Lenovo announces a $2,345 FHD smart display for video calls

Scientists have published the most accurate map of the distribution of matter in the Universe. This incredible achievement gives insights into the evolution of the cosmos as a whole and it suggests that the Standard Model of Cosmology is missing some ingredients to explain what we are seeing today.

The matter that makes stars, planets, humans, and capybaras is just 5 percent of the matter-energy content of the universe. According to our best understanding, 25 percent is made by an invisible substance we call dark matter. The remaining 70 percent is dark energy, which is responsible for the accelerated expansion of the universe.

We have no direct measurement of dark matter or dark energy. What we know about them comes from the study of galaxies near and (mostly) far and by the measurements from the cosmic microwave background, the first light that moved freely in the universe 380,000 years after the Big Bang.

This latest work employed those two focuses too. The Dark Energy Survey measured galaxies' properties to work out the expansion rate of the Universe over cosmic history, and the South Pole Telescope measured the cosmic microwave background. Both datasets were used to measure gravitational lensing. Matter, both regular and dark, warps spacetime creating lensing effects. So by measuring the lensing, you can work out where the matter is. And using the two different instruments you get a more precise map.  

“It functions like a cross-check, so it becomes a much more robust measurement than if you just used one or the other,” one of the lead authors Chihway Chang, from the University of Chicago, said in a statement.

The analysis was then compared to the Standard Model of cosmology, which incorporates the accepted physical laws and the presumed values of dark matter, dark energy, and regular matter. The model can be used to create simulations that show how the universe has evolved from 13.8 billion years ago to today. And those simulations look slightly different from the observations.

“It seems like there are slightly less fluctuations in the current universe, than we would predict assuming our standard cosmological model anchored to the early universe,” said analysis coauthor and University of Hawaii astrophysicist Eric Baxter.

The work suggests that the matter in the universe is less clumpy: it just does not cluster as much as the theory predicts. This is not the only problem with the model. A tension has emerged with the expansion rate of the universe in recent years.

“I think this exercise showed both the challenges and benefits of doing these kinds of analyses,” Chang said. “There’s a lot of new things you can do when you combine these different angles of looking at the universe.”

Current and future observatories are collecting more data and this incredible map will be refined again and again. Hopefully, an answer about the properties of dark matter (and dark energy) will emerge from this kind of work.

The findings were published in three articles in the journal Physical Review D.

Source

"The AI Text Classifier is a fine-tuned GPT model that predicts how likely it is that a piece of text was generated by AI from a variety of sources, such as ChatGPT," explains a new OpenAI blog post.

OpenAI released the tool today after numerous universities and K-12 school districts banned the company's popular ChatGPT AI chatbot due to its ability to complete students' homework, such as writing book reports and essays, and even finishing programming assignments.

According to BusinessInsider, ChatGPT is banned in NYC, Seattle, Los Angeles, and Baltimore K-12 public school districts, with universities in France and India also banning the platform from school computers.

BleepingComputer tested OpenAI's new AI text classifier and, for the most part, found it to be fairly inconclusive.

When testing OpenAI's AI text classifier against most of our own content, it correctly determined a human wrote our articles. 

OpenAI's AI text classifier response for BleepingComputer content

 

However, when analyzing content generated by ChatGPT and You.com's AI chatbot, it had a lot of difficulties detecting if the text was AI-generated.

As educators will likely use the new AI Text Classifiers to check if students cheated on their homework assignments, OpenAI warns that it should not be used as the "sole piece of evidence" for determining academic dishonesty.

"Our classifier is not fully reliable," warns OpenAI.

"In our evaluations on a 'challenge set' of English texts, our classifier correctly identifies 26% of AI-written text (true positives) as 'likely AI-written,' while incorrectly labeling human-written text as AI-written 9% of the time (false positives)."

"Our classifier's reliability typically improves as the length of the input text increases"

The classifier's success will likely improve as time passes and is trained with further data. For now, though, it is not a reliable tool for detecting AI-generated content.

Source

Discord is slowing down some Nvidia graphics cards

So far, none of the game console companies have confirmed or denied this report. If true, it would be a huge loss for E3 2023. This will be the first E3 to have an in-person event since 2019. E3 2020 was canceled entirely due to the Covid-19 pandemic threat. E3 2021 was held as an all-virtual event, while E3 2022 was also canceled completely.

In a statement to IGN, E3's organizers ReedPop, who took over the running of the trade show on behalf of the Entertainment Software Association (ESA) did not comment on the reports of the game console businesses pulling out of this year's show. It did say that it had sent out invites to attend the show across the video game industry at the start of January 2023 and that it had a lot of interest from many top game companies. The statement said it was confident that the final E3 2023 exhibitor list will "make the trip to Los Angeles well worth it for the industry and consumers alike."

Sony had previously decided not to attend the last in-person E3 event in 2019, so the report would seem to confirm the company's continued decision not to attend the show. The official lack of participation from Nintendo and especially Microsoft is more surprising as both companies have been highly supportive of the show in the past.

In a recent interview with IGN, Microsoft's Xbox head Phil Spencer did confirm that the company will be holding an in-person games showcase event in Los Angeles around the time of E3 2023. However, he stopped short of saying the company will participate in E3 itself. He did add that Microsoft is committed to the ESA trade organization and will work with the group "to help make the E3 successful."

This year, E3 will have some competition. Summer Games Fest, which has been a virtual showcase of upcoming games for the past few years, will hold its first physical event on June 8 at the YouTube Theater in LA. It's possible that one or more of the console companies could exhibit at that event rather than wait for E3 just a few days later.

Microsoft, Sony, and Nintendo may all bypass attending E3 2023

In a Twitter thread, (via BGR) Paul Buchheit, the creator of Gmail, offers this prediction:

Google may be only a year or two away from total disruption. AI will eliminate the Search Engine Result Page, which is where they make most of their money.

He predicts that typing in the typical search bar could, instead of sending you to a Google search engine page, simply use an AI-based auto-complete feature that will give you an answer to your question. That answer may also include a URL link to a website for more info.

While an AI bot like ChatGPT will still use the search engine backend at Google, The company's massive ad business could be shut down if the AI doesn't send traffic to the search page itself. Buchheit compares what might happen to Google to what happened to the print Yellow Pages, which became obsolete very quickly with the rise of the Internet.

Microsoft already has big plans to use ChatGPT. Last week, the company announced a new multi-year partnership with OpenAI that will allow ChatGPT to access Microsoft's Azure cloud servers. At the same time, Microsoft will expand the use of ChatGPT on many of its consumer and enterprise products.

Source: Twitter via BGR

The creator of Gmail says AI could disrupt Google's business within two years

Much like almost everybody else who has an internet connection, Amazon employees have been using ChatGPT for a variety of reasons, both personal and professional. Some Amazon employees have been using the AI chatbot to ask questions about everyday tasks at the job and to help them do research that may help them do their jobs better. So far, this doesn’t sound like too much of a problem for an employer, but this is where things get interesting because of how these types of tools work.

Basically, Amazon is worried about its employees sharing sensitive corporate data with the chatbot when asking questions. This is because these types of models are constantly iterating on themselves based on the new training data that feeds them. Just in case you weren’t aware, the newest training data going into ChatGPT comes from the text prompts themselves that users are putting in when talking with the bot.

In fact, the warning from Amazon originates from one of its corporate employees who actually saw ChatGPT mimicking some of Amazon’s internal data. That isn’t to say the tool recreated the data, but it was able to mimic it in a way that appeared convincing enough for the attorney to sound the alarm about possible data breaches occurring as text prompts.

So, there you have it. Amazon has warned users against using ChatGPT for help with research to help them carry out the day-to-day tasks associated with their jobs. It isn’t because of some overly controlling urge from the company or its billionaire tech boss but is legitimately based on fears of internal and sensitive data falling into third-party hands.

Why Amazon doesn’t want its employees talking to ChatGPT

In a support post on the game's official forums, the developers stated this new update would not actually change the game's software version number. It will remain at version 1.61. Another support post describes how to enable Hardware-Accelerated GPU Scheduling to add DLSS 3 support for PCs running on Windows 10. You can also check out our own article on this subject. Windows 11 PC owners don't have to worry; this feature is enabled by default.

According to a report on The Verge, enabling DLSS 3 on Cyberpunk 2077 will increase frame rates on the game by over 50 percent with Psycho ray tracing, max settings, and DLSS 2 quality features turned on. The game's developers have previously announced plans for yet another update, Overdrive, that will improve the visuals of the game's ray tracing, but a release date has yet to be revealed.

CD Projekt RED announced in September 2022 that Cyberpunk 2077 had sold over 20 million copies since its rather rocky launch in late 2020. An expansion pack for the game, Phantom Liberty, has been announced but currently does not have a release date. A full sequel to Cyberpunk 2077 is also confirmed to be in the works, although we would think it is likely many years away from being released.

Cyberpunk 2077 on Windows gets hardware GPU scheduling, NVIDIA DLSS 3 and Reflex update

Samsung’s Q4 profits plummet 69 percent, hit 8-year low

When AMD released AMD Ryzen 7 5800X3D, it completely changed how CPUs could process gaming and other things. The reason was simple, it came with an extra layer of 64MB cache stacked on top of the usual 32MB L3 cache. AMD connected these cache chips with a through-silicon via (TSV) connector, which allowed these chips to have a very small amount of space between each other.

The results were extremely impressive. AMD Ryzen 7 5800X3D was among the fastest gaming CPU of the generation. Beating other CPUs which were far faster, expensive and powerful in raw performance. Only and only because of its extra L3 cache, which many games would efficiently utilize. So impressive were the sales and results, that AMD decided to release three new chips in the AMD Ryzen 7000 CPU series. The AMD Ryzen 7950X3D, 7900X3D and 7800X3D.

Looking at the results, AMD might have thought, why not implement the same technology in its graphics cards. Looks like AMD not only thought that, but it seems they almost implemented it.

AMD Radeon RX 7000 Almost Had 3D V-Cache

Tom’s Hardware explains that a senior semiconductor expert, Tom Wassick has revealed something really interesting. When he opened up the AMD Radeon RX 7900 XT graphics card and did some imaging tests, he found two important things.

First thing he found that while AMD Radeon RX 7900 XT came with 6 cache chiplets, only 5 were active and one was just an empty silicon. Those unaware, AMD Radeon RX 7000 series comes with a new chiplet design, where cache and graphics processor are different chiplets and connected to each other forming a single GPU, if one can call that. While it was known from the start that RX 7900 XT comes with only 5 active cache chiplets, unlike the RX 7900 XTX which comes with 6, what’s next is the most interesting part.

The second thing he found was that these caches came with the same TSV connection points which AMD used in it’s X3D CPUs like Ryzen 5800X3D and others. The pitch too was the same. He wonders if AMD was considering to include stacked 3D V-Cache on the Radeon RX 7000 graphics cards too.

This massive findings bring a big question to the fore. Why AMD did implement its base but did not implement the actual 3D V-Cache on the cards. It also brings in the question whether AMD could still bring this stacked cache in current generation of graphics cards.

Reason Why AMD Radeon RX 7000 Didn’t Implement 3D V-Cache

For the answer to above, one has to look at the tweet by known Twitter based leaker @Kepler_L2. This reply is something most sites haven’t reported on.

Kepler_L2 reveals that this implementation of 3D V-Cache by AMD was known from a year. But he says that it will likely never come out because of the Navi 31, that is GPU used in AMD Radeon RX 7900 XTX and RX 7900 XT is not reaching its true performance abilities.

In Kepler_L2’s reply to another Twitter user, he explains it further that the GPU used in these cards is too slow to benefit much from increased cache. Hence, AMD never implemented it.

It is unknown if AMD still wants to use this stacked 3D V-Cache on current or next-gen Radeon graphics cards. But if AMD does implement it, the performance benefits could be something to eagerly look at.

AMD Radeon RX 7000 Graphics Cards Almost Came With 3D Cache

According to a report by the Financial Times, Twitter has begun applying for regulatory licenses to become a payment platform in the United States. Esther Crawford, Twitter's director of product management, has also been developing the infrastructure of the new service.

This development is a part of Musk's plan to turn Twitter into an "everything app." Back in November, the business mogul floated the idea of Twitter becoming a place to shop for goods, transfer money to other Twitter users, and withdraw balances to an authenticated bank account.

In a previous pitch deck to investors when Musk was looking into acquiring Twitter, the business mogul said that he aimed for Twitter to amass about $1.3 billion in payments revenues by 2028.

In line with the license applications, Twitter is also moving ahead with regulatory checks and has applied as well for the necessary state licenses.

According to people familiar with the matter, Twitter's planned payment platform will initially work with regular payment methods and currencies. However, the company plans to support cryptocurrencies in the future.

Source: Financial Times

Twitter pushes forward with plan to turn itself into a payment platform

In a post on the Chrome OS Beta Tester Community Site (via Chrome Unboxed), Google said it would soon start testing a new Microsoft 365 setup for Chromebook owners. It will be available first on the Chrome OS Dev and Beta channels before its official release, which will happen sometime in the months ahead.

So what's going to be different? Currently, Chromebook users with Microsoft 365 have to install the Office Progressive Web App from the Office.com website. However, this new installation procedure doesn't look like it will require a website visit. Rather, Chromebook owners will get a more integrated way to install the Progressive Web App. In addition, users can connect their Chromebook's Files directory to their OneDrive cloud servers. The new system will move those Chromebook Files content into OneDrive when users launch Microsoft 365.

This sounds great, and the only real problem with this integration is that it won't be available to the general public for a while. Also, Google didn't really have to make this kind of effort. Its own Google Docs and Drive online products are direct competitors to Microsoft 365 and OneDrive. The fact that it is making a special effort shows that Google knows many Chromebook users like using Microsoft 365, which could spur more purchases of Chromebooks by business users.

Microsoft 365 recently added a new and cheaper $1.99/month subscription tier. It includes, among other features, access to Outlook without ads on mobile apps and its website, along with 100GB of OneDrive cloud space.

Source: Google via Chrome Unboxed

Chromebook owners will get easier access to Microsoft 365 and OneDrive

It comes down to stress – mechanical stress to be more precise – especially during potent recharging.

“Just modest indentation, bending or twisting of the batteries can cause nanoscopic fissures in the materials to open and lithium to intrude into the solid electrolyte causing it to short circuit,” explained senior author William Chueh, an associate professor of materials science and engineering in the School of Engineering, and of energy sciences and engineering in the new Stanford Doerr School of Sustainability.

“Even dust or other impurities introduced in manufacturing can generate enough stress to cause failure,” said Chueh, who directed the research with Wendy Gu, an assistant professor of mechanical engineering.

This artist’s rendition shows one probe bending from applied pressure, causing a fracture in the solid electrolyte, which is filling with lithium. On the right, the probe is not pressing against the electrolyte and the lithium plates on the ceramic surface, as desired. Credit: Cube3D

 

The problem of failing solid electrolytes is not new and many have studied the phenomenon. Theories abound as to what exactly is the cause. Some say the unintended flow of electrons is to blame, while others point to chemistry. Yet others theorize different forces are at play.

In a study published today (January 30) in the journal Nature Energy, co-lead authors Geoff McConohy, Xin Xu, and Teng Cui explain in rigorous, statistically significant experiments how nanoscale defects and mechanical stress cause solid electrolytes to fail.

Scientists around the world trying to develop new, solid electrolyte rechargeable batteries can design around the problem or even turn the discovery to their advantage, as much of this Stanford team is now researching. Energy-dense, fast-charging, non-flammable lithium metal batteries that last a long time could overcome the main barriers to the widespread use of electric vehicles, among numerous other benefits.

Statistical significance

Many of today’s leading solid electrolytes are ceramic. They enable fast transport of lithium ions and physically separate the two electrodes that store energy. Most importantly, they are fireproof. But, like ceramics in our homes, they can develop tiny cracks on their surface.

The researchers demonstrated through more than 60 experiments that ceramics are often imbued with nanoscopic cracks, dents, and fissures, many less than 20 nanometers wide. (A sheet of paper is about 100,000 nanometers thick.) During fast charging, Chueh and team say, these inherent fractures open, allowing lithium to intrude.

A scanning electron microscopy video that shows lithium plating as it takes place on a solid electrolyte. Credit: Xin Xu, Geoff McConohy and Wenfang Shi

 

In each experiment, the researchers applied an electrical probe to a solid electrolyte, creating a miniature battery, and used an electron microscope to observe fast charging in real time. Subsequently, they used an ion beam as a scalpel to understand why the lithium collects on the surface of the ceramic in some locations, as desired, while in other spots it begins to burrow, deeper and deeper, until the lithium bridges across the solid electrolyte, creating a short circuit.

The difference is pressure. When the electrical probe merely touches the surface of the electrolyte, lithium gathers beautifully atop the electrolyte even when the battery is charged in less than one minute. However, when the probe presses into the ceramic electrolyte, mimicking the mechanical stresses of indentation, bending, and twisting, it is more probable that the battery short circuits.

Theory into practice

A real-world solid-state battery is made of layers upon layers of cathode-electrolyte-anode sheets stacked one atop another. The electrolyte’s role is to physically separate the cathode from the anode, yet allow lithium ions to travel freely between the two. If cathode and anode touch or are connected electrically in any way, as by a tunnel of metallic lithium, a short circuit occurs.

As Chueh and team show, even a subtle bend, slight twist, or speck of dust caught between the electrolyte and the lithium anode will cause imperceptible crevices.

“Given the opportunity to burrow into the electrolyte, the lithium will eventually snake its way through, connecting the cathode and anode,” said McConohy, who completed his doctorate last year working in Chueh’s lab and now works in industry. “When that happens, the battery fails.”

Co-lead authors of the new study, from left, Xin Yu, Teng Cui and Geoff McConohy sitting in front of the focused ion beam/scanning electron microscope used for this research. Credit: Xin Xu

 

The new understanding was demonstrated repeatedly, the researchers said. They recorded video of the process using scanning electron microscopes – the very same microscopes that were unable to see the nascent fissures in the pure untested electrolyte.

It’s a little like the way a pothole appears in otherwise perfect pavement, Xu explained. Through rain and snow, car tires pound water into the tiny, pre-existing imperfections in the pavement producing ever-widening cracks that grow over time.

“Lithium is actually a soft material, but, like the water in the pothole analogy, all it takes is pressure to widen the gap and cause a failure,” said Xu, a postdoctoral scholar in Chueh’s lab.

With their new understanding in hand, Chueh’s team is looking at ways to use these very same mechanical forces intentionally to toughen the material during manufacturing, much like a blacksmith anneals a blade during production. They are also looking at ways to coat the electrolyte surface to prevent cracks or repair them if they emerge.

“These improvements all start with a single question: Why?,” said Cui, a postdoctoral scholar in Gu’s lab. “We are engineers. The most important thing we can do is to find out why something is happening. Once we know that, we can improve things.”

Source

Progress in AI systems often feels cyclical. Every few years, computers can suddenly do something they’ve never been able to do before. “Behold!” the AI true believers proclaim, “the age of artificial general intelligence is at hand!” “Nonsense!” the skeptics say. “Remember self-driving cars?”

The truth usually lies somewhere in between.

We’re in another cycle, this time with generative AI. Media headlines are dominated by news about AI art, but there’s also unprecedented progress in many widely disparate fields. Everything from videos to biology, programming, writing, translation, and more is seeing AI progress at the same incredible pace.

Why is all this happening now?

You may be familiar with the latest happenings in the world of AI. You’ve seen the prize-winning artwork, heard the interviews between dead people, and read about the protein-folding breakthroughs. But these new AI systems aren’t just producing cool demos in research labs. They’re quickly being turned into practical tools and real commercial products that anyone can use.

There’s a reason all of this has come at once. The breakthroughs are all underpinned by a new class of AI models that are more flexible and powerful than anything that has come before. Because they were first used for language tasks like answering questions and writing essays, they’re often known as large language models (LLMs). OpenAI’s GPT3, Google’s BERT, and so on are all LLMs.

But these models are extremely flexible and adaptable. The same mathematical structures have been so useful in computer vision, biology, and more that some researchers have taken to calling them "foundation models" to better articulate their role in modern AI.

Where did these foundation models came from, and how have they broken out beyond language to drive so much of what we see in AI today?

The foundation of foundation models

There’s a holy trinity in machine learning: models, data, and compute. Models are algorithms that take inputs and produce outputs. Data refers to the examples the algorithms are trained on. To learn something, there must be enough data with enough richness that the algorithms can produce useful output. Models must be flexible enough to capture the complexity in the data. And finally, there has to be enough computing power to run the algorithms.

The first modern AI revolution took place with deep learning in 2012, when solving computer vision problems with convolutional neural networks (CNNs) took off. CNNs are similar in structure to the brain's visual cortex. They’ve been around since the 1990s but weren’t yet practical due to their intense computing power requirements.

In 2006, though, Nvidia released CUDA, a programming language that allowed for the use of GPUs as general-purpose supercomputers. In 2009, Stanford AI researchers introduced Imagenet, a collection of labeled images used to train computer vision algorithms. In 2012, AlexNet combined CNNs trained on GPUs with Imagenet data to create the best visual classifier the world had ever seen. Deep learning and AI exploded from there.

CNNs, the ImageNet data set, and GPUs were a magic combination that unlocked tremendous progress in computer vision. 2012 set off a boom of excitement around deep learning and spawned whole industries, like those involved in autonomous driving. But we quickly learned there were limits to that generation of deep learning. CNNs were great for vision, but other areas didn’t have their model breakthrough. One huge gap was in natural language processing (NLP)—i.e., getting computers to understand and work with normal human language rather than code.

The problem of understanding and working with language is fundamentally different from that of working with images. Processing language requires working with sequences of words, where order matters. A cat is a cat no matter where it is in an image, but there’s a big difference between “this reader is learning about AI” and “AI is learning about this reader.”

Until recently, researchers relied on models like recurrent neural networks (RNNs) and long short-term memory (LSTM) to process and analyze data in time. These models were effective at recognizing short sequences, like spoken words from short phrases, but they struggled to handle longer sentences and paragraphs. The memory of these models was just not sophisticated enough to capture the complexity and richness of ideas and concepts that arise when sentences are combined into paragraphs and essays. They were great for simple Siri- and Alexa-style voice assistants but not for much else.

Getting the right training data was another challenge. ImageNet was a collection of one hundred thousand labeled images that required significant human effort to generate, mostly by grad students and Amazon Mechanical Turk workers. And ImageNet was actually inspired by and modeled on an older project called WordNet, which tried to create a labeled data set for English vocabulary. While there is no shortage of text on the Internet, creating a meaningful data set to teach a computer to work with human language beyond individual words is incredibly time-consuming. And the labels you create for one application on the same data might not apply to another task.

Enlarge / Example entry in WordNet. Takes a lot of human work to put together!

You want to be able to do two things. First, you want to train on unlabeled data, meaning text that didn't require a human to mark down details about what it is. You also want to work with truly massive amounts of text and data, taking advantage of the breakthroughs in GPUs and parallel computing in the same way that convolutional network models did. At that point, you can go beyond the sentence-level processing that the RNN and LSTM models were limited to.

In other words, the big breakthrough in computer vision was data and compute catching up to a model that had already existed. AI in natural language was waiting for a new model that could take advantage of the compute and data that already existed.

Translation is all you need

The big breakthrough was a model from Google called "the transformer." The researchers at Google were working on a very specific natural language problem: translation. Translation is tricky; word order obviously matters, but it changes in different languages.

For example, in Japanese, verbs come after the objects they act on. In English, senpai notices you; in Japanese, senpai you notices. And, of course, French is why the International Association Football Federation is FIFA and not IAFF.

An AI model that can learn and work with this kind of problem needs to handle order in a very flexible way. The old models—LSTMs and RNNs—had word order implicitly built into the models. Processing an input sequence of words meant feeding them into the model in order. A model knew what word went first because that’s the word it saw first. Transformers instead handled sequence order numerically, with every word assigned a number. This is called "positional encoding." So to the model, the sentence “I love AI; I wish AI loved me” looks something like (I 1) (love 2) (AI 3) (; 4) (I 5) (wish 6) (AI 7) (loved (me 9).

Using positional encoding was the first breakthrough. The second was something called “multi-headed attention.” When it comes to spitting out a sequence of output words after being fed a sequence of input words, the model isn’t limited to just following the strict order of input. Instead, it’s designed so that it can look ahead or back at the input sequence (attention) and at different parts of the input sequence (multi-headed) and figure out what's most relevant to the output.

The transformer model effectively took the problem of translation from a vector representation of words—taking in words in sequence and spitting out words one after another—and made it more like a matrix representation, where the model can look at the entire sequence of the input and determine what's relevant to which part of the output.

Enlarge / An example of how transformers approach translation.

Breaking out beyond language: Dall-E, Stable Diffusion, and more

As recently as 2014, XKCD, the sage of all things technical, published this comic:

 

XKCD

 

Less than a decade later, we’ve gone from computers not knowing how to recognize a bird to... well, this:

 

Enlarge

 

How did we get here so quickly?

 

We already touched upon the starting point—ImageNet, AlexNet, GPUs, and the deep learning revolution. This combination of models, data, and compute gave us an incredible set of tools to work with images.

 

Computer vision before deep learning was a slog. Think for a moment about how you, as a person, might recognize a face. The whole is made up of the parts; your mind looks for shapes that look like eyes and a mouth and determines how combinations of those shapes fit together in the pattern of a face.

 

Computer vision research used to be a manual effort of trying to replicate this process. Researchers would toil away looking for the right building blocks and patterns (called “features”) and then try to figure out how to combine them into patterns. My favorite example of this is the Viola-Jones face detector, which worked by recognizing that faces tend to fall into a pattern of a bright forehead and nose in a T-shape, with two dark areas under them.

 

Deep learning started to change all of this. Instead of researchers manually creating and working with image features by hand, the AI models would learn the features themselves—and also how those features combine into objects like faces and cars and animals. To draw an analogy to language, it’s as if the models were learning a “language” of vision; the “vocabulary” of lines, shapes, and patterns were the basic building blocks, and they were combined higher into the network with rules that served as a “grammar.” But with vast amounts of data, the deep learning models were better than any human researcher.

 

This was immensely powerful because it gave computers a scalable way to learn rules over images. But it wasn’t yet enough. These models were going in one direction—they could learn to map pixels to categories of objects to drop them into buckets and say, “these pixels show a cat; these pixels show a dog”—but they couldn’t go in the other direction. They were like a tourist who memorizes some stock phrases and vocabulary but doesn't really understand how to translate between the two languages.

You can probably see where we’re going.

Transformers were invented as translators, going from one language to another. You can translate from English to French, English to English, pig latin to English, etc. But really, languages are just ordered sequences of symbols, and translation is simply mapping from one set of ordered sequences to another. Transformers are general-purpose tools for figuring out the rules in one language and then mapping them to another. So if you can figure out how to represent something in a similar way as to a language, you can train transformer models to translate between them.

 

This is exactly what happened with images. Remember how deep learning figures out representations of the “language” of images? A deep learning model can learn what’s called a "latent space" representation of images. The model learns to extract important features from the images and compresses them into a lower-dimensional representation, called a latent space or latent representation.

 

A latent representation takes all the possible images at a given resolution and reduces them to a much lower dimension. You can think of it like the model learning an immensely large set of basic shapes, lines, and patterns—and then rules for how to put them together coherently into objects. If you’re familiar with how compression algorithms like JPEG work, it’s like learning a codebook for representing images.

 

Latent spaces are so named because they function as a coordinate grid that represents aspects of images. For example, representing a car that looks like a face means moving to a point in the latent space that's high on the “looks like a car” axis and also “looks like a face” axis. Effectively, drawing an image (or working with images in general) just means moving about in this space. And mathematically, representing coordinates in a space simply means providing a sequence of numbers. So we have now turned creating or working with images into the problem of creating a sequence… and we know how to do that. In fact, if we imagine each “axis” of the latent space as a symbol or word, we’ve turned "drawing a picture" into something that looks like "writing a sentence."

 

Enlarge

 

We now have a language of images to work with and a tool (the transformer) that lets us do translation. All that’s missing is a set of data to act as a Rosetta stone. And of course, the Internet is full of labeled images—alt text acts as a guide to what's in an image.

OpenAI was able to scrape the Internet to build a massive data set that can be used to translate between the world of images and text. With that, the models, data, and compute came together to translate images into text, and Dall-E was born.

 

Enlarge / The basic intuition of how image models work.

 

Dall-E is actually a combination of a few different AI models. A transformer translates between that latent representation language and English, taking English phrases and creating “pictures” in the latent space. A latent representation model then translates between that lower-dimensional “language” in the latent space and actual images. Finally, there’s a model called CLIP that goes in the opposite direction; it takes images and ranks them according to how close they are to the English phrase.

The latest image models like Stable Diffusion use a process called latent diffusion. Instead of directly generating the latent representation, a text prompt is used to incrementally modify initial images. The idea is simple: If you take an image and add noise to it, it will eventually become a noisy blur. However, if you start with a noisy blur, you can “subtract” noise from it to get an image back. You must “denoise” smartly—that is, in a way that moves you closer to a desired image.

 

In this case, instead of a transformer generating pictures, you have a transformer model that takes latent encodings of an image and a text string and modifies the image so it better matches the text. After running a few dozen iterations, you can go from a noisy blur to a sharp AI-generated picture.

 

Enlarge

 

But you don’t have to start with a noisy blur. You can start with another picture, and the transformer will simply adjust from this image toward something it thinks better matches the text prompt. This is how you can have an AI model that takes rough, basic sketches and turns them into photorealistic images.

 

Enlarge

 

To sum up, the breakthrough with generative image models is a combination of two AI advances. First, there's deep learning's ability to learn a “language” for representing images via latent representations. Second, models can use the “translation” ability of transformers via a foundation model to shift between the world of text and the world of images (via that latent representation).

 

This is a powerful technique that goes far beyond images. As long as there's a way to represent something with a structure that looks a bit like a language, together with the data sets to train on, transformers can learn the rules and then translate between languages. Github’s Copilot has learned to translate between English and various programming languages, and Google’s Alphafold can translate between the language of DNA and protein sequences. Other companies and researchers are working on things like training AIs to generate automations to do simple tasks on a computer, like creating a spreadsheet. Each of these are just ordered sequences.

Generate, evaluate, iterate

These AI models are incredibly powerful and flexible, and it’s useful to talk a bit more broadly about their properties. Namely, data dependence, unpredictability, emergent behaviors, and universality.

 

If you’ve taken away anything from the above, you know that training AI is dependent on data. Text and images are plentiful, which is why image models like Stable Diffusion and Google’s Imagen were able to follow quickly after Dall-E. Stability.ai is connected to and linked to several open source AI efforts, most notably Eulethera and LAION. Eleuthera created a massive data set of text called "The Pile," and LAION created the LAION-5B set of 5 billion images with corresponding text labels. This data set let other researchers quickly catch up to OpenAI’s efforts in both text and images.

 

This means AI tools will have different impacts depending on the domain and what kinds of data are available. For example, in robotics, there is not yet an equivalent to ImageNet or LAION to train robot motion planning models with. In fact, there aren’t even good, universally shared formats for data sharing in 3D motion, shapes, and touch—it’s like we're in a world not just before LAION-5B but before JPEG.

 

Or see drug discovery, where biotech companies are training AIs that can design new drugs. But these new drugs are often exploring new areas of biology—for example, proteins that are unlike naturally evolved samples. AI design has to move hand in hand with huge amounts of physical experiments in labs because the data needed to feed these models just doesn’t exist yet.

 

Another consideration is that these AI models are fundamentally stochastic. They’re trained using a technique called gradient descent. The training algorithm compares the training data to the output of the AI model and calculates a “direction” to move closer to the right answer. There’s no explicit concept of a right or wrong answer—just how close it is to being correct.

 

The basic workflow of these models is this: generate, evaluate, iterate. As anyone who’s played with making AI art knows, you typically have to create many examples to get something you like. When working with these AI models, you have to remember that they’re slot machines, not calculators. Every time you ask a question and pull the arm, you get an answer that could be marvelous… or not. The challenge is that the failures can be extremely unpredictable.

 

The power of these models is shaped by how easy it is to check the answers they give. For example, GPT-3 has pretty good mathematical ability. It’s not just able to do simple arithmetic; it can also interpret word problems at a middle school level or higher. I asked chatGPT to solve a simple math problem, which it did well:

 

Enlarge

 

And when I extended it to bigger numbers, it kept up:

 

Enlarge

 

But it hasn’t actually learned multiplication, just a simulation of it. If I have N boxes with X crayons per box, and N * X = Y, then I have Y crayons. But if I have X boxes with N crayons per box, any kid will tell you that I still have Y crayons!

Let’s see what chatGPT thinks:

 

Enlarge

 

This answer is actually better than it would have been in the past—a previous version of GPT3 would just confidently declare a totally incorrect answer:

 

Enlarge

 

At least now it knows it can’t answer! I chose this example because it’s easy for us to see how wrong the answer is. But what about more complicated domains? While it’s easy to look at a picture that an AI generates and decide if it’s good or bad on a screen, an AI-designed drug candidate has to be synthesized and tested in the real world. This doesn’t mean “AIs are bad and unreliable”; it means that when working with these AIs, this unpredictability has to be kept in mind and designed for.

 

Indeed, this unpredictability can be good. Remember, many of the capabilities these new models are showing are emergent, so they aren’t necessarily being formally programmed. GPT2 was basically a word-association machine. But when OpenAI made GPT3 a hundred times bigger, researchers discovered it could be trained to explicitly answer questions like “when did dinosaurs go extinct?” without having to be explicitly designed to answer Q&As. The same thing holds true for its ability to obey commands—there's a huge amount of headroom in terms of pushing these models to be even more capable.

 

And these models are (almost) universal. The fact that they can be used to tie language into different domains—or map directly between different domains—makes them flexible and easy to use in a way that AI models haven’t been before. It’s been very easy to take a trained model and stretch it to many different use cases, and that ease of use is playing out now by bringing these models closer to everyday use.

 

Startups are popping up to provide AI-powered writing tools, with some finding great commercial success. Large swaths of computational biology research are being overtaken by Alphafold-like models. Programming assistants are going from glorified auto-complete tools to something that can help developers quickly flesh out code from a simple spec. The previous big AI breakthroughs, whether in self-driving cars or playing games, mostly stayed in the province of labs. But anyone can use the writing apps using GPT3 or download Stable Diffusion.

 

Ubiquity is a big deal. Today, we don’t think of the printing press or the spreadsheet and word processor as being particularly exciting technologies. But by spreading the power of creating information widely, they changed the world—for the better in many cases, but sometimes for the worse. These new AI models have the potential to be just as impactful because of the uses people will put them to.

 

Predictions are hard. Perhaps the only thing we can say is that these AI tools will continue to get more powerful, easier to use, and cheaper. We’re in the early stages of a revolution that could be as profound as Moore’s Law, and I’m both excited and terrified about what's yet to come.

 

Haomiao Huang is an investor at Kleiner Perkins, where he leads early-stage investments in hardtech and enterprise software. Previously, he founded the smart home security startup Kuna, built self-driving cars during his undergraduate years at Caltech and, as part of his Ph.D. research at Stanford, pioneered the aerodynamics and control of multi-rotor UAVs.

Source

Artificial intelligence (AI)-generated imagery has moved on impressively in the last decade, from 2014 when the most cutting-edge technology could produce a picture that looked like it was taken on a Game Boy Color. Now it can even win state fair art competitions. 

However, there are still a few things that all the main leaders in AI-generated images struggle with. As well as teeth (and remembering not to cram hundreds of them into a single horrific mouth), AI is pretty terrible when it comes to hands. 

A recent example of this was posted to Twitter by creative director for mobile games Elizabeth Sampat. "My favorite part is Josh Groban looking confused," she added.

As plainly demonstrated, the image generator used has not really nailed the concept of hands. But this is part of the reason why they are so bad – working on pattern recognition and producing patterns rather than having the concept of hands themselves.

The image generator may think it has nailed the pattern by putting an indeterminate amount of fingers clumped around a palm, or in this case sometimes through the palm, but it will not survive the scrutiny of a quick glance by humans, with our fancy concept of what hands are.

On top of that, it's incredibly difficult for humans to draw hands, too.

"Due to the complex geometry of hands, there is no standard set of lines or shapes that AI can recognize as a hand," one designer and AI expert, Jim Nightingale, explained in a blog post. "In order to generate realistic hands, AI needs to collate multiple different shapes and arrangements. There are about 30 points of geometric divergences in the human hand, from finger length and width to the metacarpals and wrist joints."

Coupled with additional human scrutiny of human hands and facial features, AI has a hell of a task on its hands to impress. For now, we'd settle for hands with five fingers that don't emerge from other fingers or the center of the palm.

Source

Scientists respond to handwringing about our failure to find evidence for life beyond the Solar System by pointing out how limited our searches so far have been. Even if our instruments have picked up a signal, however, we may have overlooked it using decades-old algorithms to comb the data.

That idea has become more plausible with the application of a machine learning-based selection method to data from a major radio telescope. The system identified 100 times as many patterns in the noise deserving investigation as had been noticed previously. 

Although none of these have yet led to strong indications of alien life – you’d have heard about it if they had – a paper in Nature Astronomy reports that eight are sufficiently interesting as to prompt follow-up observations. All this from a tiny sliver of humanity’s radio telescope recordings.

The data analyzed came from 480 hours of observations of 820 stars made by the Robert C. Byrd Green Bank radio telescope, booked by SETI Breakthrough to search for radio waves that may indicate the presence of alien civilizations.

No one is anticipating a deliberate signal directed straight at Earth like the one in Contact. Instead, astronomers are hoping for stray leakage. Identifying this is a challenge, however, since it has to be distinguished not only from natural radio emissions but from the hubbub of Earth-based and satellite sources. The apparent signal known as BCL1, initially thought to come from the direction of Proxima Centauri, shows how hard these can be to tell apart.

Breakthrough Listen’s system initially identified 2.9 million “signals of interest”, narrowed to 20,515 worthy of human attention. This compares with approximately 200 found in the same data using previous methods.

The authors followed up eight (labeled MLc1-8) from seven stars, hoping for a repeat. The re-examination struck out, but the team are encouraged by the system’s sensitivity.

“We’re scaling this search effort to 1 million stars today with the MeerKAT telescope and beyond. We believe that work like this will help accelerate the rate we’re able to make discoveries in our grand effort to answer the question ‘are we alone in the universe?’” said lead author Peter Ma, an undergraduate at the University of Toronto, in an emailed statement. 

The data Ma examined was collected years earlier, reducing the prospects of finding anything on a second look. The team are excited about being able to investigate data more quickly, shortening the time to follow up studies.

“It is a pity that, despite attempts reported by the team, these signals could not be confirmed by other instruments,” said Dr Franck Marchis of SETI, who was not an author on the study, in a statement emailed to IFLScience. “The MLc1 and MLc7 signals are very interesting because they were recorded on two different dates, suggesting that they are not known interference if they are terrestrial in nature. Such a discovery requires confirmation by other instruments before we can be sure that we have detected extraterrestrial life. Nevertheless, this scientific result shows that it is now possible to announce this kind of detection quickly enough to do the necessary follow-up.”

“The arrival of large networks such as MeerKAT and the SKA, which will produce terabytes of data per week, make it imperative that SETI research adopt powerful algorithms such as deep learning,” Marchis added. “We hope that this algorithm will be able to detect a signal more quickly than conventional methods because this will allow us to follow up with other antennas and therefore confirm whether a signal is extraterrestrial.

The paper is published in Nature Astronomy.

Source

This follows 2022’s bloodbath, when tens of thousands of jobs were lost at Meta Platforms, Twitter and Salesforce. According to an industry layoff tracker, the tech sector has eliminated some 220,000 jobs since the start of last year. If the laid-off tech workers formed a city, it’d be one of the most populous in the United States, bigger than Des Moines or Salt Lake City.

The question is: Why have many of the most profitable companies of our generation — most of which are still very much profitable — announced staggering rounds of layoffs, one after the other? And why now?

A common refrain from analysts and reporters is that the companies are “tightening their belts” after profligate pandemic hiring sprees, in order to streamline operations. The executives overseeing the cuts, for their part, cite adverse economic circumstances. “We hired for a different economic reality than the one we face today,” Google CEO Sundar Pichai said in his layoff announcement. Jassy wrote that “Amazon has weathered uncertain and difficult economies in the past, and we will continue to do so.” Microsoft’s Satya Nadella noted that “some parts of the world are in a recession and other parts are anticipating one.”

No recession has yet hit the U.S. or its tech sector. Inflation hurts, but the U.S. economy added hundreds of thousands of jobs last month. Still, certain shareholders have been vocal about their desire to see head counts trimmed — and trimmed further still.

To that end, critics argue that simple greed is driving the layoffs; they point to the tens of billions’ worth of stock buybacks the tech companies authorized last year. The Verge’s Liz Lopatto spoke with industry analysts who said that tech companies are evaluating their bottom lines differently, and concluded that they’re doing layoffs mostly because everyone else is, even though layoffs actually often cost a given company money. And the fact all these layoffs are happening in such rapid succession gives the companies some cover — making them seem elemental, inevitable.

So what’s really going on here? The answer may actually be pretty simple.

“Controlling labor costs via periodic layoffs is like breathing for Silicon Valley: cyclical, necessary for life,” Malcolm Harris, author of the forthcoming book “Palo Alto: A History of California, Capitalism and the World,” told me. The layoffs, Harris says, have “very little to do with long- or even medium-term strategy except as it pertains to cultivating an insecure workforce.”

That tracks with the economic reality we do face today, as a tech CEO might put it. Because while a recession has not yet arrived in any meaningful form, there is another economic indicator pointing to the desirability of layoffs, from a large employer’s perspective: a growing effort to organize tech workers in an unusually tight labor market.

Tech employees’ salaries have skyrocketed over the last two years, and their bargaining power has begun to grow too. Over the last half-decade, workers in the tech industry have agitated for changes that executives have found increasingly inconvenient.

At Google, they’ve spoken up against gender inequalities and pressured the company to drop a lucrative defense contract. At Amazon and Microsoft, they’ve protested lackluster climate policies and spurred those companies to make pledges to reduce carbon emissions. At Facebook and Twitter, workers rallied against content moderation decisions around keeping former President Trump on the platform after Jan. 6, 2021. A subset of Google employees formed the Alphabet Workers Union with the Communication Workers of America union, Amazonians founded the Amazon Employees for Climate Justice, and at Microsoft, employees at a video game subsidiary, ZeniMax, formed the first certified union ever to be recognized by the company.

The concrete gains won by organizing tech workers so far may be relatively small, but the rising salaries and growing organizing capacity threaten the tech giants’ bottom line and the brand of executive sovereignty that’s prized in Silicon Valley. Elon Musk’s mass firings at Twitter last year, and his public demand that only “hardcore” coders dedicated to his program remain at the company, are instructive here, not least because other tech executives have said his approach was an inspiration when culling jobs at their own companies.

Workers in an industry that had long been famously union-agnostic at best had been forming bonds, organizing and developing solidarity. Layoffs of this scale and suddenness can be a blow to that process.

Affected tech workers told me that they were struck by the randomness of the firings; senior members of staff in good standing, brilliant colleagues with sterling performance reviews, all shown the door, with little rhyme or reason. Many seemed to wonder why they were spared while their peers weren’t.

Alejandra Beatty, a technical program manager at Alphabet subsidiary Verily, told me it was “very much a surprise” when she was laid off this month. For one thing, she had known herself to be in good standing at the company. “I was high performance, considered one of the pillars of the local community in the Boulder, Colo., office. Now I’m not even allowed to go back in, not even as a visitor,” she said. Beatty was also struck by how many of those let go were performing functions crucial to keeping “core products” working.

If there’s one thing that firing people in a large-scale and seemingly random way accomplishes, it’s instilling a sense of precarity, even fear, in those who remain.

“It’s completely devastating,” said Skylar Hinnant, a senior quality assurance tester at Microsoft’s ZeniMax subsidiary, “both to the people who are laid off and their families, and their colleagues, who felt, for that day and will feel it a long time after, that they’re at risk.”

Hinnant said he knew plenty of people who lost their jobs across Microsoft — everyone does. “You can be the most important engineer at your job, you can be an awesome programmer, at the end of the day if the algorithm wants you gone you’re gone.”

“I think it’s waking people up to some realities of what the industry is really like,” Alejandra Beatty said. “We are workers. Even though we have benefits and we are highly trained — we’re still workers. We can still arbitrarily lose our jobs like anyone else.”

Beatty had been a visible member of the AWU, advocating in media interviews for reproductive rights in the workplace. She was always civil and constructive, she says, and felt her suggestions had been well-received by management. Now she thinks a lot about whether her termination was retaliatory. Ultimately, she decided the layoffs were too large, too automated to have targeted her directly. “I think I was yet another number in whatever crazy algorithm the consulting company used to figure out the cuts,” she said.

Elsewhere, there are signs that tech workers’ hard-won progress is being rolled back. Bloomberg reported that at companies such as Twitter, Meta, Amazon and Redfin that had promised to improve staff diversity, the layoffs decimated departments responsible for diversity, equity and inclusion, or DEI, inititatives.

On Tuesday, workers for Cognizant, a major Alphabet and YouTube contractor, filed a complaint with the National Labor Relations Board that they were being retaliated against for announcing a decision to join the AWU. They say Alphabet changed its policy to make relocation to Austin, Texas, mandatory for all workers, and noncompliance would result in “voluntary termination.” This, the workers say, is against NLRB rules that state that no major policy can be changed once organizing has been made public.

The tech industry has certainly shown it will go to unsavory lengths to limit worker pay and power. In 2015, Apple, Google and other tech companies agreed to pay a $415-million settlement after a lawsuit alleged the companies had colluded to keep pay low with a “non-poaching” agreement between CEOs.

But today’s massive, department-cleaving cuts do not have to be retaliatory to register a debilitating and destabilizing effect, to help the tech giants consolidate control over their workforces and underline their precarity.

“The nice comp and all the perks,” Beatty said, “none of them are really beneficial if you’re always worrying, ‘Well, am I going to be the next to be arbitrarily let go?’ ”

The clinical cruelty through which some of the layoffs were administered often served to underline that point — Google employees showing up to work to find that their keycards had been deactivated, workers locked out of email accounts and not allowed to enter the offices again, not even to say goodbye to colleagues they had worked with for years.

The tech sector seems to be betting that these massive, algorithmically orchestrated firings will not only cut labor costs, but also once again remind increasingly empowered tech workers of their insecurity, and the power the companies still hold. It’s a bet that has historically paid off and has helped transform the tech giants into some of the most profitable companies in history.

But there’s also a chance that this time, things could be different.

“It’s very galvanizing,” Beatty said. The Alphabet Workers Union held a meeting shortly after the layoffs were announced. “More than 1,000 people joined,” she said, “and another 800 tried to sign up for it after it started. It was big enough that we broke Zoom, and couldn’t do breakout rooms.”

Hinnant, the quality assurance tester at ZeniMax, is also an organizer at the newly recognized union at Microsoft. “I think it has lit a fire under a lot of people to think about organizing,” he says. “I have friends at Microsoft and Google and across the industry and I got a lot of calls.”

And unlike in tech’s mass layoffs of the past, there is now a foundation of organized workers in place, even if it is a nascent one, with the ability to provide the Salt Lake City-sized population of displaced workers some resources and support.

One of the first things that the AWU did, Beatty says, was set up a Slack channel and Discord server where laid-off workers could connect, commiserate and share information and job leads.

“It was so incredibly helpful to have a place to go and talk with other people who were impacted,” Beatty said. “And then we had people who weren’t laid off joining in to support them, to say goodbye.”

Soon, thousands of laid-off workers were joining the channel, not just from Google, but from Meta, and all across the industry. “That was a huge help in understanding the terms, what you can and can’t do, and having that place where people are like, ‘Oh, I’m freaking out,’ and talking it through,” she said. “It may become a networking space — we let a couple recruiters in. They all need employment now, you know.”

Workers provided informal, ad-hoc counseling services and helped those who could no longer access their email, computers or HR contacts get information and answers. “When you’re struggling with what just happened, it’s good to have people who can help you with that,” Beatty said.

“There are people with visas or parental leave, they’re saying, ‘How do I navigate this now?’ I was compiling some of these questions to pass along, and one guy who was here on a visa said, ‘Ccan you ask them how long until I get deported?’”

Above all, workers like Beatty and Hinnant say it’s spurring interest in further tech worker organizing. “I think this really highlights the need for the people not just in the Microsoft ecosystem, but across the industry to organize,” he says. “I think this was a wake-up call. There’s a wave coming. And there’s no stopping it.”

Beatty feels likewise; she’s still hurt by the layoff but hopes to make good use of it. “If I get nothing else out of this,” she says, “I hope to see AWU grow exponentially. Then the sacrifice will have been worth it.”

Source

A person reportedly familiar with the matter said that Baidu will introduce its yet-to-be-named chatbot in March. It will also be embedded into Baidu's main search services, meaning users could soon get conversation-style search results much like on ChatGPT.

Baidu has also reportedly invested billions of dollars in its Ernie system, a large-scale AI model that the company has been developing for several years now. This will serve as the foundation of the company's ChatGPT-like platform.

ChatGPT is a natural language processing tool powered by artificial intelligence (AI) that can respond to prompts in a conversational manner. The tool can answer questions and assist with tasks like generating content and writing code. Since its launch back in November of last year, the service has amassed more than a million users.

Earlier this month, it was reported that Microsoft is looking into supercharging its Bing search engine by integrating ChatGPT into it. The company also reportedly invested $10 billion in ChatGPT creator OpenAI recently.

The rise of ChatGPT has also worried search engine giant Google, which according to the New York Times, recently declared a “code red," akin to pulling the fire alarm. The company is seemingly set to introduce over 20 AI-related projects this year, including an AI-powered version of Google Search.

Source: Bloomberg (paywall)

Chinese search giant Baidu to launch own version of ChatGPT

Microsoft, GitHub, and OpenAI ask court to throw out AI copyright lawsuit

Declining revenues

Earlier in the week, Microsoft shares its latest financial numbers covering its FY23 Q2 period ending on December 31, 2023. Although it posted a revenue of $52.7 billion - which is a growth of 2% compared to the same quarter last year - with strong figures in the cloud and its enterprise products, it faced significant declines across consumer categories like Windows, Xbox, and Surface.

Windows OEM and Commercial revenue was down by 39% and 3% respectively, Xbox content and services shrunk by 12%, and Surface devices revenue fell 39%. Microsoft cited multiple macroeconomic and execution challenges for these declines and hinted at similar results in its ongoing quarter too. Its consumer categories are expected to shrink even further with revenue to decline a bit to $50.5-51.5 billion.

Despite these challenges, Microsoft CEO Satya Nadella highlighted the company's strength in AI. This was further underlined by an announcement extending the partnership agreement between Microsoft and OpenAI, the creator of ChatGPT. Although Microsoft didn't go into specifics, its latest investment is said to be around $10 billion with a return benefit being that OpenAI models will be deployed soon in various enterprise and consumer products.

That said, another challenge that Microsoft may be experiencing soon in the enterprise space is another antitrust probe from the EU regarding the bundling of Teams in the Microsoft 365 suite of services. The root of the probe is a complaint filed by Slack in 2020 in which it stated that Microsoft hides the "true cost" from enterprise customers when it bundles Teams with the Microsoft 365 suite of products and forces it to be installed on machines. The statement of objections from the regulator regarding Microsoft's alleged anti-competitive practices is expected soon.

Windows 11 updates, bugs, and fixes

This week saw the release of build 25284 in the Dev Channel with a new Messenger widget and tons of bugs fixes for Taskbar, system tray, windowing, and more. This was followed by the availability of the Phone Link widget as well. It also introduced a bug for VPN users but thankfully, Microsoft has already issued a temporary workaround. Beta Channel users netted build 22623.1245 to fix ReFS and file copy bugs, and more. And as usual, we got a Windows Server release with the same Dev Channel build number too.

But moving away from Insider builds, an optional non-security cumulative update was also made available as KB5022360 for Windows 11 version 22H2. Speaking of this version, you may soon be upgraded to it automatically if you're still on the original release of Windows 11, that is version 21H2. This has also led to Microsoft pushing out a compatibility update for Windows 11 version 22H2 devices.

In terms of what's next for Windows 11, it seems like Microsoft is readying bootable support for Resilient File System (ReFS). There's a new File Explorer in the works too and it reportedly has better visuals along with deeper and tighter integration with OneDrive and Microsoft 365, among many other things. Additionally, Microsoft is also working on improving its driver evaluation processes, which should hopefully result in higher quality graphics and audio drivers for both Windows 10 and Windows 11. That said, if you simply can't wait for Microsoft to roll out enhancements to the software you use daily, perhaps it's worth checking out this free app which modernizes the volume and brightness sliders in Windows 10.

Lastly, when it comes to bugs and their fixes, AMD has rolled out version 23.1.2 of its driver, fixing a launch failure in Windows 11 version 22H2. In the same vein, Microsoft has also patched an issue with Remote Desktop in Windows 11. However, some more issues in launching the Start menu and UWP apps across both Windows 10 and Windows 11 have now cropped up. The good news is that Microsoft has identified the root cause and a fix is expected soon.

Teams, Edge, and OneNote upgrades, oh my!

We'll kick this section off with some good (?) news for Teams customers. Microsoft is working on integrating insights such as birthdays, career changes, and pending meeting invites in Teams profile cards. These are expected in March 2023.

On other spectrums of Microsoft 365 - which also experienced a major outage a few days ago -, we have an update for PowerPoint that introduces the ability to export slides as animated GIFs on the iPad. Additionally, Microsoft is also testing better navigation experiences for OneNote Insiders.

There are quite a few updates on the Edge front too. Apparently, Microsoft has been working on a secret version of Edge called "Phoenix", where it tests quirky and new features internally before eventually making them available in Edge. It also features a comparatively modern interface, but it's currently unknown if this browser will ever see the light of day in terms of public release. For now though, Edge Insiders can leverage Split View in Canary and some minor improvements in Dev. Meanwhile, those utilizing Edge on Windows Server 2012 and 2012 R2 will continue getting security updates until October 10, 2023.

In terms of other app updates, Snipping Tool now offers the ability to pause screen recordings, WinGet 1.4 has zip extraction, Windows Terminal Preview 1.17 packs Mica support and process restart, and EverythingToolbar now officially supports Windows 11.

Git gud

This was a big week for gamers who prefer Microsoft's platforms, primarily due to the company's Developer_Direct showcase. We got gameplay trailers for Arkane's Redfall, Forza Motorsport, and Minecraft Legends. But perhaps the best news of all was the surprise release of Hi-Fi Rush from Tango Gameworks, the creators of The Evil Within. The rhythm action game is now available via Xbox Game Pass along with other titles like GoldenEye 007, Age of Empires II on console, and more.

There was other good news for gamers who leverage Microsoft's platforms too. Some early testing has shown that the Redmond tech firm's DirectStorage technology does wonders for the only PC game it supports right now, that is Forspoken. And if you're a fan of 343 Industries, you'll be pleased to know that the development of Halo will remain with the studio, despite previous speculation suggesting otherwise.

On the Xbox side of things, the January update for the Xbox app on Windows has a sidebar upgrade on Windows and tons of bug fixes. Similarly, Xbox console gamers in the Beta ring can also utilize customizable power settings, along with other patches.

Finally, when it comes to deals, Assetto Corsa Competizione, Lawn Mowing Simulator, and Train Sim World 3 are on offer via Xbox Free Play Days. But if you're exclusively a PC gamer, you'll want to check out this Weekend's PC Game Deals, curated by our News Editor Pulasthi Ariyasinghe.

Dev Channel

• Cybercriminals are now exploiting OneNote to spread malware

• Surface Duo and Duo 2 have been offered the January 2023 firmware update

   

Under the spotlight

During this week, we published a couple of handy guides. The first is by Neowin co-founder Steven Parker in which he explains how to stop Microsoft Edge from repeatedly creating desktop shortcuts.

The other guide is a Tech Tip Tuesday piece from forum member Adam Bottjen and it details the process to prevent a thief from disabling your iPhone's tracking.

Logging off

Our most interesting news item of the week relates to a research listin g the most impersonated brands when it comes to phishing. Yahoo tops the list with a 20% share but there are several other notable names in the list too, including Microsoft, LinkedIn, Google, Netflix, WhatsApp and more. Check out more details here.

Microsoft Weekly: Declining revenues, new Windows 11 issues, and Edge updates

Recently, a team of researchers at Stanford proposed a new method called DetectGPT, which aims to be among the first tools to combat generated text in higher education. The method is based around the idea that text generated by LLMs typically hover around specific regions of the negative curvature regions of the model's log probability function. Through this insight, the team developed a new barometer for judging if text is machine-generated which doesn't rely on training an AI or collecting large datasets to compare the text against. We can only guess this means human written text occupies positive curvature regions, but the source is not clear on this.

This method, called "zero-shot", allows DetectGPT to detect machine written text without any knowledge of the AI that was used to generate it. It operates in stark contrast to other methods which require training 'classifiers' and datasets of real and generated passages.

The team tested DetectGPT on a dataset of fake news articles (presumably anything that came out of CNET over the last year) and it outperformed other zero-shot methods for detecting machine-generated text. Specifically, they found that DetectGPT improved the detection of fake news articles generated by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline to 0.95 AUROC for DetectGPT. Honestly, this is all French to me, but it purports a substantial improvement in detection performance and suggests that DetectGPT may be a promising way to scrutinize machine-generated text moving forward.

In summary, DetectGPT is a new method for detecting machine-generated text that leverages the unique characteristics of text generated by LLMs. It is a zero-shot method that does not require any additional data or training, making it an efficient and effective tool for identifying machine-generated text. As the use of LLMs continues to grow, the importance of corresponding systems for detecting machine-generated text will become increasingly critical. DetectGPT is a promising approach that could have a significant impact in many areas, and its further development could be beneficial for many fields.

Source: DetectGPT (ericmitchell.ai)

Stanford introduces DetectGPT to help educators fight back against ChatGPT generated papers

What is OpenAI?

OpenAI is one of the top AI (artificial intelligence) labs today that has created one of the most exciting products - ChatGPT. OpenAI’s new collaboration with Microsoft will make the latter a front-runner in terms of innovation. On Monday, Microsoft announced that it was investing $10 billion into OpenAI. This is over and above the $3 billion that Microsoft has invested in OpenAI since 2019. Rumors also have it that Microsoft will add ChatGPT to its search engine - Bing. With this addition, Bing may finally compete with Google’s search engine.

Following the announcement of the investment, Daniel Ives, a Wedbush analyst, stated that ChatGPT will change the entire scenario for Microsoft. Daniel said that Microsoft will not repeat its mistakes and will be aggressive from now on.

How Did Google Move Forward?

Microsoft was once the dominant force with their operating system and web browser (Internet Explorer), one of the most widely used software. It all went downhill when the US government sued Microsoft for unfairly driving out competition and taking over the browser market. Microsoft got busy with the lawsuit, and Google took advantage by rolling out better products.

What’s Next for Microsoft?

Although Microsoft continued thriving even after their setback, it doesn’t have the same market pull as it used to. It has slowly begun rebuilding with the Microsoft 365 package and Azure. Google has also made big strides forward, with almost everyone using products like Google Chrome and Gmail. 

Google is now facing multiple lawsuits from the government and is facing antitrust issues in almost every territory. OpenAI has made significant strides and is slowly eclipsing Google’s products. While Google has been working on similar AI offerings, it has held back from giving a public demonstration. With the announcement of ChatGPT, Google has a lot to catch up on. Some even say that Google has brought back founders Sergey Brin and Larry Page to help them out.

The Race Is On

Whether Google will catch up or Microsoft will continue moving forward, only time will tell. In this war between the two software giants, one thing is for sure - the customer is the winner.

Source

In 2004, Google cofounders Larry Page and Sergey Brin engaged in a comically passive-aggressive IPO road show. They eschewed business suits for casual garb, refused to answer many questions from finance bigwigs, and warned investors that instead of focusing on profits, the newly public company might apply its resources “to ameliorate a number of the world’s problems.” Both founders dreaded the restrictions of a public company and vowed that Google would never sing to Wall Street’s tune. To ensure they could do this, the founders structured the company so that they controlled the majority of voting shares. Instead of kicking back money to shareholders, Google would pamper the talent that drove its innovations, providing perks like in-house massages, free food, and lavish compensation. For instance, at the end of 2010, Page and Brin blew their workers’ minds by announcing an across-the-board 10 percent raise, a doubling of the generous annual bonus, and a $1,000 Christmas present, just for the hell of it. The beneficiaries already had top-of-market salaries augmented by lucrative equity shares. But the founders’ largesse made clear that they meant it when they said employees were the heart of the company.

Brin and Page haven't been deeply involved for years, but in the company’s 25-year history, a lot of that convention-defying legacy has remained. At least until this month, when Google’s parent company Alphabet laid off 12,000 employees, about 6 percent of its workforce, including many senior leaders and some people who had worked there since its early days. For a company renowned for coddling its workers, the layoffs were a psychic shock. Especially since some of the victims were dispatched coldly, with their email access cut off before they could even say goodbye to long-term colleagues.

Alphabet isn’t the only company dismissing workers. Top executives at Meta, Microsoft, Salesforce, Amazon, and others are doing the same thing—dealing with what they suddenly perceive as excessive headcount by lopping off heads. Current CEO Sundar Pichai’s memo was so similar to other corporate dispatches that it seems that all of them fed the same prompts into ChatGPT: Hey sorry I was too optimistic in hiring when we were raking in dough during the pandemic, so some of you will have to go. But this is just a blip in our trajectory. I’m really excited about the future that not all of you will be part of! 

Yet, the bloodletting at Alphabet is different. Aside from letting go a few hundred sales employees in 2009, the company had never experienced a major layoff. And along with it are signals that the age of limitless perks is gone. (Among those rolfed by the cuts were 27 of the company’s in-house massage therapists.) And it’s not like the company is in financial peril. Though growth has slowed and the stock is down—like at every other tech company lately—Alphabet is still pulling in plenty of money. In the most recent quarter it reported, the company managed to eke out $14 billion in profits. It also has $116 billion sitting around in its vaults. And in the past few years it has spent over $100 billion to buy back its own stock, something Wall Street loves but that does nothing for the business itself.

Pichai does have a case to make for the layoffs and a cutback in perks. With 187,000 employees, there were undeniably thousands whose jobs were not integral to the company—likely not only the massage therapists but also hundreds of middle managers performing nonessential projects. (Brin and Page always felt that middle managers slowed down innovation.) As you might expect, those working in the hotly competitive area of AI, including the Google Brain research group, were spared from the layoffs. In fact, Pichai argued that the cuts were performed so Google could spend more resources on AI.

But in some ways the layoffs represent what seems like a gradual shift in philosophy. For years, Alphabet has funded projects—and created entire divisions—devoted to producing novel forms of technology. One of those was an in-house incubator called Area 120 that was basically shut down by this month’s cutbacks. There were also some trimming in Alphabet’s X division that works on “moonshots.” Wall Street has griped for years about the unprofitability of the company’s aspirational “other bets,” and now the company seems more focused on its more concrete businesses.

It’s certainly true that Alphabet has set fire to billions of dollars in its quest for the Next Big Thing. But they call those moonshots for a reason—one success can cancel out a hundred failures. And you can argue it’s already happened. Google Brain began at X and is now not only integrated into Google but is a key component in almost all the company’s software, and a pivotal advantage in the coming wars over generative AI.  

What’s more, investing in new in-house businesses is even more important now that the US government and the EU frown on acquisitions by Big Tech. Google’s most successful move since search itself was buying YouTube for $1.6. billion in 2006—a purchase that Federal Trade Commission head Lina Khan would squash like a dung beetle if it happened today.

It’s also disheartening that Alphabet seems more inclined to count pennies on employee perks. It’s easy to mock the grandiose goodies that Google bestows on its employees, especially when you see them laid out as lurid entitlements on TikTok videos. It’s also true that not many companies can generate the profit that pays for all that. But Brin and Page had a core belief that treating workers like royalty was good business. What a concept! A disruptive innovation in its own right, it became the template for nearly all of Silicon Valley’s contenders—not just tech giants but also well-funded startups competed for top-notch chefs as fiercely as they did for machine learning adepts. It was a grand experiment that flew in the face of Wall Street’s belief that the best workforces are ones that are brutally deprived and pitilessly culled. That experiment isn’t looking as great now, and that’s to the detriment of workers everywhere as well as those of us hungry to see some crazy idea become the next big thing. (Guess that will now be more likely to come from a startup.)

Coincidentally—or maybe not—Alphabet’s moves come as one of the company’s biggest shareholders, hedge fund mogul Christopher Hohn, has been communicating with Pichai. He has been publicly complaining that the company should drastically cut its workforce—the current layoffs of 6 percent were only “a step in the right direction” he wrote, arguing for a 20 percent evisceration. He also griped about high salaries and too much money spent on Other Bets. The whole point of Brin and Page maintaining a majority of voting shares, of course, was so they wouldn’t have to listen to hedge fund multibillionaires arguing to fire workers or cut their salaries.

While the remaining Googlers are still well paid and well fed, this episode may well lead some of them to explore other options. Though Pichai and his team attempted in a company town hall this week to provide some rationale for who was let go, people I spoke to still often had little idea why person X was cut and person Y remained. But here’s what is clear: Person Y, and everyone else in the company (except maybe its AI wizards), are now a little less certain about their status. “It feels like a shift in the company,” says one long-time software engineer who can’t figure out why he got the pink slip. “I definitely get the sense that even long-term high-performing employees who are left will now be looking over their shoulders.”

In his memo, Pichai promised that Google will continue its “healthy regard for the impossible that’s been core to our culture from the beginning.” Unfortunately, it has proved impossible to do that without firing people, freaking out the survivors, and calling into question the company’s unique values.

Alphabet's Layoffs Aren’t Very Googley

(May require free registration to view)

For one Google worker it was when the light on the card reader outside their New York office flashed red, rather than green. For a staffer at Twitter, it was when their password was changed remotely, and an unusual gray screen showed their company Macbook had been locked. For Zac Bowling, a near-eight-year veteran of Google, it was being logged out of all his devices.

Tech companies have laid off tens of thousands of workers over the past few months in an industry-wide downsizing that executives have blamed on overhiring during the pandemic. Almost without fail, they’ve handled it horribly, with casual brutality and tone deaf displays—such as at Microsoft, who hosted a private Sting concert at Davos the night before firing 10,000 people.

The disparity between Big Tech’s high spending and the callous way in which they have let go of their staff has tarnished their reputation as good employers, and reminded staff that their needs are subordinate to those of shareholders. 

“Finding out via email or auto shut out that you have lost your job is brutal—and it doesn’t have to be that way,” says Gemma Dale, lecturer at Liverpool Business School, and author of a number of books, including on employee well-being and flexible work. “It also totally disconnects with what many of these organizations say about how much they value their people.” 

Bowling eventually learned that he had been let go from Google via an email two hours after he was logged out of all his work systems on the morning of January 20. His manager had to use LinkedIn to reach him to apologize, because his access to Google Meet and other internal company communication tools had been cut. 

It was entirely unexpected—Bowling had gotten a new batch of business cards made in December. Others just received, or were expecting, glowing performance reviews but were instead given severance terms. “It caught everybody off guard,” Bowling says. “It didn’t seem like they were going after low performers, or they were going after specific projects. Someone likened it to if someone had a Tommy gun, and they were just shooting from the hip.” 

People who are still at the company aren’t sure if they’re next. Bowling said that workers who still have access to the company’s systems told him 8,000 names have disappeared from employee rolls. But Google has said it’s letting go of 12,000 people worldwide. “Everybody’s saying goodbye, just in case, because they don’t know if they’re going to get everything cut off,” Bowling says. “It’s killing morale. It was just handled terribly.”

Layoffs seem to have come as a surprise to employees at several Big Tech companies, whose failure to communicate has aggravated the anguish among those now out of work.

At Salesforce, 8,000 employees were laid off in January, but co-CEO Mark Benioff reportedly ducked questions at an all-hands meeting meant to address the cuts. Bowling says his now-ex-Google colleagues also resented that they hadn’t been able to ask any questions of the executives who had let them go. At some companies—notably Twitter, where Elon Musk has cut whole teams as part of a 50 percent reduction in headcount—the firings seem arbitrary.

“It’s personally embarrassing for myself to have to explain to friends and family members why I’m getting fired,” says one former Meta employee, who was fired as part of the company’s layoffs in late 2022 and requested anonymity to avoid jeopardizing her future job prospects. 

But it isn’t just the suddenness, but also the dehumanizing way that the announcements were made, which rankles staff who have been let go. When it finally came, the email telling Bowling he was being laid off from Google was “legalese,” and was signed off by the company’s vice president without any salutation. 

“No sincerely, no sorry, nothing,” he says. “It was written by a lawyer, so there was no implied guilt or anything in there. It was so cold. Everything about it was so cold.” 

The company has historically treated employees fairly well, even when they exit, according to Bowling. “This layoff was so different from the culture of how people leave the company,” he says.

Google did not respond to a request for comment. 

But for Susan Schurman, a professor of labor studies and employment relations at Rutgers University, the gap between how tech companies portray themselves and how they act was always there.

“It would be fair to say I’m shocked but not surprised,” Schurman says. “I’m old enough to have been brought up in a so-called 20th-century organization, where you could say workers are viewed as expendable commodities.”

Attitudes toward staff have also worsened during the pandemic, according to Cary Cooper, professor of organizational psychology at the University of Manchester Business School. Remote working created a greater separation between managers and their employees. “There was less face-to-face contact, and much more of their communications were virtual,” he says. “That could create a situation where you don’t develop a close relationship with your employees, if you’re a line manager.”

Some tech workers say that they’d already come to realize that tech companies won’t necessarily return their loyalty.

“Honestly, a couple of years ago, I started changing my mindset about the companies I work for,” says Alejandra Hernandez, a recruiting program manager at Meta, who was laid off in November after working for the company for a year. “I’m looking at it as: ‘This is a business, you hired me to do certain work.’” Hernandez points out that being employed in California means she’s employed at will, and can be terminated at any time—which helped recalibrate her thinking.

Hernandez wasn’t too upset about the way that she and her colleagues were laid off by email. “I would much rather be emailed than have someone try to butter me up on a Zoom call about letting me go,” she said.

Even for those who have survived the layoffs, the past few months have acted as a sharp reminder that their well-being will never come before executives’ fiduciary duties, and that when times get tough their positions are vulnerable.

“We were all deluded into thinking these tech companies were treating people as human beings,” says Schurman. “But I think we’ve found out that it was only possible at the time, and as soon as times get tough—boom: The boss is back.”

Big Tech Is Really Bad at Firing People

(May require free registration to view)

On January 20, news broke out of Google eliminating 12,000 workers and added that these workers would get severance packages, six months of healthcare, and more. Nevertheless, the devastating news has been a lot for individuals to handle.

According to Business Insider, a Google Engineer, who wished to remain anonymous, shared insights on the effects of the downsizing on the existing workforce. The engineer mentioned the survivors of the layoffs cried during meetings when the news of the layoffs was announced.

The East Coast engineer said that in video calls that day “some of the folks were sobbing," and "they were drying their eyes.” The engineer added that the typical interactions at the office between the existing staff had now changed saying "It's not the typical nonverbal interaction there used to be before. Now it's a meaningful nod."

A second engineer from the West Coast mentioned that the staff was “angry and sad” after the news. The currently-employed individual noted that the decision removed the sense of belonging among workers. He added:

“We truly did believe that Google was something different. This is just another big company. Now, anything that used to feel special or like you really were a part of a mission — not just a big money-making machine — that feeling is I think gone."

The two engineers mentioned that the existing staff members fear further cuts. The East Coast engineer also said that some employees stayed a part of the company due to job perks which were also stripped down over time, and now, with the layoffs, “employment didn't feel as secure anymore.”

The engineers pointed out that some of the laid-off staff learned about their elimination from their colleagues, as they experienced “cannot connect” messages on Google’s internal communication channels since the ex-employees were cut off from them on the 20th. The West Coast engineer also noted that the existing staff was not informed about who was laid off.

Among the laid-off staff was Nicholas Whitaker who was part of the company’s people development team. Whitaker got to know about the news from his colleagues who were asking about his well-being, which made him think there had been a natural disaster or a shooting.

The eliminated staff reported that they are offering to help each other; Whitaker is providing meditation and mindfulness sessions free of charge to fellow workers.

Source: Business Insider

Google engineer says workers cried at meetings as the company laid off 12,000 employees