A for AI

News Archive

Pick a date to read that day's stories.

July 2026

Days with a blue dot have stories.

DISPATCH
Sunday, July 12, 2026
  1. 01 of 10OPENAI

    ChatGPT rolls out 'Study Together' group mode

    OpenAI expanded a live group study feature that lets multiple users share a ChatGPT session with synced whiteboards and voice, targeting classrooms and tutoring groups.

    OUR TAKEGroup homework, now with a fourth roommate who never sleeps and always shows its work.

    Read more →
  2. 02 of 10GOOGLE

    Gemini 3 Ultra tops new agentic benchmark

    Google says Gemini 3 Ultra leads the latest GAIA-2 agent benchmark, edging out Anthropic and OpenAI models on long-horizon browser tasks.

    OUR TAKEAnother leaderboard, another victory lap. Wake us when it books a dentist appointment without hallucinating the address.

    Read more →
  3. 03 of 10ANTHROPIC

    Claude gets persistent project memory

    Anthropic turned on cross-session memory for Claude Projects, letting the assistant recall files, decisions, and tone across weeks of work without re-uploading context.

    OUR TAKEFinally, an AI that remembers what you told it Tuesday. Your therapist is taking notes.

    Read more →
  4. 04 of 10NVIDIA

    Nvidia ships Rubin dev kits to hyperscalers

    Nvidia began shipping early Rubin GPU development systems to Microsoft, Meta, and Oracle ahead of a broader Q4 launch, per Reuters sources.

    OUR TAKEHopper is barely warm and Blackwell just landed. The upgrade treadmill is now sponsored by your power bill.

    Read more →
  5. 05 of 10META

    Meta open-sources Llama 4 Vision weights

    Meta released Llama 4 Vision under a permissive license, matching GPT-4o on standard multimodal evals and adding native video understanding up to 30 minutes.

    OUR TAKEZuck keeps giving away frontier models while everyone else charges by the token. Strange timeline, great for builders.

    Read more →
  6. 06 of 10SECURITY

    Prompt injection worm hits enterprise agents

    Researchers at Wiz disclosed a self-propagating prompt injection that spreads through shared Slack and email agents, extracting tokens from connected tools.

    OUR TAKETurns out giving your inbox a brain and OAuth scopes was, in fact, a plot twist waiting to happen.

    Read more →
  7. 07 of 10STARTUPS

    Perplexity raises at 25 billion dollar valuation

    Perplexity closed a new round led by SoftBank at a reported 25 billion dollar valuation, doubling its mark from earlier this year despite Google AI Overviews pressure.

    OUR TAKEInvestors are still betting a chatbot can dethrone Google. Bold, expensive, and possibly correct.

    Read more →
  8. 08 of 10HARDWARE

    Apple tests on-device 30B model for iPhone 18

    Bloomberg reports Apple is testing a 30 billion parameter Apple Intelligence model that runs entirely on next year's iPhone 18 Pro with the new M-class Neural Engine.

    OUR TAKESiri may finally graduate from 'here is what I found on the web' to 'here is the actual answer.' Only took a decade.

    Read more →
  9. 09 of 10HEALTH

    FDA clears first autonomous AI radiologist

    The FDA granted clearance to Rad AI's Continuum system to independently sign off on normal chest X-rays without a human radiologist, a first for the agency.

    OUR TAKEThe robot reads your scan, the human reads the interesting ones. Efficient, unnerving, probably the future.

    Read more →
  10. 10 of 10POLICY

    EU AI Act general purpose rules take effect

    The EU AI Act's obligations for general purpose models kicked in this week, forcing frontier labs to publish training data summaries and copyright compliance reports.

    OUR TAKEModel cards just got a lot longer, and a lot more lawyered.

    Read more →