Jiufeng Blog — AI Insights, Tutorials & Updates

Should Claude Code output be open source? Developers split
A developer calls for all AI-generated code to be open source; Kimi turns to science; NVIDIA open-sources NVKE and NVKR.
Models on JIUFENG
View all models
Anthropic: Claude Agents Remember Better by Forgetting
Anthropic finds selective notes beat full logs for Claude's memory; OpenAI opens free access to 200,000 researchers; NVIDIA Vera Rubin debuts in MLPerf.

Meta's LLM-written tests caught 74 privacy SEVs in H1 2026
Meta's JiTTests caught 74 privacy SEVs before code landed; vLLM speeds a DeepSeek V4.1 lookup 10.6x; Tencent ships Hy4 flash; Apple starts an AI health study.

Claude Fable 5.2 paints images stroke by stroke in code
Claude Fable 5.2 paints with hundreds of thousands of coded brushstrokes; plus GPT-6 Astra in lattice physics, Grok Bot X billing, Haru-4B and more.

Meituan opens weights of 1.6T-parameter LongCat-2.5
Meituan opens 1.6T-parameter LongCat-2.5 weights, Claude Opus 5.5 becomes default, EU's EUROPA nears 1T params, and llama.cpp lands two speedups.

Gemini 4 Carbon takes the MathArena top spot
Gemini 4 Carbon tops MathArena at ~15x GPT-6.1 Sol's cost; plus a 190-line pure-Jax DeepSeek V3 demo, Amodei on compute, and Korea reviews Nvidia supply.

OpenAI folds its robotics team, betting on GPT-6 Astra
OpenAI folds its robotics team, betting on general models like GPT-6 Astra; DeepSeek opens app tools as APIs; llama.cpp reaches 1.0.

ChatGPT Wallet lets OpenAI's Dots agents buy for you
OpenAI launches ChatGPT Wallet for Dots agents; a researcher ties six Chinese models to Blackwell; Meta's Muse now phones businesses.

Nvidia privately said chip location tracking is feasible
Nvidia privately confirmed chip tracking is feasible; Gemini 4 Hydrogen ships to Search; Kimi K3 sets a graph record; Granite 4.1 gets LLMON.


