Whether a large language model policy is an explicit constitution or an implicit reward model, it is challenging to assess coverage over the unbounded set of real-world situations that a policy must contend with. We introduce an AI policy design process inspired by mapmaking, which has developed tactics for visualizing and iterating on maps even when full coverage is not possible. With Policy Projector, policy designers can survey the landscape of model input-output pairs, define custom regions (e.g., "violence"), and navigate these regions with rules that can be applied to LLM outputs (e.g., if output contains "violence" and "graphic details," then rewrite without "graphic details"). Policy Projector supports interactive policy authoring using LLM classification and steering and a map visualization reflecting the policy designer's work. In an evaluation with 12 AI safety experts, our system helps policy designers to address problematic model behaviors extending beyond an existing, comprehensive harm taxonomy.

A major shakeup is in the works at OpenAI. Reuters reported that the artificial intelligence research company is restructuring its business from a non-profit board into a for-profit corporation. The publication also says Sam Altman would be given equity in the new corporation. OpenAI's move to for-profit wouldn't eliminate its non-profit entity entirely. The non-profit would own a stake in the new for-profit venture but it won't have nearly the power as it did.

In a surprise move, OpenAI's chief technology officer announced on Wednesday that she would soon leave the company after six and a half years. In a note shared with the company and then posted to Twitter/X, Mira Murati wrote she was leaving the tech company behind ChatGPT. "After much reflection, I have made the difficult decision to leave OpenAI … I'm stepping away because I want to create the time and space to do my own exploration," she said. CEO Sam Altman offered kind words in response to Murati's departure, writing on X: "I feel tremendous gratitude towards her for what she has helped us build and accomplish, but I most of all feel personal gratitude towards her for the support and love during all the hard times. I am excited for what she'll do next."

Although Meta Connect 2024 lacked a marquee high-end product for the holiday season, it still included a new budget VR headset and a tease of the "magic glasses" Meta's XR gurus have been talking about for the better part of a decade. In addition, the company keeps plowing forward with new AI tools for its Ray-Ban glasses and social platforms. Here's everything the company announced at Meta Connect 2024. Today's best mixed reality gear -- like Apple's Vision Pro and the Meta Quest 3 -- are headsets with passthrough video capabilities. But the tech industry eventually wants to squeeze that tech into something resembling a pair of prescription glasses.

Hi all, I have something to share with you. After much reflection, I have made the difficult decision to leave OpenAl. My six-and-a-half years with the OpenAl team have been an extraordinary privilege. While I'll express my gratitude to many individuals in the coming days, I want to start by thanking Sam and Greg for their trust in me to lead the technical organization and for their support throughout the years. There's never an ideal time to step away from a place one cherishes, yet this moment feels right.

OpenAI chief technology officer Mira Murati resigned on Wednesday, saying she wants "the time and space to do my own exploration." Murati had been among the three executives at the very top of the company behind ChatGPT, and she was briefly its leader last year while board members wrestled with the fate of CEO Sam Altman. "There's never an ideal time to step away from a place one cherishes, yet this moment feels right," she wrote in a message to OpenAI staff that she posted on X. Altman replied to Murati's X post writing that "it's hard to overstate how much Mira has meant to OpenAI, our mission, and to us all personally." He added that he feels "personal gratitude towards her for the support and love during all the hard times." A successor wasn't immediately announced.

The more important upgrade for Meta's long-term ambitions, though, is the new ability of its models to see users' photos and other visual information. Meta today also announced Llama 3.2, the first version of its free AI models to have visual abilities, broadening their usefulness and relevance for robotics, virtual reality, and so-called AI agents. Some versions of Llama 3.2 are also the first to be optimized to run on mobile devices. This could help developers create AI-powered apps that run on a smartphone and tap into its camera or watch the screen in order to use apps on your behalf. "This is our first open source, multimodal model, and it's going to enable a lot of interesting applications that require visual understanding," Zuckerberg said on stage at Connect, a Meta event held in California today.

Large language models (LLMs) seem to get less reliable at answering simple questions when they get bigger and learn from human feedback. AI developers try to improve the power of LLMs in two main ways: scaling up – giving them more training data and more computational power – and shaping up, or fine-tuning them in response to human feedback. How does ChatGPT work and do AI-powered chatbots "think" like us? José Hernández-Orallo at the Polytechnic University of Valencia, Spain, and his colleagues examined the performance of LLMs as they scaled up and shaped up. They looked at OpenAI's GPT series of chatbots, Meta's LLaMA AI models, and BLOOM, developed by a group of researchers called BigScience. The researchers tested the AIs by posing five types of task: arithmetic problems, solving anagrams, geographical questions, scientific challenges and pulling out information from disorganised lists.

OpenAI just announced via X/Twitter that the Advanced Voice feature is finally being released to ChatGPT. This exciting update allows users to have more natural spoken-language conversations with the chatbot, bringing us yet another step closer to Her becoming reality. Advanced Voice isn't just "ChatGPT with text-to-speech" -- it uses its underlying GPT-4o technology to analyze the tone and speed of your voice and pick up on non-verbal cues like body language, then itself responds with emotion in its own voice. With five new voices, ChatGPT feels more alive than ever. The responses come faster so it feels like you're holding a real-time conversation, and the update improves accents in (certain) non-English languages.

Michael Skelly hasn't learned to take no for an answer. For much of the last 15 years, the energy entrepreneur has worked to develop long-haul transmission lines to carry wind power across the Great Plains, Midwest, and Southwest. But so far, he has little to show for the effort. Skelly has long argued that building such lines and linking together the nation's grids would accelerate the shift from coal- and natural-gas-fueled power plants to the renewables needed to cut the pollution driving climate change. But his previous business shut down in 2019, after halting two of its projects and selling off interests in three more.