Keeping up with an industry as fast-paced as artificial intelligence is a tall order. So, until an AI can do it for you, here’s a helpful roundup of recent stories in the world of machine learning, along with notable research and experiments we didn’t cover on their own.
This week at AI, Microsoft introduced a new standard PC keyboard layout with a “Copilot” key. You heard that right — from now on, Windows machines will have a dedicated key to launch Microsoft’s AI-powered assistant Copilot, replacing the proper control key.
The move is intended, one imagines, to signal the seriousness of Microsoft’s investment in the fight for AI dominance by consumers (and businesses for that matter). This is the first time Microsoft has changed the Windows keyboard layout in ~30 years. Copilot keyed laptops and keyboards are scheduled to ship in late February.
But is it all crazy? Windows users really I want an AI shortcut — or Microsoft’s taste of AI period?
Microsoft of course made a show of injecting almost all of its products old and new with the “Copilot” feature. In impressive keynotes, elegant demos and, now, an artificial intelligence wrench, the company is making its AI technology prominent — and betting on it to drive demand.
Demand is not certain. But to be fair. some vendors have managed to turn AI virus hits into hits. Look at OpenAI, the maker of ChatGPT, which According to reports to surpass $1.6 billion in annual revenue by the end of 2023. Art production platform Midjourney is also apparently profitable — and has yet to take a dime of outside capital.
Emphasis on few, although. Most vendors, weighed down by the cost of training and running cutting-edge AI models, have had to seek ever-larger tranches of capital to stay afloat. For example, it is said to be Anthropic lifting $750 million in a round that would bring its total to more than $8 billion.
Microsoft, along with its chip partners AMD and Intel, hopes that AI processing will increasingly move from expensive data centers to local silicon, commoditizing AI in the process — and it may be right. Intel’s new line of consumer chips packs specially designed cores for running artificial intelligence. Additionally, new datacenter chips like Microsoft’s could make training models a less expensive endeavor than it is today.
But there is no guarantee. The real test will be whether Windows users and enterprise customers, bombarded with what amounts to Copilot advertising, show an appetite for the technology — and will pay for it. If they don’t, it may not be long before Microsoft redesigns the Windows keyboard again.
Here are some other notable AI stories from the past few days:
- Copilot comes to mobile: In more Copilot news, Microsoft has quietly brought the Copilot clients to Android and iOS, along with iPadOS.
- GPT Store: OpenAI announced plans to launch a store for GPT, custom applications based on the AI models that generate text (eg GPT-4), within the next week. The GPT Store was announced last year during OpenAI’s first annual developer conference, DevDay, but was delayed in December — almost certainly due to the leadership change that occurred in November shortly after the initial announcement.
- OpenAI Shrinks Registered Risk: In other OpenAI news, the startup is looking to shrink its regulatory risk in the EU by channeling much of its operations overseas through an Irish entity. Natasha writes that the move will reduce the ability of some privacy watchdogs on the block to act unilaterally on concerns.
- Educational robots: Google’s DeepMind Robotics team is exploring ways to give robots a better understanding of exactly what we humans want from them, Brian writes. The team’s new system can manage a fleet of robots working in parallel and suggest tasks that can be accomplished by the robots’ hardware.
- Intel’s new company: Intel is getting away with it a new platform company, Articul8 AI, backed by Boca Raton, Florida-based asset manager and investor DigitalBridge. As an Intel spokesperson explains, Articul8’s platform “provides artificial intelligence capabilities that keep customer data, training and inferences within the security perimeter of the enterprise” — an attractive prospect for customers in highly regulated industries like healthcare and financial services.
- Dark fishing industry exposed: Satellite imagery and machine learning offer a new, much more detailed look at the shipping industry, especially the number and activities of fishing and transport vessels at sea. Turns out there are way more of them than publicly available data would indicate — a fact revealed by new research published in Nature by a team from Global Fishing Watch and several partner universities.
- Search with artificial intelligence: Perplexity AI, a platform that applies artificial intelligence to web search, raised $73.6 million in a funding round valuing the company at $520 million. Unlike traditional search engines, Perplexity offers a chatbot-like interface that allows users to ask questions in natural language (eg, “Do we burn calories while we sleep?”, “What is the least visited country? ” and so on).
- Clinical notes, written automatically: In more funding news, Paris-based startup Nabla raised $24 million. The company, which has a partnership with Permanente Medical Groupa division of US healthcare giant Kaiser Permanente, is working on a “Copilot AI” for doctors and other clinical staff that automatically takes notes and writes medical reports.
More machine learning
You may recall several examples of interesting work over the last year involving small changes to images that cause machine learning models to confuse, for example, an image of a dog with an image of a car. They do this by adding “perturbations,” small changes in the pixels of the image, to a pattern that only the model can perceive. Or at least they do thought only the model could perceive it.
An experiment by Google DeepMind researchers showed that when a flower image was perturbed to look more cat-like to AI, people were more likely to describe that image as more cat-like, even though it definitely didn’t look like a cat anymore. The same goes for other common items like trucks and chairs.
Image Credits: Google DeepMind
Why; How? The researchers don’t really know, and all the participants felt like they were being chosen at random (indeed, the influence is, though reliable, barely above chance). It seems we’re just sharper than we think — but this also has implications for security and other measures, as it suggests that subliminal signals could indeed be spread through images without anyone noticing.
Another interesting experiment involving human perception came out of MIT this week, which used machine learning they help clarify a particular system of language understanding. Basically some simple sentences like “I walked on the beach” don’t require any brain power to decode, while complex or confusing ones like “in whose aristocratic system is causing a sad revolution” produce more and wider activation, as measured by fMRI.
The team compared the activation readings of people reading a variety of such sentences with how the same sentences activated equivalent cortical areas in a large language model. They then built a second model that learned how the two activation patterns corresponded to each other. This model was able to predict for new propositions whether they would tax human cognition or not. It might sound a bit arcane, but it’s definitely super interesting, trust me.
Whether machine learning can mimic human cognition in more complex domains, such as interacting with computer interfaces, remains a very open question. However, there is a lot of research out there and it’s always worth taking a look. This week we have See Acta system by Ohio State researchers that works by painstakingly supporting an LLM’s interpretations of possible actions in real-world examples.
Image Credits: Ohio State University
Basically, you can ask a system like GPT-4V to create a reservation on a website and it will understand what its mission is and that it needs to click the “book” button, but it doesn’t really know how to do that . By improving the way it perceives well-labeled interfaces and world knowledge, it can do much better, even if it still only achieves a fraction of the time. These agent models have a long way to go, but expect a lot of big claims this year anyway! Just heard some today.
Then check out this interesting solution to a problem I had no idea existed but makes perfect sense. Autonomous ships are a promising area of automation, but when the seas are angry, it’s hard to make sure they’re on track. GPS and gyroscope don’t cut it, and visibility can be poor as well — but more importantly, the systems behind them aren’t very sophisticated. So they can miss their target or waste fuel on long detours if they don’t know any better, a big problem if you’re using a battery. I never thought of that!
Korea Maritime and Ocean University (another thing I learned about today) proposes a more robust path-finding model based on simulating the ship’s motions in a computational fluid dynamics model. They suggest that this better understanding of wave action and its effects on hulls and propulsion could seriously improve the efficiency and safety of autonomous marine transport. It may even make sense to use on human-piloted boats whose captains are not sure what the best angle of attack is for a given storm or wave form!
Finally, if you want a good recap of the last year’s big advances in computer science, which in 2023 overlap massively with ML research, see Quanta’s excellent review.
