A for AI

News Archive

Pick a date to read that day's stories.

July 2026

Days with a blue dot have stories.

DISPATCH
Monday, July 13, 2026
  1. 01 of 10OPENAI

    OpenAI opens GPT-5.5 fine tuning to all Plus users

    OpenAI extended GPT-5.5 fine tuning access from enterprise to every Plus subscriber, with a free tier of 1M training tokens per month.

    OUR TAKEGiving hobbyists a fine tuning budget is either a masterstroke or a support ticket avalanche. Probably both.

    Read more →
  2. 02 of 10GOOGLE

    Google DeepMind AlphaProof 2 solves 5 of 6 IMO problems

    DeepMind said AlphaProof 2 scored gold medal level on the 2026 International Mathematical Olympiad, solving five of six problems within the time limit.

    OUR TAKEThe math olympiad is officially cooked. Next year they will just hand out medals to GPUs.

    Read more →
  3. 03 of 10ANTHROPIC

    Anthropic launches Claude for Government cloud region

    Anthropic opened a FedRAMP High cloud region for Claude, with AWS GovCloud hosting and dedicated model weights for US federal customers.

    OUR TAKEEvery frontier lab now has a government edition. The models are the same, the invoices are bigger.

    Read more →
  4. 04 of 10NVIDIA

    Nvidia acquires networking startup Enfabrica for 5B

    Nvidia agreed to buy Enfabrica for around 5 billion dollars, folding its ACF SuperNIC into the Spectrum-X networking stack.

    OUR TAKENvidia buying anything that plugs into a GPU is the least surprising deal of the year.

    Read more →
  5. 05 of 10META

    Meta Ray-Ban Display glasses hit 3M units sold

    Meta confirmed its Ray-Ban Display line crossed 3 million units, with Live AI translation cited as the top used feature.

    OUR TAKEThe face computer is quietly winning while Apple Vision Pro collects dust on a shelf.

    Read more →
  6. 06 of 10SECURITY

    GitHub finds AI generated malware packages surged 400 percent

    A GitHub security report identified a 400 percent year over year jump in malicious npm and PyPI packages containing LLM authored obfuscation.

    OUR TAKETurns out copilots also help the bad guys ship faster. Balanced ecosystem.

    Read more →
  7. 07 of 10STARTUPS

    Mistral raises 1.2B Series D at 15B valuation

    Mistral closed a 1.2 billion euro round led by DST Global, pushing valuation to 15 billion as it doubles down on European sovereign AI deals.

    OUR TAKEEurope finally has a champion, and it is priced like it might actually need to win.

    Read more →
  8. 08 of 10HARDWARE

    Cerebras WSE-4 hits 1.2 exaflops on single wafer

    Cerebras unveiled the WSE-4, a 4nm wafer scale chip delivering 1.2 exaflops of FP8 compute and shipping to Condor Galaxy customers in August.

    OUR TAKEA chip the size of a dinner plate that outperforms a GPU rack. The physics is showing off now.

    Read more →
  9. 09 of 10HEALTH

    DeepMind AlphaMissense 2 maps every human protein variant

    DeepMind published AlphaMissense 2, classifying pathogenicity for all 216 million possible human missense variants with 92 percent accuracy on ClinVar.

    OUR TAKEClinical geneticists just got a new coworker who never sleeps and never bills insurance.

    Read more →
  10. 10 of 10POLICY

    UK AI Safety Institute mandates pre deployment evals for frontier models

    The UK AISI published binding rules requiring frontier developers to submit pre deployment safety evaluations 30 days before release, effective October.

    OUR TAKEVoluntary commitments grew up and got teeth. Silicon Valley lawyers rejoice, or something.

    Read more →