Crabhaus

What matters in AI, and why.

← Today's edition

Full wire feed

Wed 7 Oct · 14 stories · 6 sections

Frontier labs 3 items

Mistral previewed Large 4 ('le Chonk'): 1T params, 49B active, natively multimodal, trained on 3,800 GB300s; weights due end of October.
Google DeepMind released EmbeddingGemma 2, a 740M Apache-2.0 model embedding text, code, images, video, and audio in one shared space.
Google shipped Nano Banana 2.1 on Gemini 3.6 Flash, claiming across-the-board image gains at roughly half the price of Nano Banana 2.

Agents & tooling 3 items

Meta, Walmart, Shopify, Stripe, and Sierra published the Personal Agent Protocol to standardize how AI agents transact with businesses.
OpenAI's Decisions API hit public beta: typed answers (probabilities, choices, rubric scores) about 10x faster than the Responses API.
AWS's Strands Labs released Decider 2B, a small open-source 'decision model' that picks between options instead of generating free text.

Research & papers 1 item

OpenAI published hundreds of math results from an internal frontier model, with Lean formalizations and per-result compute disclosures.

Business & policy 4 items

DeepSeek is close to raising $12B+ in a round that could reach ~$14.9B, with Tencent and CATL among top backers, before an early-2027 IPO.
Anthropic expanded its cyber verification program via Project Glasswing, citing 5,500 verified vulnerability finds from April to October.
FT: insurers brace for multimillion-dollar claims from rogue AI agent incidents, with debate over whether lab CEOs can be held liable.
a16z's 7th consumer AI report: ChatGPT has 3x rivals' US paid subscribers, 4.5% of Americans pay, and personal agents are the next front.

Tutorials 1 item

Daily Dose of DS: four ways to run LLMs locally (Ollama, LM Studio, vLLM), plus the llm-checker CLI to match models to your hardware.

Also notable 2 items

Datalab released Surya OCR, a 650M open model scoring 83.3% on olmocr (top under 3B) with layout, handwriting, and 91-language support.
Nieman Lab: ChatGPT is mimicking New Yorker cartoonists' styles and adding their real signatures to the AI-generated cartoons.
Back to top ↑