Stanford University is a US research university whose AI work is organised in part through Stanford HAI. Recent coverage includes the third AI100 report, which described the past five years as a watershed as AI moved into the mainstream; CLM-8B, a contrastive language model released with NVIDIA that the authors report sets new results on DeepSWE and Terminal Bench 2.1; a self-teaching training method developed with Tel Aviv University that needs no human data; criticism over an AI-altered student photo; and an economics paper disputing a Stanford study on graduate hiring.
AI is displacing entry-level knowledge work; a Stanford study found employment of 22-25 year-olds in AI-exposed occupations is 19% below less-exposed peers, with the gap widening. The author argues seniors get more from AI due to management skills, not knowledge, and suggests graduates become AI managers early.
Stanford HAI released the third AI100 report, calling the past five years a watershed as large-scale computing, vast datasets and generative AI moved AI from a specialized technology to everyday reality. It covers research advances, infrastructure investment, concentration of frontier development, cybersecurity, military uses and international governance.
In an interview, Cognition chief security officer Alex Stamos said AI helps both private and state actors detect and exploit vulnerabilities at once, rendering the multibillion-dollar cybersecurity industry unsustainable; he expects a "dark age" before the industry adapts.
Researchers at Tel Aviv University and Stanford University have developed a training method where two AI models generate challenges for each other and improve prediction without human-generated content or external datasets. Published on arXiv, the approach could change how future AI systems are trained.
Reports claim OpenAI algorithms solved the long-standing Navier-Stokes equations, plunging mathematics into an existential crisis. Researchers reportedly moved from despair to anger and resignation, with some scholars reduced to checking AI proofs; Dana Moshkovitz of UT Austin allegedly saw a years-long problem proved by a model within hours. The claims are not independently confirmed.
Stanford PhD student Samuel King used a generative AI model to propose genetic blueprints for microscopic viruses, not yet AI-generated life but possibly the next step. MIT Technology Review will livestream an interview with King by AI reporter James O'Donnell on Oct 16, discussing his work and his selection as an Innovators Under 35 honoree.
More US adults expect AI to hurt them than help, especially the young, yet ChatGPT just hit a billion monthly users. The frustration is less about the tech itself than corporate overreach and hype; unlike social media, AI still offers consumer choice and all 50 US states have passed or proposed AI laws.
Stanford University is facing backlash for using AI to alter a dining-hall photo, replacing a Hispanic student with an AI-generated Black female in university materials. Students say the edit violates Stanford's own AI policy and that diversity need not be fabricated. The university has not yet responded.
Researchers at Caltech and Stanford present HomeBody, which lets a frontier VLM directly orchestrate composable motor skills, giving a Unitree G1 humanoid persistent spatial memory. In an unseen kitchen it cleans across the room and retrieves remembered objects from underspecified requests without environment-specific training data or policy learning.
A CESifo working paper by economists Robert Fairlie and Jane Wu says there is no evidence AI has significantly displaced or reduced hiring of recent college graduates, contradicting a Stanford study that found entry-level employment lagging in AI-impacted occupations. The authors focus on new grads because hiring cuts may appear before layoffs.
Stanford and NVIDIA introduce Contrastive Language Models (CLMs) and release CLM-8B, which matches Jev's performance on computer-use, gaming, and tool-calling tasks while achieving up to 9× lower latency. With fine-tuning, it sets new SOTA on DeepSWE (81.6%) and Terminal Bench 2.1 (87.6%), and the authors establish power-law scaling laws for contrastive loss.
That is everything