AI Updates on 2025-11-15

AI Industry Analysis

  • Warren Buffett has taken a $4.3 billion stake in Alphabet, signaling major institutional confidence in Google's AI capabilities @AndrewCurran_
  • Disney's potential AI partnership decision is viewed as a crucial signal for who will lead the AI race in 2026, with the partnership expected to legitimize AI as a creative tool and provide immense promotional power to the chosen platform @AndrewCurran_
  • Google announces $40 billion investment in Texas through 2027 to build Cloud and AI infrastructure, including new data centers and funding to double the pipeline of new electricians to power the AI era @sundarpichai
  • Databricks co-founder argues the US must go open source to beat China in AI development @TechCrunch
  • Organizations are increasingly using AI in multiple business functions, showing widespread adoption across enterprises @a16z
  • Leaked documents reveal details about how much OpenAI pays Microsoft for infrastructure @TechCrunch
  • A startup CTO reports that 14 out of 15 Meta engineers failed a practical full-stack development screening that allows AI use, while engineers from startups typically pass, raising questions about skill transferability from large tech companies @GergelyOrosz
  • Mid-sized and larger tech companies are incorporating AI usage into performance reviews to reward developers who drive efficiency with the technology and encourage innovation @GergelyOrosz
  • Fei-Fei Li discusses hardware requirements for spatial intelligence, noting that chip requirements for spatial AI will differ from LLMs, particularly on rendering and training sides @a16z

AI Applications

  • Perplexity demonstrates transparency improvements in their Comet browser agent, including asking permission before delegating tasks, showing agent traces, and clearly indicating when the agent is active @AravSrinivas
  • A Comet Android early user demonstrates using the agent inside Meta Quest 3 to code on Replit while golfing, showcasing mobile AI agent capabilities @AravSrinivas
  • Sierra partners with Redfin to build a first-of-its-kind conversational home search experience @btaylor
  • Figma introduces AI suggestions that appear after duplicating frames, automatically detecting user intent to randomize labels @brian_lovin
  • Google's Veo 3.1 now allows users to upload multiple reference images alongside video prompts to create more nuanced videos true to their vision @GeminiApp
  • Linear ships nearly 30 major updates this year with an engineering team of around 40 people, demonstrating high productivity in AI-era development @karrisaarinen

AI Ethics & Society

  • Research shows that when workers know their AI use is monitored by HR, they use it less even though it significantly hurts their performance, with workers willing to be wrong just to signal judgment, presenting a challenge for leaders seeking AI adoption @emollick
  • Anthropic's attribution of a cyberattack to a Chinese state-sponsored group is questioned for lacking evidence, with their own AI model Claude unable to find technical justification for the geopolitical attribution when analyzing their report @RnaudBertrand
  • Simon Willison criticizes poor crawler behavior from Anthropic and Google, noting that crawlers are overloading applications like self-hosted GitLab and need better rate limiting @simonw

AI Research

  • A new dLLM project introduces a unified library for developing diffusion language models, demonstrating the ability to turn any BERT into a chatbot using diffusion techniques @dawnsongtweets
  • MIT develops a robotic process that dramatically increases the speed at which scientists can characterize important properties of new semiconductor materials, potentially spurring development of more efficient solar panels @MIT
  • OpenAI and Microsoft co-design AI infrastructure with hundreds of thousands of GPUs per cluster and massive bandwidth between clusters, described as an "AI superfactory" @gdb
  • Ethan Mollick observes that 95% of practical ChatGPT problems can be solved by turning on Extended Thinking, suggesting underutilization of this feature @emollick
  • Mollick suggests Google could accelerate science by improving Deep Research and Gemini's retrieval from Google Scholar and Google Books, which contain remarkable amounts of hard-to-access academic knowledge @emollick
  • Research shows increasing progress in understanding whether whales have decipherable language @emollick