Perplexity -- a function measuring a model's overall level of "surprise" when encountering a particular output -- has gained significant traction in recent years, both as a loss function and as a simple-to-compute metric of model quality. Prior studies have pointed out several limitations of perplexity, often from an empirical manner. Here we leverage recent results on Transformer continuity to show in a rigorous manner how perplexity may be an unsuitable metric for model selection. Specifically, we prove that, if there is any sequence that a compact decoder-only Transformer model predicts accurately and confidently -- a necessary pre-requisite for strong generalisation -- it must imply existence of another sequence with very low perplexity, but not predicted correctly by that same model. Further, by analytically studying iso-perplexity plots, we find that perplexity will not always select for the more accurate model -- rather, any increase in model confidence must be accompanied by a commensurate rise in accuracy for the new model to be selected.

When you purchase through links in our articles, we may earn a small commission. Sterling Stock Picker uses OpenAI-powered AI to analyze stocks, assess risk, and help you make calmer, more informed investment decisions--and lifetime access is just $55.19. While markets don't like uncertainty, lately, there's been plenty of it. That's why Sterling Stock Picker is such a supportive tool . Designed to help investors make clearer, more informed decisions, this tool uses Finley AI, powered by OpenAI (the company behind ChatGPT), to analyze stocks and portfolios with depth and context.

NVIDIA is still planning to make a'huge' investment in OpenAI, CEO says The comment comes after a report from The Wall Street Journal suggested an earlier deal between the two companies had stalled. NVIDIA CEO Jensen Huang told reporters that the company will invest a great deal of money in OpenAI's latest funding round, according to, after on Friday reported that the two companies were rethinking a previous $100 billion deal that hasn't progressed beyond the early stages of negotiations. Speaking to reporters in Taipei this weekend, Huang reportedly said it could be the largest investment we've ever made. NVIDIA and OpenAI jointly announced in September that NVIDIA would be investing up to $100 billion in OpenAI to build 10 gigawatts of AI data centers. The companies said then that they were targeting the second half of 2026 for the first phase of the project to go online.

Will Storr: 'I was shocked to find my daily average was over four hours.' Will Storr: 'I was shocked to find my daily average was over four hours.' We have lost so much of ourselves to smartphones: can we get it back? My use of mobile phones has been compulsive - has it been for better or for worse? From a priest to a pensioner, a teenager to a tech CEO: can you guess our screen time? I n 2003, the Stanford social scientist BJ Fogg published an extraordinarily prescient book.

'I spoke to ChatGPT 8 times a day' - Gen Z's loneliness'crisis' Working from home after years spent alone over Covid lockdowns, 23-year-old Paisley said he began to feel trapped, and felt only AI could help him. I lost the ability to socialise, he said, and like many in Gen Z, he turned to AI for company. At one point, I was talking to ChatGPT six, seven, eight times a day about my problems, I just couldn't get away from it, it was a dangerous slope. He shared his experience of loneliness with 22-year-old documentary maker Sam Tullen, who told the BBC what Paisley was going through was part of a wider Gen Z loneliness crisis. Gen Z, a term used for those born between 1997 and 2012, often referred to as the first'digital native' generation.

I Let Google's'Auto Browse' AI Agent Take Over Chrome. Auto Browse can shop for clothes, plan a trip, and buy tickets for you. So, while testing Google's new "Auto Browse" feature for Chrome, I was filled with a strange sense of loss as I watched the AI agent open browser tabs and attempt to complete digital tasks with automated clicks. Sure, I felt some loss of control as the bot tapped away on my laptop screen. But also a kind of preemptive nostalgia for how the internet currently works, flaws and all, considering Google's plans to fundamentally alter the user experience.

No, the chatbot did not crash Perseverance. Since 2021, NASA's Perseverance rover has achieved a number of historic milestones, including sending back the first audio recordings from Mars . Now, nearly five years after landing on the Red Planet, it just achieved another feat. This past December, Perseverance successfully completed a route through a section of the Jezero crater plotted by Anthropic's Claude chatbot, marking the first time NASA has used a large language model to pilot the car-sized robot. Between December 8 and 10, Perseverance drove approximately 400 meters (about 437 yards) through a field of rocks on the Martian surface mapped out by Claude.

This change affects users who haven't upgraded to GPT-5.2, though OpenAI claims most users have already transitioned to the newer model. The retirement streamlines ChatGPT's offerings, leaving users with fewer but more advanced AI models for their interactions. You'll soon have fewer AI models to choose from when conversing with ChatGPT. OpenAI recently announced that it'll soon be discontinuing several of the AI chatbot's older GPT models. In two weeks--on Friday, February 13th--OpenAI will remove the GPT-4o, GPT-4.1,

Silicon Valley's power brokers spent the past year currying favor with President Trump. Two deadly shootings in Minneapolis are now exposing the price of that bargain. It was November 12, 2016, four days after Donald Trump won his first presidential election. Aside from a few outliers (looking at you, Peter Thiel), almost everyone in the tech world was shocked and appalled. At a conference I attended that Thursday, Facebook CEO Mark Zuckerberg said it was " a pretty crazy idea " to think that his company had anything to do with the outcome.

Plus: French company Capgemini has confirmed it's no longer working with ICE The US Department of Homeland Security is using AI video generators from Google and Adobe to make and edit content shared with the public, a new document reveals. The document, released on Wednesday, provides an inventory of which commercial AI tools DHS uses for tasks ranging from generating drafts of documents to managing cybersecurity. It comes as immigration agencies have flooded social media with content to support President Trump's mass deportation agenda--some of which appears to be made with AI--and as workers in tech have put pressure on their employers to denounce the agencies' activities. For the last couple of years, I've been following the progress of a group of individuals who believe death is humanity's "core problem." Put simply, they say death is wrong--for everyone. They've even said it's morally wrong.