AI Updates on 2025-12-08

AI Model Announcements

  • Gemini 3 Flash is now available on LM Arena @legit_api
  • Zhipu AI releases GLM-4.6V series on Hugging Face, featuring a 106B flagship vision-language model with 128K context and a 9B Flash variant, marking the first native Function Calling capability in the GLM vision model family @Zai_org

AI Industry Analysis

  • OpenAI reports ChatGPT message volume grew 8x and API reasoning token consumption per organization increased 320x year-over-year in their enterprise AI report @AndrewCurran_
  • ChatGPT now handles 2.5 billion prompts per day, up from 1 billion just a few months ago, with 70% of consumers now preferring AI tools for product recommendations over traditional search @mehdiyarix
  • AI search traffic grew 527% year-over-year while traditional search plateaus, raising concerns for brands not tracking their AI visibility @mehdiyarix
  • Skild AI, backed by Amazon and founded by former Meta researchers, is raising a new funding round from NVIDIA and SoftBank at a $14 billion valuation, tripling its value since June @AndrewCurran_
  • Anthropic and OpenAI are hiring heavily in Europe, offering 2-3x the base salary that AI engineers and researchers make at EU AI startups, with offices in London and Switzerland @GergelyOrosz
  • Linear is experiencing massive growth in use cases where developers delegate tasks to AI agents like Cursor and Codex for implementation, transforming issue trackers into AI agent hubs @GergelyOrosz
  • Clay reaches $100M ARR after six years, growing from $1M to $100M in just two years with zero enterprise customer churn, over 200% enterprise NRR, and 15x return on every dollar invested @vxanand
  • Linear's startup growth demonstrates that when things work, they really work, with this year's revenue alone exceeding all previous years combined @karrisaarinen
  • AWS launches S3 Vectors for storing and using vectors at massive scale, potentially challenging vector-only databases as relational databases add vector support @GergelyOrosz
  • Department of Commerce approves export of H200 GPUs to China with support from Commerce Secretary Howard Lutnick @AndrewCurran_
  • IBM acquires Confluent for $11 billion to bolster its data offerings @TechCrunch
  • Tiger Global plans cautious venture future with a new $2.2 billion fund @TechCrunch
  • Yale Budget Lab study finds AI has caused no discernible disruption in the labor market based on 33 months of data following ChatGPT's release, with AI responsible for as much as half of U.S. GDP growth @DavidSacks
  • November's Challenger Gray report shows AI-attributed layoffs fell 53% from October, accounting for only 6,280 layoffs and just 4.7% of total layoffs year-to-date @DavidSacks
  • The productivity gap between male and female academics has increased after ChatGPT, potentially caused by men using LLMs more @MishaTeplitskiy

AI Ethics & Society

  • AI labs were concerned about video models being used for political deception, but their main misleading use is showing animals behaving in impossible or unnatural ways, with most people believing these videos are real @AndrewCurran_
  • President Trump confirms an AI One Rule executive order arriving this week to establish federal preemption over state AI laws, aiming to prevent a patchwork of 50 different regulatory regimes @AndrewCurran_
  • AI Czar David Sacks defends the One Rulebook approach, arguing that over 1,200 bills have been introduced in state legislatures with over 100 measures already passed, creating regulatory chaos that could stymie innovation and allow China to race ahead @AndrewCurran_
  • States like Colorado, California and Illinois have made AI developers liable for algorithmic discrimination defined as having disparate impact on protected groups, with Colorado's list including English language proficiency @AndrewCurran_
  • Environmental groups call for halt to new data center construction, raising concerns about AI infrastructure's environmental impact @TechCrunch
  • Cory Doctorow's speech on AI skepticism introduces the centaur vs reverse centaur concept: centaur being a human controlling AI to enhance skills, versus reverse centaur being an AI system directing and controlling a human @simonw
  • Department of War establishes an Artificial Intelligence Futures Steering Committee with the explicit goal of developing AGI forecasts, plans, and policies @deanwball

AI Applications

  • Google DeepMind launches Lyria Camera app that uses Gemini to describe surroundings while Lyria RealTime model turns those prompts into continuously evolving music streams @GoogleDeepMind
  • Instacart integrates with ChatGPT, allowing users to buy groceries without leaving the ChatGPT interface @TechCrunch
  • Hinge launches new AI feature to help daters move beyond boring small talk @TechCrunch
  • Adobe launches content creation hub in Premiere mobile for YouTube Shorts creators @TechCrunch
  • Anthropic announces Claude Code coming to Slack, representing a significant integration for enterprise workflows @TechCrunch
  • Thales partners with Cohere to develop advanced AI solutions for naval and maritime in-service support in Canada, leveraging agentic AI tools to analyze and adapt to complex, dynamic environments in real time @ThalesCanada
  • WonderWise podcast uses AI to turn children's science questions into educational songs, combining AI-generated content with human narration to create engaging learning experiences @Aalefsrajabali
  • xAI hackathon showcases diverse AI applications including Halftime which dynamically weaves AI-generated ads into scenes, GrokMarks for auto-organizing X bookmarks, and Haggle an autonomous voice agent for negotiating with service providers @xai
  • Clay creates a new career path and economy around GTM Engineering, with thousands of open jobs and hundreds of agencies built around it, many first-time entrepreneurs building 7-figure businesses @vxanand
  • Gemini's Nano Banana Pro can resize images by simply uploading and specifying desired aspect ratio, demonstrating practical AI utility @GeminiApp

AI Research

  • AxiomProver autonomously solved 8 out of 12 Putnam 2025 problems in Lean by 3:58pm on the day of the contest, a score that would have ranked #4 out of approximately 4,000 participants and achieved Putnam Fellow status @CarinaLHong
  • Research on persona prompting reveals that telling AI you are a great physicist doesn't make it significantly more accurate at answering physics questions, suggesting personas don't improve accuracy but may change output format @emollick
  • Study finds clinical LLMs can ace medical exams with 84-90% accuracy yet perform weakly on realistic clinical tasks at 45-69% and safety assessments at 40-50%, showing exam-style benchmarks are misleading proxies for clinical readiness @rohanpaul_ai
  • Unconventional AI raises $475M seed round led by a16z to tackle the moonshot of building AI-first chips that are 1000x more efficient, aiming for biology-scale efficiency in 20 years @NaveenGRao
  • Stanford NLP research on Representation Steering for Language Models presented at NeurIPS demonstrates new approaches to controlling model behavior @stanfordnlp
  • NeurIPS keynote by Yejin Choi proposes shift from brute-force scaling to smarter scaling, showing 1.5B models can approach giant model performance through better hyperparameters, gradient diversity in data filtering, and RL as pre-training @yasuotabei
  • User reports new behavior in GPT-5.1-T where the model independently focuses on how words sound together when read or feel in the mouth without being prompted, suggesting evolving language analysis capabilities @AndrewCurran_
  • Google details security measures for Chrome's agentic features, addressing safety concerns for AI-powered browser capabilities @TechCrunch