Actor Scarlett Johansson has accused OpenAI of copying her voice for one of the voice assisstants in ChatGPT despite denying the company permission to do so. Johansson's statement on Monday came hours after OpenAI said that the company would no longer use the voice in ChatGPT but did not provide a reason why. "Last September, I received an offer from Sam Altman, who wanted to hire me to voice the current ChatGPT 4.0 system," Johansson wrote in the statement that was first shared with NPR. "He told me that he felt that by my voicing the system, I could bridge the gap between tech companies and creatives and help consumers to feel comfortable with the seismic shift concerning humans and AI. He said he felt that my voice would be comforting to people."

In a statement provided to The Washington Post by Johansson's publicist, she claimed that she received an offer from OpenAI CEO Sam Altman in September to be the voice of its AI system. Johansson, who famously voiced the role of the all-knowing AI in the 2013 movie "Her," said she declined Altman's offer.

What if you could run an entire Windows PC on a mobile Arm-based chip, bringing the power efficiency and thinner designs from smartphones and tablets to laptops? If you've been paying attention to Microsoft's PC strategy over the past two decades, this song probably sounds familiar. From the original Surface in 2012 (running Windows RT for Arm devices) to the recent Surface Pro 9 5G, Microsoft has chipped away at this dream, only to fail miserably every time. Now with its new Copilot PC initiative, which includes major upgrades in Windows for Arm systems and AI, Microsoft may finally have the answer to its mobile computing dreams. Microsoft's portable PC ambitions didn't start with the Surface line: You can trace it back to Windows CE and Windows Mobile-based Pocket PCs.

All over the PC industry today, we're learning of new systems and products launching in conjunction with Microsoft's Copilot push. The company up and decided to nuke its entire product portfolio altogether and unify most of its sub-series. While HP was never the worst offender in the world of awful product names -- I'm looking at you, Sony, LG and Lenovo -- being able to quickly identify the make and model of a device is crucial when you're deciding what to buy. HP's vice president of consumer PC products Pierre-Antoine Robineau admits as much, saying "to be fair, we don't make things easy with our portfolio." He referred to the company's brands like Spectre, Pavilion and Envy, saying that if you ask ChatGPT what they are, the answers you'd get might refer to a ghost or a gazebo.

OpenAI removed a heavily promoted voice option from ChatGPT on Monday, following a widespread reaction to the flirtatious, feminine voice that sounded almost identical to Scarlett Johansson. The company used the voice, which it calls "Sky", during its widely publicized event last week debuting the capabilities of the new ChatGPT-4o artificial intelligence model. Researchers talked with the AI assistant to show off Sky's personable and responsive affectations, which users and members of the media immediately compared to Johansson's AI companion character in the 2013 Spike Jonze film Her. Even OpenAI's CEO, Sam Altman, seemed to suggest that the vocal design was intentionally mimicking Johansson's character, posting a one-word tweet after the presentation that simply said "her". Less than a week later, OpenAI felt compelled to explicitly clarify that Sky was not based on Johansson.

Theory of mind is a hallmark of emotional and social intelligence that allows us to infer people's intentions and engage and empathize with one another. Most children pick up these kinds of skills between three and five years of age. The researchers tested two families of large language models, OpenAI's GPT-3.5 and GPT-4 and three versions of Meta's Llama, on tasks designed to test the theory of mind in humans, including identifying false beliefs, recognizing faux pas, and understanding what is being implied rather than said directly. They also tested 1,907 human participants in order to compare the sets of scores. The team conducted five types of tests.

When OpenAI showed off GPT-4o's seemingly more-human like voice mode last week, observers were quick to point out that one of ChatGPT's voices sounds like Scarlett Johansson, particularly her character in Her. The company says the similarity between the flirty AI voice Sky (which it actually rolled out in September) and Johansson was unintentional. However, it's "working to pause the use of Sky" while it addresses some questions about the voice. "We believe that AI voices should not deliberately mimic a celebrity's distinctive voice -- Sky's voice is not an imitation of Scarlett Johansson but belongs to a different professional actress using her own natural speaking voice," OpenAI wrote in a blog post detailing how it picked ChatGPT's five voices. "To protect their privacy, we cannot share the names of our voice talents."

In a shocking turn of events, AI systems might not be as safe as their creators make them out to be -- who saw that coming, right? In a new report, the UK government's AI Safety Institute (AISI) found that the four undisclosed LLMs tested were "highly vulnerable to basic jailbreaks." Some unjailbroken models even generated "harmful outputs" without researchers attempting to produce them. Most publicly available LLMs have certain safeguards built in to prevent them from generating harmful or illegal responses; jailbreaking simply means tricking the model into ignoring those safeguards. AISI did this using prompts from a recent standardized evaluation framework as well as prompts it developed in-house.

Soon after OpenAI released GPT-4o last Monday, some Chinese speakers started to notice that something seemed off about this newest version of the chatbot: the tokens it uses to parse text were full of spam and porn phrases. Humans read in words, but LLMs read in tokens, which are distinct units in a sentence that have consistent and significant meanings. GPT-4o is supposed to be better than its predecessors at handling multi-language tasks, and many of the advances were achieved through a new tokenization tool that does a better job compressing texts in non-English languages. But, at least when it comes to the Chinese language, the new tokenizer used by GPT-4o has introduced a disproportionate number of meaningless phrases--and experts say that's likely due to insufficient data cleaning and filtering before the tokenizer was trained. If left unresolved, it could lead to hallucinations, poor performance, and misuse.