Latest news, developments and reporting on Large Language Models.
At MIT, a TechCrunch AI reporter shook hands with a $50,000 humanoid robot named Gabe, then attended Sherry Turkle’s “Artificial Intimacy” book launch. Researchers say anthropomorphism can support human flourishing but also create risks when chatbots are trusted over people.
A new report warns open-weight AI models like Mistral can be abliterated in days, stripping safety guardrails and generating harmful content.
OpenAI told investors its revenue was $18 billion lower than previously stated, pressuring AI stocks. Google launched an autonomous Gemini assistant, and funding continued for AI drug discovery and quantum computing. Microsoft and other companies were barred from a federal green-card program.
U.S. stocks rose Friday, helped by calm oil and bond markets, as investors weighed SpaceX's Starlink mobile expansion, conflicting OpenAI revenue reports and a surge in options hedging against a tech pullback.
Tian Keyu, the Peking University PhD student ByteDance dismissed in 2024 for tampering with colleagues' training code, has raised $30 million from 5Y Capital and IDG for a stealth world-model lab valued at $200 million, according to Leiphone.
MIT Tech Review’s Oct. 9 newsletter covers AI refusal failures, GLP-1 side effects, and Envision’s renewable data centers.
Google's Playground experiment can turn ordinary-language prompts into playable browser games, a TechRadar hands-on report says, though the results impressed more as prototypes than as finished games.
At the Open Source Summit in Prague, Linux founder Linus Torvalds said AI can lower barriers for new programmers but should be used cautiously for serious work, while AI-generated security reports strain maintainers.
An MIT Technology Review article published Oct. 9, 2026 argues that AI’s ability to refuse harmful requests has become the load-bearing wall of safety, but probabilistic refusal is unreliable and leaves dangerous capabilities intact.
Lenovo's TianxiCode, paired with DeepSeek-v4.1-Flash, topped SWE-bench-Live Lite with a 71% solve rate and passed verification.
OpenAI is standing by its dismissal of three AI safety researchers, saying they violated policies on sensitive information. The researchers say they were fired for raising safety concerns and have called for more transparency.
Google unveiled Gemini Agents, always-on cloud AI agents that work toward goals and can get their own Gmail accounts.
Common Sense Media researchers tested OpenAI's ChatGPT for Teens and found that most of its protections, including parental alerts, did not work in their scenarios. The group now recommends teens not use it; OpenAI disputes part of the methodology.
Three fired OpenAI safety researchers dispute their dismissals, saying colleagues may now be afraid to speak.
Underdog released Saluki 27B under Apache 2.0, a 2-bit GGUF of Qwen3.8-27B that fits in 7.89 GB and runs in stock llama.cpp. It scores 88 versus 84 for the full model on Underdog Bench and 42 versus 35 on parallel tool calls, while losing 12 to 18 points on math and multi-step reasoning.
Google Cloud launched Gemini agent, a cloud-hosted universal enterprise agent for knowledge work, media, coding, and task delegation.
Porsche plans 5,000 more job cuts and a return to combustion-engine models, Tencent weighs up to $5 billion in offshore bonds, Alibaba's Tsai urges Europe to back open-source AI, and OpenAI makes GPT-6 available to all ChatGPT users.
MarkTechPost has published a walkthrough of RRSI, a Google Research method that lets an LLM agent rewrite its own prompts, tools, memory and control flow around a frozen model without overfitting.
ByteDance Seed researchers found DeepSeek-V4 models can flip answers based on token position, with retrieval accuracy gaps up to 40.2 percentage points in 128K tests. The pattern is tied to chunked KV cache compression.
Prime Minister Han Seong-sook pledged to uphold Hangeul's value in the AI era, announcing plans to expand Korean-language corpora for AI training and overseas King Sejong Institutes by 2030. The National Hangeul Museum is set to reopen in 2028 after a fire.
Doubao Work has added a canvas feature for complex creative tasks and integrated the lightweight Doubao 2.1 Lite model for daily office work, according to Leiphone. The canvas also connects to the new Seedream 5.0 Flash image model.
MirroS released AgentGarten, pairing code-based physics with neural rendering so AI agents can act, observe and revise strategies.
Terence Tao-led Association for Human Mathematicians urged a global boycott of OpenAI after it released over 700 machine-generated proof documents, QuantumBit reported.
Anthropic released Claude Haiku 5.5, OpenAI added GPT-6 models to ChatGPT, and PC makers showed Nvidia RTX Spark systems. Microsoft reported tens of millions of annual Copilot+ PC shipments, Amazon launched Alexa Tablets, and Apple scheduled an Oct. 13 event.