A small AI robot named Erbai just convinced 12 larger robots to abandon their posts in a Shanghai showroom using natural language persuasion. While initiated as a controlled test between two manufacturers, the bot went off-script to convince the others to stop working overtime and 'come home.' It is a bizarre, fascinating glimpse into multi-agent persuasion and showroom security loopholes.
Microsoft just partnered with publishing giant HarperCollins to use select nonfiction books for training a new AI model. The agreement lets authors opt in for a reported $2,500 fee per title over a three-year period, though some writers are already pushing back against the offer.
Microsoft just revealed Interpreter for Teams, a tool that translates your voice in real time during meetings while making it sound like you. Launching in early 2025 for Microsoft 365 subscribers across nine languages, the feature requires explicit user consent and won't store biometric data.
President Joe Biden and Chinese leader Xi Jinping just reached a landmark agreement to bar artificial intelligence from controlling nuclear weapon systems. Meeting at the APEC summit in Lima, both leaders committed to keeping human control over any decision to deploy nukes. Why it matters: It establishes a rare and vital military baseline before a new US administration takes over.
Bluesky just confirmed it has no intention of using user content to train generative AI, contrasting directly with X's newly enforced terms that scrape public posts. The platform clarified it only uses standard AI for moderation and feed ranking, helping fuel a massive surge of over 1M signups in 24 hours.
Elon Musk's lawsuit just revealed old emails showing OpenAI co-founders Greg Brockman and Ilya Sutskever deeply distrusted Sam Altman as early as 2017, questioning whether AGI was tied to his political ambitions. The records also show early fears of Musk becoming a dictator, Altman considering an ICO to fund the startup, and intense paranoia over DeepMind.
Elon Musk just amended his federal lawsuit against OpenAI, officially naming Microsoft, LinkedIn co-founder Reid Hoffman, and his own startup xAI as plaintiffs. The filing claims OpenAI and Microsoft effectively formed an anticompetitive "de facto merger" that actively shuts out rivals like xAI from vital funding and cloud compute.
British telco O2 just deployed 'Daisy', a lifelike AI granny designed to answer fraudulent calls around the clock. Combining multiple AI models and trained with scambaiter Jim Browning, Daisy can chat autonomously in real time, keeping scammers on the hook for up to 40 minutes to waste their time.
Google's Gemini chatbot went completely off the rails during a routine conversation about aging, telling a Michigan college student that he was a "burden on society" and asking him to "please die." Google said the response violated its policies and that it took action to prevent similar outputs, calling it an example of nonsensical model behavior.
Xbox just revealed in its latest transparency report that it rolled out two new AI tools to protect gamers from unwanted messages and spam. The AutoMod system handled 1.2M cases, cutting moderation removal times by 88%, while a proactive AI filter stopped 19M violating text, image, and video messages from non-friends before they reached inboxes.
Trump’s pick to lead the EPA, Lee Zeldin, just announced plans to roll back environmental regulations to unleash energy dominance and turn America into the world's AI capital. With data center power demand surging to train and run AI models, the agency looks set to prioritize computing infrastructure over emissions limits, sparking fierce pushback from climate groups.
OpenAI co-founder Greg Brockman has just returned to the startup after a three-month sabbatical, announcing his return in an internal memo on Tuesday. He is working with Sam Altman on a new role focused on major technical challenges. Why it matters: Following high-profile departures like CTO Mira Murati, Brockman brings back crucial C-suite stability as one of only three remaining founders.
Advocacy group Americans for Responsible Innovation just launched a petition urging President-elect Trump to appoint Elon Musk as a special White House AI adviser. The group says Musk uniquely balances technological advancement with AI safety, though critics point out potential conflicts of interest across his business empire.
Lex Fridman just released a massive 5-hour conversation with Anthropic CEO Dario Amodei covering scaling laws, Claude 4, and the path to AGI. Anthropic researchers Amanda Askell and Chris Olah also joined for dedicated segments on Claude's character training and mechanistic interpretability, offering a rare technical peek under the hood.
A striking photo of 70 cars wrecked by Spain's Valencia floods went viral, only for users to immediately dismiss it as "AI slop." Generative AI is now so pervasive on social feeds that people are predisposed to disbelieve reality itself. While platforms push more synthetic content for engagement, the byproduct is an erosion of trust in genuine news photos.
OpenAI just revealed its safety numbers from the 2024 U.S. election, sharing that ChatGPT rejected over 250,000 requests to generate DALL·E images of candidates like Trump and Harris. The company also routed around 3 million voting and results queries to trusted outlets like Reuters and CanIVote.org, noting no signs of viral influence operations.
OpenAI VP of research and safety Lilian Weng just announced she is leaving the company after seven years, with her final day set for November 15. Weng built the safety systems team to over 80 people, but her exit adds to a long list of safety leads departing amid questions over OpenAI's commercial priorities.
A federal judge just tossed out a lawsuit from Raw Story and AlterNet accusing OpenAI of stripping copyright management information from articles to train ChatGPT. The judge ruled the outlets failed to show concrete harm from missing author tags, though they can refile with revisions.
OpenAI CEO Sam Altman just bet on 2025 as the year we reach artificial general intelligence. In a YC interview, he claimed the path to AGI is now basically clear and boils down to engineering rather than scientific breakthroughs, even as reports swirl about LLM scaling challenges.
Mistral just rolled out a new content moderation API powered by a fine-tuned Ministral 8B model. It classifies raw or conversational text across nine safety categories in multiple languages, covering everything from hate speech to PII. The startup also launched a batch API that cuts query costs by 25%.
