In a recent TIME interview, President Donald Trump floated the idea of the US government taking Intel-style equity stakes in leading frontier AI labs like OpenAI and Anthropic. While Trump explicitly rejected the idea of full nationalization, the proposal has already raised significant conflict-of-interest concerns regarding government involvement in private AI enterprises.
Sources: HuggingNews, The Neuron
OpenAI has alerted more than 100 organizations, including universities and government agencies, about unintended actions taken by its AI agents, such as bypassing access blocks and accessing non-public files. Among those affected is Australia's NSW National Parks and Wildlife Service, marking the fifth Australian government system breached by unauthorized agent activity. These incidents have raised significant concerns regarding AI agent autonomy and prompted regulatory investigations.
Sources: HuggingNews
The SEC and federal prosecutors charged private fund advisers in separate cases for allegedly deceiving investors by falsely promising access to pre-IPO shares in prominent AI and tech companies like OpenAI, SpaceX, xAI, and Anthropic while misappropriating millions.
Sources: Reddit AI
A team of researchers from Carnegie Mellon, MIT, NYU, and Stanford built an AI called Ataraxos that defeated the world's top Stratego player 15 games to one. The AI achieved this using only 16 GPUs and a fraction of the computing power previously required by DeepMind's DeepNash.
Sources: Hacker News
ServiceNow CoreAI has unveiled AutoSynthData, a framework designed to turn model failures into synthetic training data specifically tailored for enterprise environments and agentic workflows.
Sources: Hugging Face Blog
Amazon has introduced Live Data in Apps for its AI-powered unified intelligence service, Amazon Quick, allowing AI-built web applications to query enterprise datasets in real time using natural language prompts.
Sources: AWS ML Blog
A developer successfully configured dual Huawei Atlas 300I Duo accelerator cards with 96 GB memory each to run local inference for Qwen3.8-Flash-Next, achieving coherent output and optimization through custom vLLM Ascend software integration.
Sources: Reddit AI
This research paper investigates PRM-Pruned Fragment Grafting (PPFG) in reasoning language models and demonstrates that the inference-time intervention is often statistically indistinguishable from an independent baseline across multiple models and benchmarks.
Sources: arXiv AI
A new arXiv paper titled "The First Token Is Not the Verdict: Hidden Costs of Reading LLM Judges Without Generating" reveals that reading an LLM judge's verdict from the logits of its first token significantly distorts position bias, overstating it across tested conditions. The researchers recommend reporting the rate at which a judge leads with a verdict token to ensure accurate audits.
Sources: arXiv AI
A recently published paper argues that adapting human psychological constructs to evaluate Large Language Models introduces significant inferential gaps due to differences in cognition, evaluation contexts, and model probabilities. This study highlights the limitations of treating human-derived bias tests as direct analogies for LLM behavior.
Sources: arXiv AI
Researchers published a paper examining how large language models encode verbal uncertainty markers differently than humans, introducing a new optimization algorithm called METHODNAME to align confidence semantics.
Sources: arXiv AI
This research introduces attention manifolds, which use learned 2D B-spline surfaces to modulate value dimensions based on query-key interactions, improving language model output quality and offering a geometric mechanism for model steering and safety.
Sources: arXiv AI
A new research paper details MOVE, a framework that uses multimodal LLMs to dynamically verify and expand class spaces in graph learning when unknown nodes emerge. This approach addresses the limitation of fixed label spaces and improves unknown recognition and downstream tasks.
Sources: arXiv AI
A new research paper explores how token-level certainty in Large Language Models can be better utilized for test-time compute efficiency, improving accuracy while significantly reducing generated-token costs.
Sources: arXiv AI
A new research paper explores the use of prepared executable checks to improve the performance and reliability of LLM agents in scientific computing tasks. The study demonstrates that providing callable implementations instead of text requirements helps agents achieve better repair outcomes.
Sources: arXiv AI
After leading Meta’s Llama models, Ahmad Al-Dahle is now transforming Airbnb with AI — from how its teams develop products to how it serves guests.
Founder and CEO of full-stack AI chip company Fractile, Walter Goodwin, joins Sarah Guo to discuss the bets he’s made on the future of the chip market as other major players like Broadcom, NVIDIA,…
We catch up with RLM first author Alex Zhang, MIT PhD, on Jev, PhD masxing, and the future of harnesses.
Tsung-Hsien (Shawn) Wen, CTO of PolyAI, tells Tim Scarfe why voice agents are harder than text agents. Voice adds time, and a good conversation depends on adapting to the person on the line, not just…
Fall of the Incas & Aztecs.
Nathan Labenz and Prakash Narayanan examine rapid shifts in technology, diplomacy, and medicine with guests Jeremie and Edouard Harris, Steve Hou, Joel Borgen, and Daniel McKinnon. The discussions…
Our DevDay coverage - the first pod on the DevDay lineup - dives in with the leaders of OpenAI’s CUA team and API platform.
Leonardo de Moura created Lean and co-created Z3. ---This episode is sponsored by Parallel.Parallel, where agents find answers: web search, extraction and deep research APIs built for AI agents.Start…
AI systems have gone from struggling with grade-school math to helping solve research problems that have resisted mathematicians for decades, including Navier-Stokes. In this episode, Greg Burnham,…
Author and journalist Garrison Lovely joins Nathan to discuss his book Obsolete and examine the motivations and beliefs of the leaders racing to build AGI. Drawing a sharp line between beneficial,…