AI Updates on 2025-11-14
AI Model Announcements
- OpenAI releases GPT-5.1 in their API with new reasoning options and adaptive reasoning capabilities for instant responses, though some users note regressions in certain tasks like the pelican example compared to GPT-5 @simonw
- Perplexity makes GPT-5.1 available to Pro and Max subscribers @perplexity_ai
- Alibaba ships Qwen Code v0.2.1 with major improvements including free web search (2000 searches/day for OAuth users), smarter code editing with fuzzy matching, better IDE integration, and multi-stage normalization pipeline for zero-overhead matching @Alibaba_Qwen
- OpenAI launches group chats in ChatGPT as a pilot in Japan, New Zealand, South Korea, and Taiwan, enabling collaboration with friends, family, or coworkers alongside ChatGPT in the same conversation @OpenAI
- Google announces SIMA 2, a general agent that can understand and reason about complex instructions and complete tasks in simulated game worlds, even ones it has never seen before, learning through self-play @demishassabis
- Google AI Plus expands to 53 additional countries, making productivity and creativity tools available in 130 regions worldwide @GeminiApp
- Google rolls out Deep Research update to all Gemini app users on mobile (Android and iOS), allowing users to select sources, enter prompts, and generate reports @GeminiApp
- Claude API now supports structured outputs through native tool support, eliminating the need for previous workarounds using single tool calls with schemas @simonw
AI Industry Analysis
- Mira Murati's startup Thinking Machines Lab is in early talks to raise funding at a valuation of roughly $50 billion, more than 4x its valuation from a few months ago @shiringhaffary
- Disney CEO indicates plans to deploy AI not just in production processes but across the entire company, including for generative short-form user-created content on Disney+ platform, signaling Disney's transformation into an AI company @AndrewCurran_
- Analysis suggests Disney's hard-nosed YouTube negotiations may be pressure tactics by Google to push Disney to choose Veo over Sora as their AI partner @AndrewCurran_
- A new category of startups called "Neolabs" emerges in Silicon Valley, with 9 out of 10 achieving $1B+ valuations at seed stage, likely under $10M in revenue, founded by ex-model lab AI researchers who have made $10-100M+ in personal wealth @deedydas
- Venture capitalists are abandoning old rules for a "funky time" of investing in AI startups, reflecting changing investment patterns in the AI sector @TechCrunch
- Harvey, built by a first-year legal associate, becomes one of Silicon Valley's hottest startups, demonstrating AI's impact on the legal industry @TechCrunch
- Cisco acquires EZDubs, a speech translation technology company, to embed their technology in Cisco's videoconferencing products for use by millions @snowmaker
- NVIDIA GPU rental rates for H100 and A100 have stabilized after price drops over spring and summer @a16z
- Gamma CEO Grant Lee emphasizes that real product market fit beats brute force marketing, noting their growth came from organic word of mouth rather than advertising spend @a16z
AI Ethics & Society
- AI Now Institute releases report "Fission for Algorithms: The Undermining of Nuclear Regulation in Service of AI" examining how nuclear regulation is being compromised to serve AI infrastructure needs @AINowInstitute
- Anthropic's Amanda Askell discusses the challenge of making Claude approach political topics fairly, suggesting existing norms around respect and professionalism can inform how AI models should navigate these issues @AmandaAskell
- Anthropic releases open-source political bias evaluation materials to promote transparency in AI model behavior @AnthropicAI
- Apple updates App Review Guidelines to clamp down on apps sharing personal data with third-party AI systems @TechCrunch
- A federal judge denies Apple and OpenAI's motions to dismiss Elon Musk's antitrust lawsuit @AndrewCurran_
- Perplexity's Comet Assistant introduces transparency features showing exactly what actions it's taking, asking permission before sensitive actions like logging in or completing purchases, and allowing users to control browsing behavior @perplexity_ai
- Francois Chollet outlines the "ladder of intelligence" from memorization to metacognition, arguing that achieving compounding AI requires reaching Level 4 (discovering general principles and metacognition) through symbolic program synthesis rather than parametric learning @fchollet
- Ethan Mollick raises concerns about expectations for open weights models keeping pace with closed ones, citing rising costs without clear revenue paths, government pressure on capable systems, and questioning long-term viability of Chinese frontier models remaining open @emollick
AI Applications
- OpenAI confirms Sora cameos already work with fictional characters at a high level, requiring only IP permission for use @AndrewCurran_
- Claude Code's new front-end design Skill improves vibe-coded apps by considering audience and moving beyond default purple gradients and Arial fonts @emollick
- GPT-5 Pro proves incredibly useful for social science research, allowing researchers to analyze datasets and papers, check work, perform alternative specifications, and verify findings through provided code and statistical results @emollick
- Claude Code combined with Playwright MCP creates a powerful combination for development tasks @brian_lovin
- Microsoft demonstrates "vibe coding" enabling anyone, regardless of experience, to build apps using AI assistance @Microsoft
- Microsoft Copilot introduces Learn Live featuring Mico, an AI study buddy that helps break down complex ideas and maintain focus @Copilot
- Google Photos launches six new AI-powered features for editing, creating, and searching, including Nano Banana for photo remixing @GoogleAI
- Google Shopping integrates directly into Gemini App for convenient holiday shopping, with agentic AI features including checkout and calling stores for availability @GoogleAI
- NotebookLM receives major updates including custom video overview styles, chat history, images as sources, and Deep Research capabilities @GoogleAI
- ChatGPT now respects custom instructions to avoid using em-dashes in responses @sama
AI Research
- Google DeepMind's SIMA 2 demonstrates ability to play games in the mind of Genie 3, showing advanced agent capabilities in procedurally generated worlds @demishassabis
- ChatGPT demonstrates ability to recognize when problems are too difficult to solve, as evidenced by reading MathOverflow posts and declining to attempt overly complex problems @aryehazan
- Reuters reports that OpenAI offered to collaborate with DeepMind on AI research in 2019 but was rejected, with speculation that an "OpenMind" collaboration could have significantly advanced timelines @AndrewCurran_
- Stanford researchers introduce new AI model for computer vision that recognizes object parts, understands their function, and transfers skills between objects, advancing toward real-world usefulness @StanfordHAI
- An LLM-generated paper reaches top 17% of ICLR submissions by average reviewer score (receiving two 8's) despite containing BS jargon and hallucinated references, though one reviewer gave it a zero after actually reading it @micahgoldblum
- Ethan Mollick notes custom system prompts may degrade LLM results without users knowing, as accuracy improvements are being built into models rather than requiring prompt engineering @emollick
- Google's Android team reports moving from C++ to Rust yields 1000x reduction in memory safety vulnerability density (1 per 5M lines), 4x lower rollback rate, and 25% less time in code review @deedydas
- Francois Chollet argues that all great scientific breakthroughs are forms of symbolic compression, taking complex observations and reducing them to simple rules expressed as mathematical equations @fchollet
- MIT announces new platform for designing metal compositions with previously unattainable properties, representing an entirely new approach to making metals @MIT
- Andrej Karpathy expresses excitement about self-driving technology's potential to terraform outdoor physical spaces, reduce parking infrastructure, improve safety, decrease noise pollution, and free up human attention from lane following @karpathy