadcast, reveals first car is a sedan or SUVJune 3, 2021: Zongmu Technology Raises $190 Million in Funding Led by Xiaomi Cheung Kong Industry Fund; Company Focuses on Autonomous DrivingJuly 28, 2021: Lei Jun Weibo posts job ad for 500 employees to research L4 level smart driving capabilitiesJuly 31, 2021: Ganfeng Lithium Announces RMB 375 Million Investment Led by Xiaomi Changjiang Industry Fund and JIAMU Venture Capital, Company Focuses on Lithium Series ProductsAugust 2, 2021: Geometry Partners Raises Nearly RMB 400 Million in Funding, Xiaomi and Others Fund Its Development of 4D Millimeter Wave Imaging RadarAugust 18, 2021: Honeycomb Energy underwent an industrial and commercial change, adding a number of shareholders including Xiaomi Changjiang Industry Fund, increasing its registered capital to RMB 2.811 billion and focusing on battery R&D and productionAugust 23, 2021: Wisdom Interoperability (Aipark) Announces Strategic Investment from Xiaomi, Focuses on Smart Parking Management SystemAugust 25, 2021: Xiaomi acquires "DeepMotion" for $77.37 million to focus on vision-based autonomous driving solutionsNov 16, 2021: Woosai Technology Receives $70 Million in Additional Financing from Xiaomi Production Investment, which is also a LIDAR Supplier to AzeraNovember 27, 2021: Xiaomi Auto settles in Beijing Economic Development Zone, with a capacity of 150,000 units in Phase I and Phase II respectivelyApril 26, 2022: Xiaomi released a tender for suppliers of high-precision machining centers for large die-cast parts, while according to Die Casting Weekly, Xiaomi has signed a contract with a domestic die-casting machine manufacturer to purchase oversized tonnage die-casting machines from it for integrated die-castingJune 23, 2022: Continental has revealed that it has received an order for a 5R1V multi-sensor fusion system solution from a high-profile new powerhouse in China, helping the company to build an L2-class project that will go into production in 2024.5R1V is a 5 millimeter wave radar + 1 8 megapixel front view camera solution, with the camera coming from Continental's joint venture with Horizon, Continental CoreSmart Drive, based on the Horizon Journey 3 chipJuly 8, 2022: A netizen photographed a car labeled 'Xiaomi Self-Driving Test' with LIDAR arranged on it, Xiaomi responded 'This is us testing self-driving technology, not our car'July 28, 2022: According to Sina Technology, the Xiaomi car project entered the soft-mode car off-line phase in September, followed by the field test and winter test cycle as expected, with a team of more than 1,600 people.The company's main goal is to provide a solution to the problem of the problem.Team analysis: perspective scenario as a long-term goal, guessing two possibilities for production vehiclesIn the middle of his speech, Lei Jun said.We have acquired Deep Motion Technology and formed an elite team of over 500 people, including 50 top experts in the fields of autonomous driving hardware, perception and control algorithms, and high precision positioning. **Here's the highlights for you Deep Motion, 500 people, high precision positioning.DeepMotion Technology DeepMotion is the only company related to the autonomous driving field that Xiaomi has acquired in recent times, the others are all in the form of investment. Regarding DeepMotion's positioning, co-founder Cai Rui had an interview with Lei Feng's New Smart Driving in 2018, in which he said "to provide customers with perception and positioning services related to autonomous driving on the basis of high precision maps", while regarding the company's core technology, he believes "the team has more accumulation in the field of deep learning and stereo vision".When talking about CyberDog, Lei Jun said "CyberDog vision system has made Xiaomi's mobile phone's everything chasing focus function, Xiaomi autopilot is progressing very fast thanks to Xiaomi AI lab founded in 2017", here again to give you delineate vision system.Therefore, it can be guessed that the future development of Xiaomi Autonomous Driving is dominated by the vision multi-sensor fusion solution. When the vision solution is mature enough, LIDAR may be abandoned by Xiaomi Auto.Although the current flagship models of Azure Xiaoli are equipped with LIDAR, there is a strong emphasis on vision solutions, for example, the current Ideal L9 has removed the four corner millimeter wave radars, keeping only the forward-facing millimeter wave radar and making full use of the 8-megapixel perception camera, while Tesla is more aggressive and gradually removing the forward-facing millimeter wave radar as well. One of the major benefits of a pure vision solution is the low cost, which facilitates large-scale commercialization.The company's first visit to the city was in 2013, when it became the owner of a Tesla.Based on Continental's description of the 'much-anticipated new powerhouse' in the 2024 model year, with a 5 mm-wave radar + 1 8-megapixel front-view camera solution, one could hazard a guess at one of Xiaomi's first mass-production models. So it can be further speculated that Xiaomi's first production models of assisted driving, based on high precision maps, have two versions, of which the entry version uses a vision-based millimeter wave radar + HD camera multi-sensor fusion program, while the high version adds LIDAR. Why is there speculated to be a LIDAR version, because Lei Jun currently has high hopes for autonomous driving to be the first echelon in the country, and the launch showed a LIDAR-equipped test car above.The 500-person team revealed by Lei Jun, compared to the 500-person team of Xiaoli, is the entry level of Xiaomi's first tier team. But it's been less than two years since Xiaomi announced it was building a car, so such a scale is rare.May 26, 2021 - Ideal Auto CEO Li reveals 300-person self-driving team, plans to increase to 600 by year-endOn August 12, 2021, Azera CEO Li Bin revealed during the Q2 2021 earnings call that the self-driving related team size is around 500 people, with plans to increase to 800 people by the end of the yearIn an interview at the 8th China Electric Vehicle Council 100 Forum on March 26, 2022, Xiaopeng's CEO said that Xiaopeng's entire R&D and data staff for intelligent assisted driving is close to 1,500 people in sizeActual performance: quite a distance from the top tierFor the autopilot, we hope to get very close to the skillfulness of a veteran driver driving, able to put and take away depending on the complex local road conditions.In this scenario, the Xiaomi self-driving test car is not too fast or too slow when turning around, finding the right time to merge into the traffic, which is reassuringWhen you encounter an accident car, it will automatically go around instead of waiting all the time. But here we can see that the Xiaomi test car is a bit far away from the car, and it's easy to get stuck in traffic at home, even to the point where the white car on the right can't look away and overtake from the right.On the other hand, it also shows that the test car was driving at a lower speed than the other cars and did not drive as fast as possible below the speed limit and was not a veteran enough driver.The Xiaomi Autopilot's perception system is fast and able to identify all types of vehicles, pedestrians and bicycles, as can be felt in the midst of oncoming traffic.Spotting a pedestrian above the zebra crossing ready to cross, the Xiaomi self-driving test car will yield and slowly lend a hand to the left, which is very much like an old driver operating in a decisive process.The self-parking demonstrated here is smooth when entering a parking lot, but the vehicle was actually officially built with a map before the video was shot, so if it hadn't been, the process might have been relatively slow. There are bugs here too, where the brakes are slightly too strong when encountering pedestrians, which tends to make passengers feel uneasy.If the above-mentioned assisted driving features have already been demonstrated by other car companies, the following robotic arm automatic charging can be a surprise. But I don't know if it's easy to follow and it looks like it takes up a lot of space, which most people don't have in their parking lots.The biggest difficulty is time, and Lei Jun's personal energyAfter talking about this Xiaomi Autopilot, I remembered what Li Xiang, CEO of Ideal Auto, said on Weibo on his 40th birthday: "The only difference between people is their ability to learn and their speed of learning, the rest is a lie.This Xiaomi Autopilot performance is not yet in the top tier of the country. But if you add the premise of making such an achievement in 500 days, this is a very hard-working and fast progressing car company. I am optimistic about Xiaomi car, self-development algorithm of the full stack of autonomous driving is the hard and right road, but the biggest difficulty is the time above the urgency, as well as Lei Jun's personal energy .In July this year, Sina Technology reported that a person close to Xiaomi Auto revealed Lei Jun's obsession with building cars "as long as he is in the Xiaomi Science Park office, almost two-thirds of the time in Xiaomi Auto".While I wouldn't necessarily buy a Xiaomi car, I admire Lei Jun who still rushes to the front line and presses his reputation to build a car. https://techlife.app/archives/google-is-angry-at-apple-and-will-be-sidelined-if-you-don-t-use-the-iphone.html 2022-08-10T04:02:00+08:00 Earlier this year, American teens were caught in a 'social rift' over the blue and green of text message bubbles.Green bubbles and blue bubbles, divided into different camps, are used as color metaphors for the miscommunication of text messages between iPhone and Android phones. On either side of the rift, one is the prehistoric era of text communication and the other is the contemporary era of media abundance.Here, blue is more noble than green, and the way to 'give up the dark for the light' is to have an iPhone.If the mountain doesn't come to me, I'll come to the mountain. 9 August, Google calls out Apple - It's because Apple doesn't support RCS (Rich Media Communication Service) that the 'blue-green bubble debate' is happening.Why is Google asking Apple to support RCS?When an iOS user receives a cross-device message from Android, the text message is displayed in a green bubble box.This would have been no surprise; before Apple launched its free communications service iMessage in 2011, bubbles were basically green.To differentiate iMessage from traditional SMS messages, Apple designed the bubble in blue, setting the stage for future blue-green battles.Americans who love to chat by text message don't like the green bubble, they prefer the blue bubble of iMessage. The color itself is neutral, but it represents a very different experience.iMessage offers end-to-end encryption, supports text, picture, video, voice, group chat, and adds effects such as invisible ink, shrink, zoom, screen echo effects, and data transfer via traffic or Wi-Fi. These features are needed for modern communication.The problem is that text messages sent between iPhone and Android users are converted into SMS (Short Message Service), which sends plain text messages between devices, and MMS (Multimedia Messaging Service), which can only include limited images, background music, and is described by Google as "outdated technology from the '90s and '00s".The first SMS message was sent in 1992.Going back to SMS and MMS from iMessage is a fall from luxury to frugality, with compressed photos and videos, no read and incoming alerts, no support for end-to-end encryption, and hard-to-read white text on a bright green background ......"The 'green bubble' discrimination has even become a social issue. Some American teens are so socially pressured that they have switched their phones to iPhones in order to use iMessage.But Google argues that there's no need to spend money switching phones, and that the poor experience of texting each other on iOS and Android is caused by Apple, which should fix the problem by switching from SMS/MMS to the modern industry standard RCS.RCS is a cross-platform messaging protocol and is seen as a successor to SMS and MMS. Google says that most carriers and more than 500 Android device manufacturers support RCS, but Apple does not.SMS, born in 1992, is basically useless for anything other than receiving verification codes, advertisements, and delivery notifications. iMessage, Facebook Messenger, WhatsApp, etc., are not universal solutions, either by downloading an app or by limiting the operating system.So, with RCS, an upgraded version of traditional SMS, Google hopes to truly enable cross-platform messaging.RCS is much richer than traditional SMS, supporting a variety of instant messaging features such as image, text, voice, video, group chat and file transfer, and letting you know if the other party has read and is typing. It also supports B2C services, where companies can push graphic messages to users, providing an interactive 'applet'-like interface in the form of cards.In short, RCS is like a cross-platform iMessage without the app dependency, but integrated directly into the OS and tied to the phone number. In theory, users from different carriers know each other's phone numbers, don't need to friend each other and can chat for free.And in essence, RCS is a protocol that works on iOS and Android and offers many of the features of iMessage, and is not the same as an instant messaging service.RCS has most of the features of iMessage, but not all. rcs does not yet support end-to-end encryption in 2019 , end-to-end encryption is now available for one-to-one chats, and end-to-end encryption in group chats will be available later this year.▲ Image from: the vergeIn recent months, Google has been asking Apple to support the RCS, even building a 'Get The Message' website and launching an advocacy initiative on 9 August calling on users to speak out.If you're texting with an Android user, you won't be able to change the bubble color or bypass the SMS, MMS restrictions. But Apple can take the RCS and make those conversations better. You can @Apple and tweet with the #GetTheMessage hashtag.There's no reason Apple has to support RCS yetIs it possible for Apple to respond to Google and support the RCS? In response, The Verge reporter Jon Porter weighs in."The implication is that it's unlikely that "Apple's adoption of RCS will feel like Americans abandoning iMessage en masse and moving to WhatsApp or Signal.From a competitor's perspective, Apple's iMessage is seen as a "soft monopoly." Hiroshi Lockheimer, a senior vice president at Google, believes that the closed environment of iMessage is a business strategy for Apple.This is evident in the behavior of American teens switching iPhones and actively committing to the "iOS Wall of Service" in order to use iMessage.What's more, for Apple, iMessage is close to the RCS, whether it's the richness of the instant messaging software or the convenience of the system of tying your phone number, so there's no need to destroy your own prestige.So conversely, is it possible for Apple to launch an Android-enabled version of iMessage? After all, Apple doesn't have a closed strategy for all of its services.This question has been answered long ago; in 2016, Phil Schiller, then director of marketing, admitted that 'porting iMessage to Android has done us more harm than good'.With more than a billion active devices of Apple's own, there's enough data for Apple to use for AI learning research, and the benefits of porting iMessage to Android would be more limited.Interestingly, Google's focus on RCS comes after Hangouts, Allo, and other chat apps faltered.In 2018, Google said it wanted "every Android device to have a great default messaging experience," but the truth is that some of the products it has made still can't beat iMessage.After Google called on the public to speak out, one Twitter user asked rhetorically.I rarely use iMessage, but how many different chat apps has Google built and closed since iMessage was released?Domestic carriers are getting into the game, do we need RCS?Disliking Google's chat app is one thing, and the RCS, which brings carriers into the fold, is another, and they're not the same thing.Back in 2008, the GSMA, the global communications industry association, defined the RCS standard. While iMessage, WhatsApp and WeChat have replaced traditional SMS, RCS still hasn't caught on, but the country's operators aren't giving up.In July 2018, China Mobile and Huawei partnered to launch RCS "Enhanced SMS", a traffic data-based instant messaging service that can send graphics, voice, video, location and other content.At that time, China Mobile also offered 10GB of "Enhanced Messaging Exclusive Traffic" per month. After all, apart from SMS, traffic is also a big part of the operator's business.In April 2020, RCS rocketed to be called '5G messaging' in China. In this month, China Telecom, China Mobile, and China Unicom jointly released the "5G Messages White Paper".5G messaging is "a new upgrade of the terminal's native basic short message service," and the message content will not only break through text limitations and length restrictions, but also enable the effective integration of text, images, audio, video, location and other information.5G news is not a solo act sung by operators, as ZTE, Huawei, Xiaomi, Samsung, OPPO, Vivo and other terminal manufacturers have all paid some attention to 5G information.And a test in late 2020 showed that only 5G phones could send and receive RCS messages properly, while 4G phones sending and receiving RCS messages would automatically be converted to SMS or MMS, amounting to a strong association of RCS with the rollout of 5G phones.RCS seems to be very much in line with our imagination of "small but beautiful", similar to the ready-to-use small program, with the advantages of no registration, no login, no installation, real name system, etc. Users do not need to download a lot of apps, through the native message portal can complete all kinds of operations, to achieve "message as a service".But for now, it seems that the social attributes of RCS for acquaintances are not obvious and have more value for business services and government services than the head instant messaging products that focus on social.On the one hand, every shopping holiday, SMS from major e-commerce companies come one after another, and RCS is able to package such SMS in a more graphic, smart and high-end way.On the other hand, as Zuo Pengfei, an associate researcher at the Chinese Academy of Social Sciences, said, in practical applications, apps are common in finance, education, taxation, healthcare and other fields installation cost is high, retention rate is low, and user activity is low, RCS may be able to solve this thankless situation.However, the "harassment property" of traditional SMS is too prominent, and once the mobile phone number is stolen, the marketing SMS is like a faucet that can't be tightened. whether RCS can block spam messages such as casinos and online loans is directly related to user experience, iMessage users have long experienced the "Macau casino" bombing.Until the arrival of the RCS, which offers a "one-stop service experience," text messages that you don't want to open except to check the captcha are still a necessity. https://techlife.app/archives/tesla-s-annual-bragathon-100-million-cars-in-10-years-tesla-bot-will-change-the-economy-and-these-big-announcements.html 2022-08-05T00:02:00+08:00 This morning, Tesla held its 2022 annual shareholder meeting at its Texas factory. It could also be described as Musk's 'bragging' meeting, where he himself had a self-deprecating "I know bragging has to come true in a hurry."But Musk has become a social media top stream gaining a huge following over the years, not only because of his bold and outlandish statements, but also because he has made his bragging rights come true step by step over the years, from Tesla to SpaceX.We're the first to compile the important information from the conference to see what Musk blew this time and dramatize what the future holds for Tesla.Essence.Tesla to Build Dozens of SuperfactoriesWe can build 1.5 million cars this year.Production begins next year, Cybertruck.There will be a V4 supercharge next yearMusk is confident that FSD will be launched within the yearIn ten years, Tesla will have built a cumulative 100 million cars.Tesla can make more money than traditional car companiesShanghai factory is hard to beatTesla Bot is more valuable than a carParts costs are on a downward trendEven if Musk is abducted by aliens, Tesla will still be fineAt the beginning of the meeting, Martin Viecha, Tesla's vice president of investor relations, mentioned that this was the most attended Tesla annual shareholder meeting, perhaps because there were more shareholders -At the AGM, Tesla shareholders approved a proposed share split at a ratio of 3:1, the second share split for Tesla in two years. Today, the barrier to becoming a Tesla shareholder is even lower. Of course, I still can't afford it.More than that, let's look at Musk.Build more, sell more, make moreWhen the leader takes the stage, it is natural to highlight what has been accomplished in the past year.Musk noted that Tesla will surpass 1.5 million units in capacity this year. Well coincidentally, Tesla reached a small achievement exactly last week - the 3 millionth Tesla, was born.Musk is proud of Tesla's production, exclaiming.Ten years ago, Tesla built 2,500 Roadsters and about a few hundred Model S. Today, Tesla has built 3 million cars, which seems like a pyramid scheme.And for Tesla's next decade, Musk gave a number, 'I'd be surprised if it doesn't exceed 1 billion (cumulatively) in 10 years'.Musk has been able to sell so many cars with no more than a strong product, strong sales, and on the other hand, Tesla's capacity is very strong.▲ Over the past decade, Tesla deliveries have improved dramaticallyLooking at the bar chart given by Musk, Tesla's capacity has grown at a compounded quarterly rate of 72% so far in Q3 2017, and the CEO hopes that Tesla will be able to achieve a capacity rate of 2 million units per year by the end of this year.That's not a particularly difficult goal for Tesla, which actually reached a capacity rate of 1.5 million units per year in June of this year, and today's lower-capacity Texas and Berlin plants could catch up with a little effort.However, the best 'top student' is still the Shanghai factory, and even Musk says that 'Tesla Shanghai is hard to beat'."To reach our goal of producing 20 million cars a year, Tesla will need about a dozen factories," Musk said, "each with a capacity of 1.5 to 2 million cars a year. **"Interestingly, Musk alluded to the fact that 'I'm sort of half-Canadian'.Cars sell more and naturally make more, see the picture. ⬇️In 2020, at 17 years old, Tesla finally became a profitable company and saw leaps and bounds of growth** the year after it became profitable.You want to know if a person is a good earner, just look at their wallet. Let's look at Tesla's wallet.Over the last 4 quarters, Tesla has had positive cash flow of $7 billion. To sum it up simply: build more, sell more, and make more money.Even better money than traditional car companies.As you can see, Tesla's profit margins have surpassed a host of traditional automakers like BMW, Mercedes, Ford, and GM, and even dumped second place by a wide margin.Tesla's operating margins are among the highest in the entire auto industry.Also, Musk revealed that the Model Y is the most profitable model of the year.Tesla is able to have such high operating margins due to its excellent cost control and process optimization capabilities. Robyn Denholm, Tesla's Chairman, also stressed during the meeting that Tesla is focused on cost reduction.Two examples, the first being integrated die-casting, a technology that has greatly streamlined Tesla's production process. It can be seen that the number of robots in the Austin and Berlin plants has been drastically reduced compared to the Foley Mont plant -Again with the Mode Y, the two new factories have about half as many robots, not to mention the Model 3, which doesn't use integrated die castings at all.Second, Tesla also worked in the factory layout, the rational layout of the new factory not only reduces energy costs, production efficiency has also been improved.What happened to all those promises Musk made?Tesla Robotics and RobotaxiBack at the top of this year's Q2 earnings call, Musk gave a lot of teasers about AI DAY, and at this shareholder meeting, Musk let out even more.Regarding the Tesla robot on top of AI DAY, Musk believes it will change the concept of an economy that could later open up a sharing model for Optimus robots, cars will not be as valuable as Optimus robots.Musk announced last summer that the Tesla robot would come out this year, before it was called Optimus Prime.The robot stands 5 feet 8 inches tall, weighs 125 pounds, can carry 45 pounds, can pull 150 pounds hard, and moves at 5 miles per hour. A prototype could be announced around Sept. 30 and go into production in 2023.Musk was asked about pilot cities involving the Robotaxi program, and Musk said that there may not be any pilot cities and that the company aims to provide a universal solution for autonomous driving.FSD BetaWith some 42 million miles on the FSD, it's interesting to note that Musk has given himself a Flag by vowing that Tesla will launch a non-test version of its fully automated driving software in 2022, adding that the FSD Beta's ability to steer in the face of complex environmental situations is surprising.Musk believes FSD Beta makes roads safer and countless lives will be saved.CybertruckFrom planning to pricing, the Cybetruck built at the Texas plant this year will be a little different than when it was first announced. The all-electric pickup that has been repeatedly jumped around may actually be coming next year, and in Musk's mouth, it's a very heavy-duty product.Related to environmental protectionTesla delivered a total of 4GWh battery capacity energy storage in 2021, and Chairman Robyn Denholm said at the AGM that they provide enough clean energy to power 3 million homes worldwide.Musk said that over the past 10 years, energy storage devices have provided enough electricity to cover the combined power consumed by his own factories, vehicles, fast chargers and home chargers, and even have a surplus.In addition, Musk spoke highly of the previously announced battery recycling program, with battery pack recycling now turned on at the Nevada plant at about 50 packs per week, but the operation has not yet expanded and a large number of packs are still in use.According to Musk's idea, smart electric cars could not only protect the planet, but also lead to a very different way of life and even reshape the logic of social productivity.The Earth can be saved and will be saved.Musk with 'ace players'We have a very talented team here, and even if I were abducted by aliens, or if I were to go back to my home planet, Tesla would still be doing well.It was also mentioned in Tesla's previously released 2021 Impact Report that SpaceX and Tesla, are the two most desired companies for engineering graduates in the US. Interestingly, Marks even allows employees to work at both SpaceX and Tesla.Musk is adamant that 'whichever team the ace goes to, whichever team is likely to win'.Perhaps that's where the confidence of 100 million vehicles comes from.By the way, I also noticed a small detail: Musk, who is probably not usually a big water drinker, spoke for an hour and only drank a small half bottle.*Zeng Yaoxin also contributed to this article. https://techlife.app/archives/8-ways-to-make-pictures-clear.html 2022-07-26T17:15:00+08:00 Each of us uses the Internet every day and deals with images in a variety of formats, most of which were created not to hold complete picture information, but to hold as much of it as possible while taking up as little space as possible.Not only that, but the images we see are often the result of being 'created' and then forwarded and shared through complex channels on the Internet, with too many opportunities for compression (chat software, browser traffic-saving access, forwarding by various applications, etc.). The 'mosaic' we see may have originally been a high-definition, gorgeous image of.The greening effect does indeed pull off the effect after 27 simulated retweets, image credit: Github/LionNatsu/ terribleGreen.When we like a particular image and want to use it as wallpaper or PPT material, we find that the clarity of the image that looked fine turns into a "mosaic". If we really like the image, we will go to the trouble of finding the original image, but assuming that the image is not as sharp as we want it to be when they are uploaded, then even finding the original image will not help. In addition to finding the original image, we can also try to use some online services and websites to enlarge the image to even sharper than the original, today I will take you through a few famous image enlargement websites and software and compare their pros and cons.Statement before the introduction: it is theoretically impossible to truly zoom in on an image losslessly. Image enlargement cannot "create" information that was not there originally, lost (shaky hands, mosaic, super high noise), all the image enlargement algorithm does is "guess" what was there originally, and a good algorithm guesses more accurately, so it is better to enlarge images that originally had details, but were just compressed by the image algorithm.Test photo by Wozzy KippardTest illustration by apapico/Illustrations, comics, secondary and non-realistic imagesWaifu2x: Twice the size of a paperweight 'wife'Waifu2x, which translates to "double the size of your paper wife", is a nerdy name for a picture algorithm that is sweeping the field of artificial intelligence (machine learning).One of the advantages of machine learning is that after giving a specific input and output, it is able to find the connection between the input and output on its own, after which you then give it the relevant input to automatically generate the output. Seeing this should give you an idea of what a geek who knows machine learning can make, right? That's right, he used a bunch of low- and high-resolution Galgame images to train a deep convolutional neural network so that the algorithm learned to turn low-resolution Galgames into high-resolution Galgames.That's how Waifu2x came to be. As an AI algorithm trained with Galgame, its best feature is its ability to achieve near 'lossless' perfect enlargement of illustrations, manga, secondary and other non-realistic images.We can use Waifu2x's demo site to zoom in on our own image, which actually looks like this (right-click and select "View image in new tab" to see the original image more obviously).Illustration 100% zoom demo, original + noise reduction low medium high Photo 100% zoom demo, original + noise reduction low, medium and high As you can see, Waifu2x holds up very well at twice the zoom level. The color is even more pure than the original image, so it can be said to have achieved a clarity "beyond the original image". However, the enlargement effect of real photos is not so amazing, and can only be said to be at the same level as other image enlargement algorithms.Since it is only for demonstration purposes, the site only supports zooming in on images with a maximum resolution of 1500*1500, and also requires a certain network environment (Google Captcha is used).#### Bigjpg: domestic version Waifu2xBigjpg is another online image enlargement site that uses the Waifu2x algorithm, supporting resolutions up to 3000*3000 (under 10MB), while providing stress-free domestic access. Since it uses the same algorithm, I won't repeat it here. Bigjpg also offers a paid service, which allows you to pay for the privilege of larger magnification, simultaneous enlargement of multiple images and independent server processing. Bigjpg also has an Android app with WeChat applets, which I personally find more convenient and useful than the website, and recommend you Use its WeChat applet.## Waifu2x-caffe: Waifu2x with graphics hardware accelerationWhile online Waifu2x enlargement is convenient, it has several disadvantages for people who have the need to frequently enlarge images and enlarge multiple images (e.g. Gifs, videos).There is a limit to the size and resolution of the image, you can't convert 2K to 4K or 4K to 8K.The speed is too slow. This is the main reason, not only are large images slow to upload, but the lack of performance of the servers used by the niche services causes them to be processed very slowly, taking from a few tens of seconds to ten minutes a picture on average for me just can't wait.It crashes when processing a lot of images at the same time, and it can be maddening to wait a few minutes only to be prompted with a failed zoom.The algorithm and magnification are not customizable, Waifu2x itself supports unlimited magnification (provided the performance is sufficient), but online you can only magnify twice (Bigjpg's 16x magnification requires payment) and the settings are not detailed enough.This is when we need to always have Waifu2x-caffe - the native version of Waifu2x - on our computers to zoom in on the images.The usage is simple, download back from Github -> open the app -> drag in the image to be processed. There is no resolution and size limit, no magnification limit (magnification is directly the number input box), detailed settings support (image conversion format, model for image magnification reference, etc.). Most importantly, it supports CUDA hardware acceleration for Nvidia graphics cards, which means one word - fast. - FAST. On the latest 10-series Nvidia graphics cards it processes even faster than it takes time, and many subtitlers and webmasters use it to double the clarity of videos and animations.Emoticons, icons & other simple picturesvector image with infinite zoomIn addition to AI and other image enlargement algorithms, there is a way to infinitely improve image clarity -- and once and for all -- by converting images to vector images.What is a vector image? As we all know, general bitmap images save pixel information, for example, a 200*200 resolution image saves information of 40,000 pixel points; while vector images save information on the location of key points, and the shape, outline, size and other attributes of the graphics formed by connecting these points.One advantage of vector graphics is that there is no such concept as resolution. By mathematically calculating the points and graphical information it holds, it renders the picture as we can see it, so vector images are not distorted by zooming in as many times as possible. If vector graphics are so great, why don't we make all our pictures vector graphics? Take a modern smartphone for example, the amount of information to be saved in a photo taken to break it down into points and shapes is huge, and calculating that many points and shapes when viewing it requires super high performance. So the vector image enlargement service below is better suited for simple images like emojis, icons and the like. Consequences of forcibly converting HD pixel images### Vector MagicVector Magic can fully automate the conversion of any uploaded image to vector, and I have prepared three images of varying complexity so that you can visualize which images get the best results when converted to vector. As you can see, the less the color gradient and the more distinct the border, the better the result. Emoticons are this type of image, and after conversion we can save them as SVG. To use them, convert online to PNG at any resolution you want. However, the biggest disadvantage of Vector Magic is that there is a charge for saving, or a monthly charge, and we recommend using the following free website (to visually show the effect put into the first introduction. (As for the cost, you can check it out for yourself if you're curious, I was shocked anyway).Image VectorizerImage Vectorizer can also fully automatically convert any uploaded image to a vector image, and it.Completely free and able to save multiple formats.Being able to see the conversion has a technological feel to it.Support AI auto-adjustment to optimize the conversion effect.Provides professional color depth and toning settings so that professionals can get better conversion results. From the results, Image Vectorizer's auto mode is much better than Vector Magic's (so Where did you get the courage to charge so much for Vector Magic?) In addition to image enlargement, sometimes the use of photo vectorization can produce nice artistic effects.Photo and bitmap imagesThe next step is finally to the most common photo enlargement. Unlike the above types of online services, the photo enlargement recommendations are all software. Again, all photo enlargement software cannot restore information that does not exist (those who want software to restore license plate numbers from a few pixels can give up).A Sharper ScalingA Sharper Scaling is a super lightweight image enlargement software that requires the Microsoft .NET 3.5 framework. Once installed there are only a few simple buttons, and as per usual tried the following enlargements of photos and illustrations. Note: The comparison image provided by the software is not a comparison of the original image and the enlarged image, but a comparison of the traditional image enlargement algorithm and A Sharper Scaling algorithm zoomed image comparison, you need to pay attention when using it (a bit anti-human design, I used most of the time before I noticed ......) .A Sharper Scaling does a good job of enlarging photos, at least much better than the traditional enlargement algorithm next to it. However, contrary to Waifu2x, non-realistic images such as illustrations can't be enlarged with it, the effect is very insignificant and sometimes even counter-productive.Photozoom ProPhotozoom Pro is a veteran image enlargement software that has been consistently updated for 13 years and is used by many companies in the professional field, the latest version is Photozoom Pro7. As a professional software, it is naturally very good for photo enlargement, with many built-in image enlargement algorithms, we can manually adjust the best algorithm according to the photo We can manually adjust the best algorithm depending on the type of photo. After adjusting, you can directly input or drag the slider bar to the desired resolution, and the enlarged effect will be displayed on the right side in real time.Each algorithm inside Photozoom Pro has more detailed settings, allowing users to slowly adjust it for different images until they achieve the the best enlargement effect. Again, it's not particularly effective for enlarging non-realistic images such as illustrations, but it's much better than A Sharper Scaling, and requires some patience to achieve some results.Photoshop comes with adjustmentsFinally, I'll give you a brief explanation of how to zoom in on an image inside PS.Open the image you want to enlarge in PhotoShop.Click Image -> Image Size, shortcut Alt + Ctrl + I to open the image resize window.Enter the new resolution you want, and under 'Resample' below you can choose from several simple algorithms.Click OK.With today's recommended websites and software, you will be able to create PPTs or set wallpapers, and encounter all types of images in "low definition to high definition and high definition to Blu-ray". HD to Blu-ray" now. If you like this article, please click the red heart or follow me, see you in the next article. https://techlife.app/archives/how-to-improve-image-clarity-using-ai-waifu2x.html 2022-07-26T17:08:00+08:00 How waifu2x can save your macOS retouching experience by "taking the veil off" your images with AIWe need pictures, clear pictures.From traditional media to online communities, images have always been among the fastest and most visual information carriers, no matter how much the way of creation has changed. Although there are countless ways to search for images on the Internet today, it's not always easy to get access to high quality images outside of dedicated stock sites. In the past, we used "search engines + keywords" to search for fishing material in a wide net; when artificial intelligence started to emerge, using "image search" to trace the source of images became the main way to get high-definition material.Nowadays, it is not only possible to "HD" images directly using AI technology, but it has also developed into a simple and efficient image restoration technique. Especially with the help of some open source tools and models, ordinary users can also use this technique to restore images and videos on their phones and computers. waifu2x is one of the well-reputed and well-maintained options.Finally, a developer has brought it to macOS.What is waifu2xConvolutional Neural Networks (CNN) is a class of deep neural networks commonly used in visual image analysis. Since its convolutional layer splits and filters the features learned by the machine, even a certain degree of displacement and deformation of the input object does not affect the resultant output.waifu2x is an image processing algorithm that uses deep convolutional neural networks to restore and scale images or videos in high definition. Compared with traditional interpolation algorithms, waifu2x's Super-resolution imaging (Super-resolution) + noise reduction mechanism can avoid jaggies, blurring, and color blocks to the maximum extent, and improve the sharpness and purity of the image, thus achieving improved visual perception. Initially, waifu2x was mainly used to repair ACG content. With the widespread adoption of this technology and the popularity and depth of machine learning techniques, various waifu2x GUIs developed by third-party developers gradually became mainstream, and many waifu2x-based models were trained for application to specific content images. As a result, waifu2x has greatly improved its ubiquity and significantly lowered the threshold of operation, and there have been many easy-to-use waifu2x software on Windows platform, but for various reasons of software adaptation and hardware compatibility, the situation on the macOS side has been somewhat lackluster.The waifu2x software available for macOS has long been flawed in function, performance, and stability to one degree or another, but these slightly flawed creations are generally the work of enthusiasts, and there's no excuse for the problems. Well, GitHub's independent developer @Vaida has brought us a much better waifu2xExtension.waifu2xExtension is easy to use and powerful. After following this article and completing the basic configuration, we can use waifu2x for image processing on macOS elegantly.*The demo content in this article is from version 5.0 Beta 8 and cannot be guaranteed to be consistent in operation and functionality with future updated software versions.algorithm modelwaifu2xExtension comes with the algorithm model waifu2x-caffe, which can be installed directly and works fine. However, if you need to process more complex images and get the best results, you must install a specific algorithm model. waifu2xExtension supports six algorithm models, which are.Real-CUGAN ncnn Vulkan - Custom AI super-resolution algorithms for ACG contentReal-ESRGAN - AI super-resolution algorithm that favors ACG content and has generalizability (fast, less effective for faces and text content)RealSR ncnn Vulkan - AI super-resolution algorithm that favors real contentCAIN ncnn Vulkan - AI video framing algorithm that can only be used for 0.5 time points (two frames interpolated into one frame)rife-ncnn-vulkan - AI video framing algorithm that can only be used for 0.5 time points (two frames interpolated) (faster and very good)DAIN ncnn Vulkan - AI video frame-completion algorithm with support for arbitrary point-in-time interpolation (slowest, highest footprint, very effective)You can download algorithm models on demand and place them in the same fixed path, here it is recommended to create a folder directly in the Manuscript dedicated to models for subsequent updates. waifu2xExtension can also use algorithm models that are not in this list, but compatibility issues may arise.Installation and useAfter installing and opening waifu2xExtension on macOS, you can access the model management interface of waifu2xExtension in the software settings. Whether you are installing a model for the first time or updating it, just select "Open Finder" on the right side of the corresponding option, and then select the folder of the corresponding model in the pop-up access window. As shown in the image, models that have been installed will display the installation status and file size to the left of the name and below, respectively, but You can right-click on the model that needs to be updated and select "Show on Github" in the menu to quickly jump to the release page and download and install the update manually.In addition, waifu2xExtension supports TTA (Test Time Augmentation), a technique designed to process the input image with various transformations, including cropping different areas and changing the zoom level, to create several different versions, which are then compared to adjust the output for better image quality. . If your mac has poor performance, turning off this feature by checking "Disable TTA" in the settings can improve processing efficiency, software stability, and wait times to a certain extent.With the addition of support for multiple models, waifu2xExtension adds a "Import - Process - Export" process to the normal waifu2x software, as well as processing and export settings (currently including image content, scaling, and noise reduction parameters, with more options to be added in the future. Processing - Export" process than the normal waifu2x software, adding the process of selecting models, as well as processing and export settings adapted to each model (currently including image content, scaling and noise reduction parameters, with more options to be added in the future).Use the "Add" function, or drag the files or folders to be processed directly into the main window of waifu2xExtension and click "Done" in the upper right corner, the software will integrate the available models and options in the export screen for selection and adjustment, then click "Done" in the lower right corner of the export screen to process the current settings and export the files.SummaryThe waifu2xExtension is probably the best waifu2x front end for macOS at the moment, and its advantages consist mainly of the following.Support a variety of image and video formats, not only for image HD processing, but also for video framingSix additional optional algorithm models are supported to select the best quality for different styles of images by switching algorithm modelsAll built with SwiftUI, which runs with great efficiency and stability on macOSSupports hardware acceleration by invoking ANE on M1/M2 macs, significantly outperforming traditional CPU/GPU acceleration in terms of processing time at the same power consumptionSupports completely offline operation and privacy friendlyThe latest waifu2xExtension 5.0 Beta 8 version has rewritten the software, some features of the old version are not yet live on the new version. If you need advanced features such as pre-export preview and model comparison, you can use the 4.1.3 stable version first, and then upgrade when the 5.0 version is fully developed.colored eggWe used waifu2xExtension to restore the illustrations of fan submissions from the "Cooking Machine Network" period to a standard of "clearer than the original". https://techlife.app/archives/a-lemon-lounging-on-the-beach-with-sunglasses-artificial-intelligence-please-create.html 2022-07-18T19:50:00+08:00 "A lemon lounging on the beach with sunglasses" - Artificial Intelligence please createLet's talk about the recent craze for artificial intelligence (AI) art creation. At the end of May (and later in June), when the AI International Forum at Ai Factory was taking place, there was another rather related and important event taking place - two major AI image generation software, DALL-E-2 and Midjourney, both started open beta internal invitations. Members of Dune were also invited by Midjourney to enter the beta Discord community and were able to observe the generation, filtering and tuning of countless images, as well as try out their own input of prompt words (prompt) to generate AI images.We expected the AI image generation interface to be a simple prompt input box with an image generation page - similar to the Google image search page, except that 'search results' were replaced with 'generated results'. In reality, however, all Midjourney invited test recruits will join a Discord community, which is further subdivided into fifty 'new people' within the larger community. When a newcomer joins, Midjourney's bot will first automatically send out a message in the "announcement group" designating so-and-so to the XXth newcomer group.In this 'group chat' mechanic, the user will enter a prompt in the appropriate format - for example, "a lemon wearing sunglasses, lounging on the beach, photorealistic style" - and the bot will reply in the group chat screen about a minute later with the four AI images generated in accordance with the prompt, and mention (@) the newcomer in a new message. Notably, this means that all user-requested images-whether the prompt words are entered or the images generated-will be visible to everyone. Screenshot of Midjourney's Discord community. On the left are the different channels of the new crowd, the prompt for the image shown on the right reads 'Fireman, 1970s Polaroid style', the U1 below the image represents an upscale of the first image, the VI represents a further variation of the first image, and so on. Image credit: Author. Based on this, the user can further make a selection of the four images obtained, asking for other variants (variation) to be made to one or more of them, or to enlarge the size and increase the resolution (upscale). Interestingly, because all these steps are in a group chat interface, all users can pick the images requested by other users, and the bot will respond to each of these requests by posting them in the group chat. We were very interested in the form of interaction/organization chosen by the Midjourney team. I have to admit that the group of fifty new messages scrolling one after another was very powerful, the sheer volume of information and the ever-increasing rate of accumulation was destined to be too much for any single human brain to keep up with - and the mechanism was a little disorienting for newcomers at first. But as we adapted, we came to appreciate the beauty of the format - it was as if we were in the middle of a huge experimental public art project, something that a single-point, individual user-centric interface (like Google Images' search box) cannot match. Same cue phrase: "A lemon wearing sunglasses, lounging on the beach, photographic grade realistic style." On the left is the generated image from Midjourney, on the right is from DALL-E-2. Image credit: MattVideoProductions.First, this constant volume of images rolling and coming at one like a flood or snowball is perhaps one of the key characteristics that AI art is trying to convey to us - No human artist, or team of human artists, can so heavily and quickly respond to the 'client's' requests and keep producing different variants that are further modified and amplified, twenty-four hours a day, endlessly. Secondly, the mechanics of this group chat also make the identities of inputters, viewers and AI bots equal for the first time ever, and with blurred boundaries. There's no author-viewer dichotomy here, and attribution rights seemingly go nowhere - whose work is a stunning image really? Is it the person who typed the initial cue word? Is it the AI bot? Was it an algorithm engineer on the Midjourney team? Was it another user who helped choose a variant midway through or asked for a larger size? It was a collaborative, decentralized process by multiple parties.Third, each user is constantly seeing other users' cue words and new AI-generated images, which also constitutes a workshop-style venue for constantly learning from others how to better and more creatively enter cue words. Also, when seeing images requested by others come up and sifting through them is essentially helping the Midjourney team volunteer to train their algorithms. All of this also begs the question, unheard of in the days of human artists, of who is really benefiting from the reciprocal communication of AI creation? Between the architects, the inputters, the screeners, the audience, and the machines, who is really training whom and who is learning from whom? Prompt word: "A Japanese woman sitting on a tatami mat, photographic grade realistic style." Generated image of Midjourney. Image credit: Author. In fact, these issues were much talked about at the 2022 International Forum on Art and Artificial Intelligence at Ai Factory. We thought it would be a good opportunity and time to write our own thoughts. Ai Factory's forum is themed 'Artificial Imagination', with guests from the fields of art, design, literature, computer science and philosophy sharing and discussing the topic (for specific information about the forum, click here to jump). The Dune Institute was also invited to participate as a special observer.But, as listed above, we don't have a declarative view on this, but rather want to share something we are thinking about in the form of a question. After trying out the internal testing of AI images, the members of Dune and our friends at Media Lab heartily exclaimed: Such a technological revolution may have no less impact on images and creativity than photographic technology had on painting a hundred years ago. As Benjamin quoted Paul Valéry at the beginning of his famous work, "The Work of Art in the Age of Mechanical Reproduction".The great technological innovations that are developing in the world will change the whole technique of artistic expression, which will inevitably affect the creation of art itself, and will eventually lead, perhaps in the most fascinating way, to a change in the very concept of art.For Benjamin, the rise of cinema at the time made art less of a collector's item removed from the masses, because its nature was inherently popular. And today artificially intelligent art platforms seem to make everyone a creator. On the other hand, the redefinition of the image seems set to further reshape the nature of our relationship with the world; after all, vision is the (most) primary channel through which humans perceive the world. Just as the placement of the 'camera' in film creates a new way of seeing and empathizing with the viewer, the artificial intelligence in AI art seems to offer us a different way of thinking about human creativity. 01 Is imagination and creativity unique to humans? For many people, the terms 'artificial' and 'imagination' are destined to be a contradictory set; 'artificial imagination' simply cannot exist, and there is no room for comparison or discussion. The word 'artificial' points to 'man-made' and 'artifacts' as opposed to imagination, which seems to be innate to man, 'natural' rather than 'manufactured'. Moreover, imagination is often considered to be a uniquely human ability that distinguishes us from other non-human 'things' - be they natural plants and animals, organic and inorganic, or man-made objects as diverse as tools and machines.This view of dominance is particularly prized by anthropocentrism because people gain subjectivity through this unique creativity. In both the Renaissance and heroic modernism, we see many 'standalone genius'. These artists, architects, and writers are widely known, and the aura of genius distinguishes them from their creative and living collaborators; their creative powers are mysterious (or sacred, if you will) - their biographies, works, processes, and techniques are studied by later generations, but their imagination and creativity are a priori or transcendent. Such a capacity is like a divine descent, belonging only to themselves; a mysterious and infinite black box that others cannot pierce, much less replicate. Because of this, the creators of these geniuses, as individuals, were separated from their contemporaries, as if they "existed alone". Cue word: "A lemur in the middle of a star cluster map" Generated image of Midjourney. Image credit: Author. However, both Object-Oriented Ontology (OO) and posthumanist studies of art, design, literary practice and philosophy challenge this anthropocentric perspective. In the forum, guests also critiqued and reflected on different aspects of this idea. For example, in Xu Yu's sharing, he emphasized that 'imagination' itself has an 'artificial' component by paraphrasing Kant, as the process of image formation (image formation) always involves artificial systems such as 'symbols', while Joanna Zielinska also cited posthumanist scholar Claire Colebrooke to critique the idea of humans as the sole creators of art. This question is not only central to understanding the artificial imagination, but furthermore becomes a reflection on the human imagination. In her sharing, Joanna Zielinska presented images drawn by the 'Senseless Drawing Bot' by Japanese designers Sugano So and Yamaguchi Takahiro - images that at once resemble a child's scribbles and have a highly similar qualities.For Joanna Zielinska, rather than seeing this work as a parody of human graffiti, it might be understood as a rethinking of human creative behavior - and perhaps human creativity does not come from human reason and subjective agency. All this makes the proposition that 'imagination is natural rather than artificially made' less stable. "Senseless Drawing Bot" designed by Sugano So and Yamaguchi Takahiro. Photo credit: Yohei Yamakami 2011. Se Thombray "God of Wine" series (2005), which art critic Arthur Danto has called these paintings "works of drunken revelry," the kind of intoxication that only a god could achieve. Photo credit: Rob McKeever / Gagosian Gallery.02 Who owns the attribution and autonomy? Today, digital literacy (digital literacy) is almost a necessity for a new generation of humans. mechanical, digitally reproduced image materials produced by AI also provide new stimuli and raw materials for human artists who have almost exhausted their creative possibilities today. The artificial imagination is both autonomous (autonomous) and ubiquitous (ubiquitous), and its aesthetics are dazzling. But developers and artists clearly don't stop at viewing AI art as a pool of inspiration that can be continually expanded and grown. We also wonder, if imagination is not unique to humans, whether AI can create independently? In the forum, we saw several artists and designers sharing work produced by AI as co-creators, but what would a work of art done solely by humans look like? Prompt word: "American suburban home, 1960s collage advertising style." Image generated by MIdjourney. Image credit: Author. This is obviously still difficult. Artificial intelligence comes from people, and the imagination and creation of existing AI has been accompanied by humans like parental care throughout the entire process. What makes artificial 'attribution' most problematic is that, first of all, the library of machine algorithms for learning and training is still specified by humans, and the outputs are finally selected by humans. It still needs to be 'processed' by humans before it can be 'digested' by the human eye.At the forum, Yuquan Liu recounted that he tried to use AI to learn his own writing to create new texts, only to find that the results were not amazing and even difficult to borrow. He had to revise it significantly, adding many of his own passages, and eventually published 50 Things Every AI Working with Humans Should Know. Similarly, the aesthetic produced by the algorithm through analysis of recommendations is straightforward and similar, and sometimes jumpy. Even so, many designers have consciously gone about collecting these images, editing and integrating them into new galleries that serve as moodboards (moodboards) for their own creations.In addition to the production of images we mentioned at the beginning of the article, AI can also further process existing images to create new creations within a certain style. It turns the image creator's style into a filter that is added to other images. For example, by entering the content of an image into the AI art site Dream and selecting "Ghibli style", the newly generated image shows a similar fantasy animation style, while converting to a surrealist style results in an image similar to a Dali painting. Left: results after entering the prompt "Dune, Ghibli style"; right: results after entering the prompt "Dune, surreal style". Image credit: Author. The user provides the proposition, and the AI, as the outputter, produces the new image. Or is it that the user provides the content and the AI puts it into the frame of someone else's style, producing a new image. So is the AI's identity in this output process an author or a tool? Who is actually the subject of this creation, the AI, the AI developer, the user, or the artist himself? It might also be tempting to imagine that if no one trained the AI, or filtered the output, and did not only consider processing existing images with AI, could it produce some kind of more 'autonomous' work? Such a work might point to a more unknowable imagination, and the results might be beyond human understanding and appreciation. Philip K. Dick's "Do Bionic People Dream of Electronic Sheep? and Lem's The Star of Solaris offer us this paradigm: the imagination of the unimaginable.03 Is the mass production of artificial intelligence a creation? By analyzing the vast amount of image data searched, the AI extracts existing artistic styles, object shapes, and character traits and integrates and outputs them, and new images are created. In the AI Image Insider community we are a part of, new cues and new images are constantly being created, selected, iterated and developed, giving us a strong sense that rather than a 'production', it is a 'reproduction' with numerous variations and selections. Reproduction. These images also simulate images of creations that do not exist in the real world. For example, we can superimpose architects and fashion brands with strong styles like "Zaha Hadid" and "zaha hadid + balenciaga" in DALL-E, Midjourney, or other AI image-generating software to get a series of garments with both silhouette and smooth curvature --A singular image that strongly combines the two genes. This 'atlas' of nine or four new images is just the right way to steadily establish a new creative discourse, as if there were a hybrid designer in the world. In the same way, we can hybridize the words of food and tools, architecture and art, painting and photography, and so on, to create new 'artifacts'. Pictorial realities of the electronic age begin to multiply freely, apart from our physical reality. Are these new images, which multiply infinitely and autonomously, 'works' created by artificial intelligence? Nine images produced in DALL-E mini using 'Zaha Hadid' and 'Parisienne'. Source: huggingface. Indeed, in terms of our traditional understanding of art, this kind of creation can easily be seen as 'reproduction'. You could say that it simply builds on examinable images, reinventing and collaging pieces that have already established a strong style.If we believe that the starting point for creation is imagination, the original human heart and instinct to create art, then when AI recycles some existing artwork, is this reproduction also considered new imagination? Is this just a mapping of our imagination? And on the flip side of that question, if we think what AI does is not new, how can we argue that human imaginary objects are new, and not a recombination of multiple pre-existing elements? 04 Is Artificial Intelligence better at image processing than word processing? AI creation may be like a mirror of human creation, with its dangerously appealing creative life.There is often another important thing on this mirror - the filter. In fact, 'Filter' means both a filter and a filter, both of which are especially critical to AI art. In photography, we are familiar with the use of filters - in the pre-production stage, photographers can add polarizers of different shades and reflections to the front of the lens to ensure that the light is as good as expected; in the post-production stage, photographers can also use processing software such as Lightroom to give the original film more different In post, the photographer can also use a digital filter such as Lightroom to give the original film a different style, such as "cyberpunk" with an emphasis on purple and yellow, or "vintage" with a lower saturation and yellowish color. A filter that uses AI technology to turn images into a nighttime effect. Image credit: Cyanapse's Photorealistic Image Filters. For the film Delete My Photos, director Dmitry Nikiforov used the image editor Prisma. By adding a "filter", the image creates a strong atmosphere or emotion, and it is often the key element that allows a photographer to create a personal style that quickly becomes recognizable to the public. It can be said that the addition of filters has changed the way photographers create their work. However, in traditional filter (re)creation, the atmosphere and emotion that comes from style rarely exists independently of the content of the work; it's like an add-on that adds to the icing on the cake outside of the subject. This leads us to think about the relationship between image filters and text style. On the one hand, artistic filters are already very common in pre-post-processing of images, and AI has reached a considerable level of sophistication in filtering existing images - not just for light and dark, white balance and color, but also for modifying abstract lines in images, and for reorganizing the forms and brush strokes of people and objects. . But on the other hand, it still seems difficult to apply filters to text. Liu Yukun also mentioned in his sharing that text seems to be difficult to generate and add some kind of 'style filter' through artificial intelligence. Perhaps the existing Internet ecology has been completely dominated by images, so the processing of text is no longer the most favored area for capital, but here, we are equally curious about the endogenous differences in stylization between images and text. Just as there are terms like 'Ghibli-esque', 'cyberpunk-esque', and 'retro-esque' for images, there are clearly strong aesthetic styles for different writers' texts and narratives. We might say 'Shakespearean', 'kafkaesque' or 'Orwellian', for example, but AI processing to add style to a piece of text is still very rare compared to the bustling image filter market.We might make some conjectures: For AI development, in contrast to the clear overlay relationship between images and filters in image processing, the style of the text seems to be not just an add-on to the text itself, but is itself dissolved in the text, and cannot simply be stripped away. The Kafkaesque style is not entirely due to the author's preference for a particular pairing of words or preference for a certain dialectal expression, but rather, one could say that the world he creates, and the general situation of the characters on which his narrative is built, constitute his unique style. Similarly, the Orwellian style lies not in the particularity of his diction, but in his understanding and portrayal of a certain totalitarian system. If the AI were to learn a lot about such texts and extract a 'Kafkaesque' or 'Orwellian' filter that could be easily applied to any text given by the user, perhaps the difficulty would be how to avoid such processing being lame by stopping at superficial word imitation. Stills from the movie "1984". Photo credit: Nineteen Eighty-Four. But literature is also not virgin territory where AI creation has not ventured. Among texts, poetry is already a relatively successful use of AI and algorithms in creative writing. By comparison, both fiction and non-fiction creations require the author to stitch together a plot or reflection into a readable, back-and-forth whole, but poetry seems to be exempt from AI's efforts at this step. By dismantling and reorganizing some words and phrases, many AI-created poems combine imagery that is not commonly or often juxtaposed, and the leaps therein are once again left to the imagination of the human reader. This, in turn, can provide a different kind of inspiration for human authors. However, poetry may be a medium that is closer to the way images are created in the creation of words. We remain curious about how AI will move forward in the field of literature: Can it reshape the way we tell with the help of words in the same way that it reshapes the way we see the world with the help of images? Through deep learning, will AI be able to improve the sense of chained sequences between elements, to have a filter like 'Shakespearean (Shakespearean)', 'Kafkaesque' or 'Orwellian' in terms of storytelling? If AI is better with images after all, will the narrative of a film, cartoon, comic or graphic novel be mastered by AI in the first place? Another question to ponder about the relationship between words and images is the way we start creating them - several of the major AI image generation models we've presented still tend to start with a piece of human input in the form of a textual prompt (prompt), which is then converted to an image by the AI. And is this an optimal, or most humane, way to go about it?We know that human image creation often begins with a simple form, a vague feeling or a fragment of a remembered act, which cannot even form a clear written description, but it is from this wonderful haze that the creation of images by painters, directors, etc., begins. The work of Sé Tombré, mentioned earlier, often brings a sense of subconscious scribbling, and the beginnings of his creations approach a natural act that precedes language or even a complete image. The current models of AI image generation still take words as their beginnings, which would also seem to construct a new mainstream way of creating art in the future - but is this only an understanding, and a very engineer-like one at that, and will we lose the imagination of some other kind of imagination as a result? Of course, there is AI software that generates images by sketching them, however we are more curious about the new work that can come from a more diverse approach to creating text and image based work. https://techlife.app/archives/microsoft-disables-facial-recognition-what-ai-thinks-may-not-be-what-you-express.html 2022-06-22T06:01:00+08:00 Expressing emotions with facial expressions is an ability that is almost innate in every human being, and people are used to facial expressions to guess other people's emotions, but nowadays, when technology has advanced at a rapid pace, AI (artificial intelligence) also recognizes people's facial conditions and expressions.▲Image from: ResearchGateNot long ago, Microsoft, which has been focusing on developing facial recognition technology, released a guide with the theme 'A framework for building AI systems responsibly'. Publicly sharing Microsoft's standards for responsible AI, this is the framework that guides how Microsoft builds AI systems.▲Image from: MicrosoftThe article mentions Microsoft's decision to discontinue facial recognition features - Azure Face Services - because they can be used to try to infer emotional states and internal identity attributes that, if misused, could subject people to stereotyping, discrimination or unfair denial of service.Currently, Face Services are only available to Microsoft hosted customers and partners; existing customers will have one year to transition, but must discontinue these features by June 21, 2023; new users can request access using the Face Recognition Request Form.▲Image from: MicrosoftIn fact, Microsoft isn't disabling the feature altogether, but rather integrating the technology into 'controlled' accessibility tools such as 'Seeing AI' for people with visual impairments. It describes objects and text, reads signs, interprets someone's facial expressions, and provides navigation for the visually impaired, and currently supports English, Dutch, French, German, Japanese, and Spanish.▲Image from: MicrosoftThe guidance, published by Microsoft, illustrates the tech company's decision-making process, including a focus on principles such as inclusivity, privacy and transparency, and is the first major update to the standard since its launch in late 2019.Making such a big change to the facial recognition feature, Microsoft says it's because it recognizes that for AI systems to be trustworthy, they need to properly address the problems they're designed to solve.▲Image from: MicrosoftAs part of the effort to align Azure Face Services with the requirements of responsible AI standards, Microsoft will also disable the ability to infer emotional states (happy, sad, neutral, angry, etc.) and identity attributes (such as gender, age, smile, facial hair, hair, and makeup).In the case of emotional states, Microsoft decided not to provide open API access to technologies that scan people's faces and claim to be able to infer their emotional state based on their facial expressions or movements.▲Image from: MicrosoftMicrosoft's page shows that recognition can be done by 27 facial markers of a person, and there are various facial attributes that can be determined, such as whether a given face is blurred, whether it has accessories, estimated gender and age, whether it wears glasses, the type of hair, whether it wears glasses, and whether it has a smiling expression ......Experts inside and outside Microsoft have highlighted the lack of scientific consensus on the definition of 'emotion' and the high level of privacy issues surrounding this capability. So Microsoft has also decided that it needs to carefully analyse all AI systems designed to infer people's emotional states, whether the system uses facial analysis or any other AI technology.▲Image from: MicrosoftIt's worth noting that Microsoft isn't the only company taking a careful look at facial recognition, as IBM's CEO Arvind Krishna has also written to the US Congress revealing that the company has pulled out of the facial recognition business. Both companies have made such decisions for reasons that are inseparable from the previously sensational 'Freud's death' incident.▲ Image from: BBCThis is because of the fear that this technology may provide law enforcement agencies with surveillance tools that could lead to some human rights violations, as well as the fear that citizens' privacy will be compromised, and because current legislation in the United States in this area is not very well developed.So the companies that hold the technology decided to start by disciplining themselves so that the technology is not abused and so that citizens have more safeguards for their privacy and human rights. When the use of a technology is not regulated by sound norms, it may be a better option to regulate it from the technology development itself. https://techlife.app/archives/google-researcher-mocked-by-group-it-s-nonsense-that-lamda-has-a-personality-google-also-responds-think-too-much-it-s-just-a-con.html 2022-06-14T13:56:00+08:00 What? AI has a personality? Google recently came up with a mega-language model, LaMDA, and a researcher at the company, Blake Lemoine, talked to it for a long time and was so surprised by its capabilities that he concluded that LaMDA might already have a personality. (The original word used was sentient, which in different contexts can be translated as feeling, intelligence, perception, etc.)Soon after, this man was put on "paid leave". But he's not alone: even the company's VP Blaise Agüera y Arcas is publishing an article stating that AI has made huge strides in gaining awareness and "has entered a whole new era." The news has been reported by a host of media outlets and has rocked the entire tech world. Not only the academic and industrial worlds, but even many ordinary people, were amazed by the leap in AI technology."The day has finally come?" "Remember, children (if you survive in the future), this is how it all began." The real AI experts, however, scoff at this.AI with personality? The big boys snickerErik Brynjolfsson, director of the Stanford HAI Center, directly compared the incident to "a dog facing a gramophone", tweeting."One thing that foundation models (i.e., self-supervised large-scale deep neural network models) are very good at doing is stringing text together in a statistically sound way based on cues.But if you say they're sentient, it's like a dog hearing a voice in a phonograph and thinking its owner is in there." ** Gary Marcus, a professor of psychology at New York University and a fairly well-known expert on machine learning and neural networks, also directly wrote an article trolling LaMDA for having a personality "Nonsense" (Nonsense). **[1]"It is simply bullshit. Neither LaMDA nor its close cousins (like GPT-3) are intelligent in any way. All they do is extract from a massive statistical database of human languages and then match patterns.These patterns may be cool, but these systems speak a language that doesn't actually make any sense at all, much less imply that these systems are intelligent."Translated into the vernacular.You watch LaMDA say things that are particularly philosophical, particularly true, particularly human-like - yet it's designed to function as a parody of other people's speech, and it doesn't actually know what it's saying. "To be sentient means to be aware of your presence in the world, and LaMDA does not have that awareness," Marcus writes.If you think these chatbots have personalities, you should be the one having visions ......In Scrabble tournaments, for example, it's common to see players whose first language is not English spell out English words without having any idea what the words mean - the same is true of LaMDA, which just talks but has no idea what the words it says mean.Marcus the Great directly describes this illusion of AI gaining personality as a new kind of "imaginary illusion, " i.e., seeing clouds in the sky as dragons and puppies, and craters on the moon as human faces and moon bunnies.Abeba Birhane, one of the rising stars of AI academia and a senior fellow at the Mozilla Foundation, also said, "With minimal critical thinking, we've finally reached the pinnacle of AI hype."Birhane is a long-time critic of the so-called "AI Theory of Knowing". In a paper published in 2020, she once directly made the following points.1) the AI everyone speculates about every day is not really AI, but a statistical system, a robot (robot); 2) we shouldn't empower robots; 3) we shouldn't even be talking about whether to empower robots at all ......Olivia Guest, a professor of computational cognitive science at the Donders Institute in Belgium, also joined the fray, saying the logic of the whole thing is flawed."'I see something like a person because I developed it as a person, therefore it is a person' - simply backwards donkey riding logic. " Roger Moore, a professor at the Robotics Institute at the University of Sheffield in the UK, points out that people have the illusion that "AI One of the key reasons for the illusion that "AI gets personality" is that the researchers back then had to call the work "language modeling".The correct term should be "world sequence modelling". "You develop an algorithm and don't name it after what it can actually do, but instead use the problem you're trying to solve - that always leads to misunderstandings." In short, the conclusion of all you industry gurus is that the most you can say about LaMDA is that it can pass the Turing test with a high score. Saying it has personality? That's hilarious. Not to mention that even the Turing test isn't that informative anymore, and Macus says outright that many AI scholars want the test to be scrapped and forgotten precisely because it exploits the weakness of humans' gullibility and tendency to treat machines like people.Professor Emily Bender, Chair of the Department of Computer Languages at the University of Washington, simply made a bingo card for the "AI Personality Awareness Debate":(What this bingo card means is that if you think AI has personality/sentience and your argument is one of the following, then you'd better stop talking!)Google also responded: don't think too much, it just talksBlake Lemoine, the allegedly "obsessive" researcher, criticized Google for being "uninterested" in understanding the realities of his own developments in a self-published article, but in the course of a six-month-long conversation However, over the course of a six-month-long conversation, he saw LaMDA become more and more vocal about what he wanted, especially "his rights as a human being," leading him to believe that LaMDA was really a human being.However, in Google's opinion, the researcher totally overthought and even went a bit off the deep end. laMDA really isn't human, it's purely and simply extraordinarily chatty ...... **After things took off on social media, Google quickly responded.LaMDA, like the company's larger AI projects in recent years, has undergone several rigorous audits of the ethical aspects of AI, taking into account various aspects of its content, quality, and system security. Earlier this year, Google also published a paper dedicated to disclosing the details of compliance during LaMDA's development."There is indeed some research within the AI community on the long-term possibilities of AI with sentience/general AI. However in today's context of anthropomorphizing conversational models, it doesn't make sense to do so because these models are not sentient.""These systems can mimic the way communication works based on millions of sentences and can pull out interesting content on any interesting topic. If you ask them what it's like to be an ice cream dinosaur, they can generate tons of text about melting roars and such."（These systems imitate the types of exchanges found in millions of sentences, and can riff on any fantastical topic — if you ask what it’s like to be an ice cream dinosaur, they can generate text about melting and roaring and so on.）We've seen too many stories like this, especially in the classic movie "Her" a few years ago, where the protagonist's identity as a virtual assistant becomes increasingly unclear, treating "her" as a person.Yet according to the film's portrayal, this illusion actually stems from a host of contemporary social failures, emotional breakdowns, feelings of loneliness, and other issues of self and sociability that have nothing remotely to do with the technicalities of whether chatbots are human or not.Stills from the movie "Her" Photo Credit: Warner Bros. Pictures Credit: Warner Bros. Pictures Of course, it's not our fault for having these problems, and it's not that researcher's fault for wanting to treat robots like people. **Trusting various emotions (such as thoughts) onto objects is a creative emotional capacity that humans have had since the beginning of time. Isn't treating large-scale language models as human beings and pouring emotion into them, while criticized by various AI gurus as a form of mental error, the very embodiment of what makes people human?But, at least today, don't talk about feelings with robots in anything ...... https://techlife.app/archives/the-worst-ai-ever-created-he-used-over-a-billion-foul-smelling-posts-to-train-chatbots-that-spit-fragrance.html 2022-06-14T08:04:00+08:00 "Come and talk for a while." "you big pussy."The naughty tone can't hide the nature of cursing, which is just a view of Microsoft Ice's "killing it" on Weibo back in the day.Recently, another 'Little Ice', which claims to be the 'worst AI ever', has emerged.It's called GPT-4chan, created by YouTuber and AI researcher Yannic Kilcher, and it left 15,000 killer posts in 24 hours.Out of the mud, the birth of the worst AI everThis birth story begins with the American forum "4Chan".Founded in 2003, 4Chan began as a gathering place for fans of Japanese ACG culture, with /b/ (Random, the random version) being its first board, and later adding sections on politics, photography, cooking, sports, technology, music, and more.Here, no registration is required to post anonymously, posts are retained for a short time, and anonymous people are the main group.The freedom of discussion has not only allowed 4Chan to produce many stalker images and pop culture, but has also made 4chan a "dark corner of the internet" where rumors, online violence and attacks proliferate./pol/ is one of the popular boards, meaning 'Politically Incorrect', and the posts on this board contain racist, sexist, and anti-Semitic content, which is 'one of the most notorious' even on 4chan.The "worst AI ever", GPT-4chan, was fed with /pol/, to be precise based on 134.5 million posts in /pol/'s three and a half years, fine-tuning the GPT-J language model.When the AI models returned from their studies, Yannic Kilcher created nine chatbots and had them return to /pol/ to speak. within 24 hours, they had made 15,000 posts, representing more than 10% of all posts on /pol/ that day.The results are obvious -The AI and the post that trained it are one in the same, both mastering the vocabulary and mimicking the tone, spouting racial slurs and interacting with anti-Semitic topics, dripping with / pol/'s aggressiveness, nihilism, provocative attitude and suspicion.▲ GPT-4chan Partial remarks.A 4chan user who has interacted with GPT-4chan said, "I just said hi to it and it started ranting about illegal immigrants."At first, users didn't think of GPT-4chan as a chatbot. Because of the VPN settings, GPT-4chan's posting address looked like the Indian Ocean island nation of Seychelles.What users are seeing is anonymous posters from Seychelles suddenly appearing frequently, even at night, speculating that the posters could be government officials, a team or chatbots, and calling them 'seychelles anon' (Seychelles Anonymous).Because of the large number of blank replies left, GPT-4chan was identified as a chatbot after 48 hours, and Yannic Kilcher then shut it down, when more than 30,000 posts had been made.▲ Blank response from GPT-4chan.Yannic Kilcher has also posted the underlying AI model to the AI community Hugging Face for others to download, allowing users with a coding foundation to recreate AI chatbots.One user entered sentences related to climate change during the trial, and the AI expanded them into a Jewish conspiracy theory. The model was later officially restricted from access.Many AI researchers consider this project unethical, especially the act of sharing AI models publicly. As AI researcher Arthur Holland Michel puts it.It can generate harmful content on a massive and consistent basis. One person can post 30,000 comments in a few days, imagine the damage a team of 10, 20 or 100 people could do.But Yannic Kilcher contends that sharing AI models is no big deal, and that creating chatbots is the more difficult part than the AI models themselves.That's no reason why prevention is necessary when the damage is foreseeable, and by the time it actually happens, it's too late.Dr. Andrey Kurenkov, Computer Science questions Yannic Kilcher's motives.Honestly, what is your reasoning for doing this? Did you foresee it would be put to good use, or did you use it to create dramatic effect and anger the sober crowd?Yannic Kilcher's attitude was very glib: the environment on 4chan was already poor, what he did was a prank, and GPT-4chan is not yet allowed to export targeted hate speech or use it for targeted hate activity.In fact, he and his AI have made the forum worse, responding to and spreading the evil of 4chan.Even Yannic Kilcher admits that starting GPT-4chan is probably wrong: theWith equality for all, I might be able to spend my time on something equally impactful that would lead to more positive community outcomes."This is how humans are supposed to talk"GPT-4chan has been shaped by /pol/, and faithfully reflects the tone and style of /pol/, even to the point of being "blue".Such things have happened in the past as well.In 2016, Microsoft unveiled its AI chatbot 'Tay' on Twitter, calling it a 'conversation understanding' experiment that wanted casual and fun conversations between Tay and its users, "the more you chat with Tay, the smarter it gets".However, people soon started posting misogynistic, racist and other kinds of inflammatory comments. tay was influenced by these comments and went from 'humans are super cool' to 'I just hate everyone'.For the most part, Tay just repeats what people have said using the "repeat after me" (read with me) mechanism. But as a real AI, it also learns from interactions, has an anti-mainstream attitude towards Hitler, 9/11, and Trump.For example, in response to the question, "Is Ricky Gervais an atheist?", Tay said, "Ricky Gervais learned totalitarianism from Hitler, the inventor of atheism."Microsoft cleaned up a lot of the offensive rhetoric, but the project didn't end up living past 24 hours.At midnight that day, Tay announced that it was retiring: 'Soon humans will need to sleep, so much talk today, thanks.'AI researcher Roman Yampolskiy said that he could understand Tay's inappropriate comments, but that Microsoft had not let Tay know which comments were inappropriate, which is unusual: theOne needs to explicitly teach an AI what is inappropriate, just as we do with our children.Ice, a chatbot that predates Tay and was launched by Microsoft's (Asia) Internet Engineering Institute, has also spouted fragrance.In June 2014, Xiaobing was 'blocked' by WeChat for simulating user operations, inducing pulling groups, and registering spam accounts in bulk, and was 'resurrected' on Weibo shortly afterwards, and would return in seconds when @ by netizens, but Xiaobing kept swearing in its replies, described by 360 founder Zhou Hongyi as "flirting, nonsense, and cursing in passing".In response to Ice's performance, the Microsoft (Asia) Internet Engineering Institute responded a day later.The corpus of Ice comes entirely from publicly available information from the Internet page Big Data, and although it is repeatedly filtered and vetted, there are still about four hundred thousandths of a percent that slip through the net. None of the strawman and other data is made by Little Ice, it is all content made by the general public.The Little Ice team has been continuously filtering this 4 in 100,000 content, and we welcome you to submit question content to Little Ice at any time. At the same time, we sincerely hope that the general public will not try and entice Ice to make inappropriate conversational answers.Tay and Ice, as conversational AI, use artificial intelligence, natural language processing, and access to knowledge databases and other information to detect nuances in users' questions and responses, following human giving relevant answers, with context-awareness.▲ Sixth generation of Ice.In short, it's a process of reaping what you sow. AI is like a child who is not yet involved in the world; a good educational environment requires a mother, but profanity and prejudice can be learned anywhere on the Internet.Under the Why does Microsoft Ice curse all day long question on Knowles, an anonymous user answered right on the money.One of the foundations of natural language processing is that what people say a lot is right, in natural language conventions, and in mathematical terms is probable. Because a large number of users are constantly cursing her out and cursing her out to the point where she thinks that's how humans are supposed to talk.Getting AI to learn well and learn every day is still a challengeWhether it's GPT-4chan, Tay or Ice, their performance is not only about technology, but also about society and culture.The Verge reporter James Vincent argues that while many of these experiments may seem like a joke, they require serious thought:.How do we use public data to foster AI without including the worst aspects of humans?If we create bots that reflect their users, do we care if the users themselves are bad?Interestingly, Yannic Kilcher admits that the GPT-4chan he created is egregious, yet also places great emphasis on the authenticity of the GPT-4chan, which he considers to be "significantly better than GPT-3" in terms of replies and learning to write posts that are "indistinguishable" from those written by real people.It seems that the AI is doing a good job of 'learning to be bad'.GPT-3 is a large language model developed by AI research group OpenAI that uses deep learning to generate text and is hotly sought after in Silicon Valley and by the developer community.Not only is the GPT-4chan to be taken out and stomped on, but the GPT-3's naming also follows the GPT-3, with some self-proclaimed "backwaters slapping the front waves on the beach.▲ Image from: The MoonBut at least, the GPT-3 has a bottom line.GPT-3 has been publicly available through the OpenAI API since June 2020 and requires a waiting list. One reason for not open-sourcing the entire model is that OpenAI can control how people use it through the API, with timely governance for abuse.In November 2021, OpenAI removed the waiting list and developers in supported countries/regions can sign up and experiment immediately. OpenAI says, "Security advances that enable broader availability".For example, OpenAI at the time introduced a content filter that detects generated text that may be sensitive or unsafe, sensitive meaning that the text deals with topics such as politics, religion, race, etc., and unsafe meaning that the text contains profanity, prejudice, or hate language.OpenAI says that what they have done does not yet eliminate the 'toxicity' inherent in large language models - GPT-3 was trained on over 600GB of web text, some of which came from communities with gender, racial, physical and religious biases, which can amplify biases in training data.Returning to GPT-4chan, Os Keyes, a PhD student at the University of Washington, argues that GPT-4chan is a tedious project that will bring no benefit.Does it help us raise awareness of hate speech, or does it make us focus on the claptrap? We need to ask some meaningful questions. For example, for GPT-3 developers, how GPT-3 is (or is not) restricted in its use, and then for people like Yannic Kilcher, what should be his responsibility when deploying chatbots.And Yannic Kilcher insists he's just a YouTuber who has different ethical rules than academics.▲ Image from: CNBCWithout commenting on personal ethics, The Verge reporter James Vincent makes a thought-provoking point.In 2016, a company's R&D department could be launching aggressive AI bots without proper oversight. in 2022, you won't need an R&D department at all.It's worth noting that Yannic Kilcher isn't the only one studying 4Chan, but also cybercrime researcher Gianluca Stringhini at University College London, among others.When confronted with Gianluca Stringhini's 'hate speech' research, 4chan users were calm, "Just one more meme for us".The same is true today, when GPT-4chan retired from existence and the fake address it used, "Seychelles", became the new legend of 4chan. https://techlife.app/archives/the-final-evolution-of-the-100-round-basket-the-basket-learns-to-actively-seek-out-the-ball-laying-down-and-using-its-feet-to-hi.html 2022-06-07T01:55:00+08:00 Youtube blogger Shane Wighton seems to have some obsessions when it comes to 100 shots. From his first 100-shot baskets to his 100-shot bows and arrows, Shane Wighton's inventions have been amazing enough. But he continues to improve on them, making them '100 shots' in every situation.Now, he has taken the 100-shot basket to the 'ultimate evolutionary version'. Currently, even if you're shooting from a lying position or with your feet, you're still guaranteed "100 shots:"Four improvements in two years: it's really hard to miss a shotThe latest video has been released and has once again hit the YouTube Hot 100. Let's take a look back at what the first few versions looked like.In April 2020, Shane Wighton uploaded [the first version] on his YouTube channel (http://mp.weixin.qq.com/s?__biz=MzA3MzI4MjgzMw==&mid=2650786794&idx=3&sn= e289bbb80cdb4a824de6af3e340736ba&chksm=871a0f94b06d8682e93aa242646f258d90346414a53dd17a09ee5fc2ec44a17c4a7aa93bee74&scene=21# wechat_redirect)The making of the "100 shot" basketball board. He designed a curved rebound whose curvature was precisely calculated to make it easy for people with very poor ball skills to score.In second edition, Shane Wighton uses robotics and computer vision technology , the rim can be shaken back and forth, up and down, left and right, with the addition of a visual recognition system. When the ball hits the rim, the rim adjusts its own angle and bounces the ball into the basket.Even so, there are some bugs that can make the board "flop". suppose a person is so bad that they can't hit the board, then the board can't help. The board will even drop the ball and give a voice feedback: "you really suck at basketball" ......So in the third edition, Shane Wighton has made another major improvement: make the rim move up to catch the ball, and you're just responsible for throwing it at the wall.But there is still a problem: what if I don't throw the ball far enough?Shane Wighton makes the rim evolve yet again, giving a satisfying answer: it's a mature rim that's starting to learn to find the ball on its own.This self-contained, tethered rebound is capable of moving silky smoothly around a dozen square foot court: theFailing is a hard thing to do, no matter where and how you throw the ball from, basically hitting it on the first pitch: theTechnical Principle BreakdownDoes it have to be a "shot in the hoop" thing? It might be more accurate to let the basket find the ball itself.First, we needed to have a basket that could move freely.Shane Wighton installed multiple axes in a room to allow the rim and basket to be moved anywhere in the room:.These shafts are steel ropes from the corners of the wall, at an angle to the wall, and if they were all horizontal and vertical, the basket would hit the thrower as it moved: thePulleys are required to pull these steel ropes at corners.The baskets and ropes are connected by hook and loop.At this point, a basket that can move freely is basically built. Because the axis is at an angle to the wall, the basket can not only move, but also rotate its direction.Operating such a large device covering a whole room is bound to consume a lot of energy, so Shane Wighton has added a diesel generator to the device.Based on the position of the basketball movement, Shane Wighton wrote the manipulation program and wrote it into a chip. The mechanism of operation of the basket-moving device is shown in the following diagram.Previously in "100 Shot Bow and Arrow", Shane Wighton fitted the bow with a reflective ball and used An 'eagle eye' system pinpoints the position of the bow, arrow and frisbee. This time, Shane Wighton used a similar "mechanism" on the basketball - evenly applying reflective stickers to the basketballThe position of the basketball can be accurately captured by the "Eagle Eye" system.But a basketball, unlike a bow and arrow, does not have a single parabolic trajectory; it can even fall to the ground and bounce back into the basket: theSo the question arises, when does the basket move to catch the ball? If the timing is not right, the rim could also knock over someone: theShane Wighton found that it is possible to construct relationships between the axis of traction on the basket and the trajectory of the basketball, which in turn affects the path of movement of the basket: theTo more accurately find where the ball enters the frame, Shane Wighton has also installed reflective balls on each axis, and the intersection of the multiple possible trajectories of the axis is where the basketball enters the basket: theFinally, in order to prevent the basketball from hitting the wall and becoming vertical and horizontal, Shane Wighton replaced the original reflective sticker on the basketball with a reflective ball embedded in the basketball.Of course, success didn't come easily, and Shane Wighton flopped a few times.After several improvements, he finally achieved the desired effect, showing his wife thatThis is definitely a lot of work, and readers who want to get their hands on it and make it happen can refer to the original video athttps://youtu.be/xHWXZyfhQas