Agentic AI news, research notes, and industry analysis.
9 stories in this month.
Ricursive Intelligence released Ripple-65B on November 4, a dense 65B open-weight model that scores 80.52 on TerminalBench 2.1 — beating models more than ten times its size. The story isn't just the benchmark: it's how they got there with $32M, 1.4M RL rollouts, and no purchased training data.
Read more →Anthropic published FlashProve on November 2 — an open-source prototype that uses zkSNARKs to let users verify which model weights actually produced a given response. At ~110 tokens/sec proof generation on a single H100 (with 70–95x GPU overhead), it's not production-ready, but the problem it's solving is real and underexplored.
Read more →Simon Willison's LLM CLI hit 0.37 on November 2, adding gpt-6.1 models, a simpler OpenAI-compatible endpoint setup command, and multimodal embeddings. The more interesting note is that nearly all 12 contributing authors used AI assistance, and Willison uses Opus 5.5 or GPT-6 Astra to triage those submissions — a small but clear snapshot of where open source maintenance is now.
Read more →Figma's October 29 writeup describes how it trained its own canvas-editing model using synthetic SFT data followed by RL with designer-calibrated reward models. The results — 20% faster tasks, 60% fewer undos — are more interesting than the benchmark numbers, because undo rate is the one metric that tells you whether a model actually understood what a user meant.
Read more →A blog post from this weekend describes using GPT-6 Astra to diagnose and edit the weights of Qwen3.5-0.8B without gradient descent, raising MATH-500 accuracy from 7.9% to 48.1% by removing a single looping direction from two MLP layers. It's early and constrained, but the concept of an LLM doing targeted surgical weight editing is new territory.
Read more →Nolan Lawson measured the actual dollar cost of having Claude Opus 5.5 build the same app across twelve JavaScript frameworks. The spread is nearly 2x. A separate observation from the PostgreSQL project suggests AI is now surfacing latent bugs faster than maintainers can fix them — two data points on how AI is reshaping the economics of software work.
Read more →Balestriero and LeCun formally prove the Platonic Representation Hypothesis for JEPAs: models trained separately on any modality learn the same representation up to rotation, provided they use good positive pairs. The result reframes next-token prediction as a JEPA with structurally bad views, explains the slow alignment of language models with vision, and shows a 109M text JEPA outperforms every LLM tested up to 72B parameters on cross-modal tasks.
Read more →Cisco Talos published a detailed breakdown of Volta, an infostealer that routes victim data through an LLM to produce a ranked monetization assessment — with a three-model refusal-bypass fallback chain built in. It's a small but concrete sign of how AI components are becoming standard modules in commodity malware toolkits.
Read more →The GVS5H paper showed that five Qwen3.8-27B instances coordinating through a shared file system outperform Fable 5 on LiveCodeBench Hard at roughly 9% of the cost. Memdoor, an open-source harness, independently reproduced the result — 49 of 50 runs solved, $0.39 median per problem.
Read more →