We tackle the question of whether Large Language Models (LLMs), viewed as dynamical systems with state evolving in the embedding space of symbolic tokens, are observable. That is, whether there exist multiple 'mental' state trajectories that yield the same sequence of generated tokens, or sequences that belong to the same Nerode equivalence class ('meaning'). If not observable, mental state trajectories ('experiences') evoked by an input ('perception') or by feedback from the model's own state ('thoughts') could remain self-contained and evolve unbeknown to the user while being potentially accessible to the model provider. Such "self-contained experiences evoked by perception or thought" are akin to what the American Psychological Association (APA) defines as 'feelings'. Beyond the lexical curiosity, we show that current LLMs implemented by autoregressive Transformers cannot have 'feelings' according to this definition: The set of state trajectories indistinguishable from the tokenized output is a singleton. But if there are 'system prompts' not visible to the user, then the set of indistinguishable trajectories becomes non-trivial, and there can be multiple state trajectories that yield the same verbalized output. We prove these claims analytically, and show examples of modifications to standard LLMs that engender such 'feelings.' Our analysis sheds light on possible designs that would enable a model to perform non-trivial computation that is not visible to the user, as well as on controls that the provider of services using the model could take to prevent unintended behavior.

If you're looking to understand the philosophy that underpins Silicon Valley's latest gold rush, look no further than OpenAI's Scarlett Johansson debacle. The story, according to Johansson's lawyers, goes like this: Nine months ago, OpenAI CEO Sam Altman approached the actor with a request to license her voice for a new digital assistant; Johansson declined. She alleges that just two days before the company's keynote event last week, in which that assistant was revealed as part of a new system called GPT-4o, Altman reached out to Johansson's team, urging the actor to reconsider. Johansson and Altman allegedly never spoke, and Johansson allegedly never granted OpenAI permission to use her voice. Nevertheless, the company debuted Sky two days later--a program with a voice many believed was alarmingly similar to Johansson's.

Yet this month, it antagonized someone much more powerful, and is already retreating just a touch. On Monday evening, Scarlett Johansson issued a statement to NPR's Bobby Allyn about OpenAI's GPT-4o announcement, which the company showcased in a live demonstration just last week. Specifically, the multimodal computer interaction model centered around a voice assistant named Sky, whose timbre really, really resembled ScarJo's. "Last September, I received an offer from Sam Altman, who wanted to hire me to voice the current ChatGPT 4.0 system," Johansson wrote. "After much consideration and for personal reasons, I declined the offer. Nine months later, my friends, family and the general public all noted how much the newest system named'Sky' sounded like me. When I heard the released demo, I was shocked, angered and in disbelief that Mr. Altman would pursue a voice that sounded so eerily similar to mine that my closest friends and news outlets could not tell the difference."

Scarlett Johannson has gone to war with OpenAI, and in the battle for public opinion, OpenAI is losing--badly. Last week, OpenAI released an update of its AI chatbot called ChatGPT-4o, which featured a female voice talking to its users. Many people pointed out that the voice, which sometimes seemed to veer into flirtation, was eerily similar to Scarlett Johannson's in the 2013 dystopian sci-fi film Her. OpenAI CEO Sam Altman has long talked about how much the movie inspired the company's products, and even made the connection clear last week by tweeting the title of the movie. But on Monday, Johannson released a statement saying OpenAI had asked her to be the voice of the chatbot, and when she refused, they found a soundalike.

The company said so again last week when it unveiled a chattier ChatGPT that featured the Johansson sound-alike. The same day, OpenAI CEO Sam Altman posted on X a one-word reference to the 2013 movie "Her," in which Johansson was the voice of an emotional companion AI.

One of the worst parts of online shopping is trying to figure out whether an item of clothing will actually fit. While some brands have begun hiring models with more diverse body types, the process still often requires a leap of faith--or making a lot of returns. Google announced Tuesday that it's rolling out a new way to tackle the fit guessing problem using generative artificial intelligence. Brands that run ads for women's or men's shirts will now have the ability to show shoppers how the products look on dozens of different real models, without taking additional photos. The new feature means that typing in "eyelet crop top" on Google's search engine, for example, could return an ad with a clickable gallery that shows what an item looks like on women with different skin tones and body types.

For the past decade, AI researcher Chris Olah has been obsessed with artificial neural networks. One question in particular engaged him, and has been the center of his work, first at Google Brain, then OpenAI, and today at AI startup Anthropic, where he is a cofounder. "What's going on inside of them?" he says. "We have these systems, we don't know what's going on. That question has become a core concern now that generative AI has become ubiquitous. Large language models like ChatGPT, Gemini, and Anthropic's own Claude have dazzled people with their language prowess and infuriated people with their tendency to make things up. Their potential to solve previously intractable problems enchants techno-optimists. But LLMs are strangers in our midst. Even the people who build them don't know exactly how they work, and massive effort is required to create guardrails to prevent them from churning out bias, misinformation, and even blueprints for deadly chemical weapons. If the people building the models knew what happened inside these "black boxes,'' it would be easier to make them safer.

An AI chatbot often recommends neighbourhoods to potential home buyers and renters based on race. This may be a reflection of bias stemming from generations of US housing discrimination. "A lot of people think that generative AI and large language models are the emerging technologies of the future," says Eric Liu at the Massachusetts Institute of Technology. "But of course they're being trained on data from the past." How this moment for AI will change society forever (and how it won't)

It just revealed a bunch of new hardware and plans for Windows. Copilot PCs were the big announcement, designed to run generative AI processes locally instead of in the cloud. Of course, Microsoft had new Surface devices to showcase these features, but the usual PC suspects also have new laptops that meet the spec requirements -- and include Copilot in their name for added chaos. The company also claims Copilot PCs are 58 percent faster than the M3-powered MacBook Air. We'll drill into some other announcements down below.

I'm in Seoul for the International AI summit, the half-year follow-up to last year's Bletchley Park AI safety summit (the full sequel will be in Paris this autumn). While you read this, the first day of events will have just wrapped up – though, in keeping with the reduced fuss this time round, that was merely a "virtual" leaders' meeting. When the date was set for this summit – alarmingly late in the day for, say, a journalist with two preschool children for whom four days away from home is a juggling act – it was clear that there would be a lot to cover. The inaugural AI safety summit at Bletchley Park in the UK last year announced an international testing framework for AI models, after calls … for a six-month pause in development of powerful systems. There has been no pause. The Bletchley declaration, signed by UK, US, EU, China and others, hailed the "enormous global opportunities" from AI but also warned of its potential for causing "catastrophic" harm.