News
A reflection on AI advances in the past decade and how scaling and time-horizon trends might point to far greater capabilities in the decade ahead (Zhengdong Wang)
4+ hour, 56+ min ago (71+ words) [Techmeme permalink] Zhengdong Wang: A reflection on AI advances in the past decade and how scaling and time-horizon trends might point to far greater capabilities in the decade ahead" " People love to predict the future." Back in June, Peter Thiel…...
OpenAI introduces a framework to evaluate chain-of-thought monitorability and a suite of 13 evaluations designed to measure the monitorability of an AI system
1+ week, 3+ day ago (70+ words) [Techmeme permalink] OpenAI: OpenAI introduces a framework to evaluate chain-of-thought monitorability and a suite of 13 evaluations designed to measure the monitorability of an AI system" " We introduce evaluations for chain-of-thought monitorability and study how it scales with test-time compute, reinforcement…...
Mistral launches Mistral OCR 3, featuring improvements in processing forms, scanned documents, complex tables, and handwriting, priced at $2 per 1,000 pages
1+ week, 5+ day ago (70+ words) [Techmeme permalink] Mistral AI: Mistral launches Mistral OCR 3, featuring improvements in processing forms, scanned documents, complex tables, and handwriting, priced at $2 per 1,000 pages" " Overview" " Mistral OCR 3 is designed to extract text and embedded images from a wide range of documents…...
UK AI Security Institute report: AI models are rapidly improving at potentially dangerous biological and chemical tasks, and show fast jumps in self-replication
1+ week, 5+ day ago (68+ words) [Techmeme permalink] Shakeel Hashim / Transformer: UK AI Security Institute report: AI models are rapidly improving at potentially dangerous biological and chemical tasks, and show fast jumps in self-replication" " UK AISI's first Frontier AI Trends Report finds that AI models are…...
Zoom says its “federated AI” model, combining its SLM with open- and closed-source models, got 48.1% on Humanity's Last Exam vs. 45.8% for Gemini 3 Pro w/ tools
2+ week, 4+ day ago (74+ words) [Techmeme permalink] Xuedong Huang / Zoom: Zoom says its "federated AI" model, combining its SLM with open- and closed-source models, got 48.1% on Humanity's Last Exam vs. 45.8% for Gemini 3 Pro w/ tools" " Federated innovation driving breakthrough results in complex AI testing" " In…...
An Ai2 research scientist argues that AGI, as commonly conceived, will not emerge because it ignores, among other things, the physical realities of computation
2+ week, 5+ day ago (67+ words) [Techmeme permalink] Tim Dettmers: An Ai2 research scientist argues that AGI, as commonly conceived, will not emerge because it ignores, among other things, the physical realities of computation" " If you are reading this, you probably have strong opinions about AGI, superintelligence,…...
Google DeepMind plans to open its “first automated science laboratory” in the UK in 2026, focused on using AI tools to develop new materials for chips and more
2+ week, 6+ day ago (72+ words) [Techmeme permalink] Melissa Heikkil" / Financial Times: Google DeepMind plans to open its "first automated science laboratory" in the UK in 2026, focused on using AI tools to develop new materials for chips and more" " Big Tech group will work with Sir…...
An overview of AI in 2025, including arguments for and against above-trend model capabilities growth, the state of evals, and the safety of reasoning models
3+ week, 19+ hour ago (74+ words) [Techmeme permalink] Gavin Leech / LessWrong: An overview of AI in 2025, including arguments for and against above-trend model capabilities growth, the state of evals, and the safety of reasoning models" " This is the editorial for this year's "Shallow Review of AI…...
The International Committee of the Red Cross, which runs major research archives, warned that AI models are fabricating research papers, journals, and archives
3+ week, 1+ day ago (70+ words) [Techmeme permalink] Dan Vergano / Scientific American: The International Committee of the Red Cross, which runs major research archives, warned that AI models are fabricating research papers, journals, and archives" " The International Committee of the Red Cross warned that artificial intelligence…...
How Pathway, a startup developing an alternative to the transformer, aims to use its Dragon Hatchling architecture to create a new class of adaptive AI systems
3+ week, 1+ day ago (71+ words) [Techmeme permalink] Steven Rosenbush / Wall Street Journal: How Pathway, a startup developing an alternative to the transformer, aims to use its Dragon Hatchling architecture to create a new class of adaptive AI systems" " The architecture underlying large language models revolutionized…...