AI Updates on 2026-02-03

AI Model Announcements

  • Alibaba releases Qwen3-Coder-Next, an open-weight language model designed for coding agents and local development, featuring 800K verifiable training tasks, 80B total parameters with 3B active, achieving strong results on SWE-Bench Pro and supporting 256K context with 370+ languages @Alibaba_Qwen
  • OpenAI launches Codex desktop app for Mac with integrated development capabilities, doubling rate limits for paid plans for 2 months to celebrate the launch @sama
  • OpenAI introduces Prism, a scientific tooling platform where GPT-5.2 works inside LaTeX projects with full paper context @OpenAI
  • Anthropic integrates Claude Agent SDK directly into Apple's Xcode, giving developers full functionality of Claude Code for building on Apple platforms @AnthropicAI
  • Allen AI releases SERA-14B, a new 14B-parameter coding model with major refresh of open training datasets @allen_ai

AI Industry Analysis

  • SpaceX acquires xAI in a merger valued at $1.25 trillion, with xAI valued at $250B despite annualized revenue of $428M and annualized loss of $5.84B, planning to IPO at $1.5T+ valuation @deedydas
  • Wealthsimple transitions from GitHub Copilot to Cursor and finally to Claude Code for all 600 engineers, cancelling Copilot subscription after finding better productivity with Claude @GergelyOrosz
  • Companies are building sophisticated internal AI tools rather than launching more external features, with developers becoming much more productive but focusing on better internal tooling and eliminating existing SaaS products @GergelyOrosz
  • Software reliability is declining across the industry with increased failure rates and larger batch sizes, as AI generates larger changes that research shows tend to result in more failures @GergelyOrosz
  • Sam Altman predicts 10x growth in AI capabilities from current levels by the end of 2026, with increasing demands for locally running private models @AndrewCurran_
  • Over 200,000 people downloaded the Codex app in the first day with strong positive reception @sama
  • Waymo raises $16B at $126B valuation to scale robotaxi fleet internationally, planning to add 20+ new cities across US and internationally in 2026 @TechCrunch
  • Y Combinator announces startups can receive their $500k funding in stablecoins like USDC, citing growing adoption and passage of the GENIUS Act @ycombinator
  • OpenAI confirms NVIDIA as their most important partner for both training and inference, with entire compute fleet running on NVIDIA GPUs, scaling from 0.2 GW in 2023 to roughly 1.9 GW in 2025 @sk7037
  • Goldman Sachs CEO predicts it could be the biggest M&A year in history, citing improved regulatory environment shifting from "answer was no" to "answer is maybe" @a16z

AI Ethics & Society

  • Anthropic research finds that AI models become more incoherent rather than systematically misaligned as they reason longer, suggesting AI failures may resemble industrial accidents rather than coherent pursuit of wrong goals @AnthropicAI
  • Nature commentary by linguists, computer scientists and philosophers declares that by reasonable standards including Turing's own, artificial systems that are generally intelligent exist, stating "the long-standing problem of creating AGI has been solved" @emollick
  • Sam Altman expresses feeling "a little useless and sad" after Codex suggested better feature ideas than he conceived, noting nostalgia for the present while confident better ways to spend time will emerge @sama
  • AI Now Institute launches essay series examining narratives shaping India AI Impact Summit, questioning whether positioning countries as "data rich" creates new path to exploitation and whether AI for climate obscures material impacts @AINowInstitute

AI Applications

  • Microsoft partners with ALERT California and UC San Diego, combining Azure cloud and AI with camera network to give first responders earlier situational awareness before first 911 call, helping stop small fires from becoming devastating @BradSmi
  • CoreWeave transforms customer support in 90 days using Cohere's agentic platform North @cohere
  • Ramp builds internal revenue stack powered by customer data platform processing millions of records with agents embedded in workflows, with over 80% of sales workflows now powered by Ramp Revenue @GergelyOrosz
  • Fitbit founders launch AI platform to help families monitor their health @TechCrunch
  • Lotus Health raises $35M for AI doctor that sees patients for free @TechCrunch
  • Google launches nationwide randomized study with Included Health to evaluate AI in real-world virtual care, assessing capabilities and limitations responsibly @GoogleResearch
  • Phylo raises $13.5M seed round to build first Integrated Biology Environment (IBE) where hypotheses are generated, experiments planned, and data analyzed in auditable and reproducible way @a16z

AI Research

  • Anthropic research shows smarter models are often more incoherent, with incoherence increasing as models reason longer across every task and model tested, measured by reasoning tokens, agent actions, or optimizer steps @AnthropicAI
  • MIT researchers create AI model that guides scientists through materials synthesis by suggesting promising routes, helping make theoretical materials from generative AI libraries @MIT
  • IBM researchers implement paged attention in Helion, achieving 97% end-to-end performance versus highly optimized Triton attention backend with naive implementation @PyTorch
  • World Labs releases world model that outputs persistent 3D scenes users can build on top of, allowing extended interaction beyond 60 seconds @theworldlabs
  • Baidu's GLM enters OCR field with 0.9B parameter model using multimodal GLM-V architecture, achieving #1 on OmniDocBench v1.5 with 94.62 score @AdinaYakup
  • H Company releases Holo2-235B-A22B, achieving #1 on ScreenSpot-Pro with 78.5% and #1 on OSWorld-G with 79.0% for GUI localization @hcompany_ai