LLM

ACE and IBM's ALTK-Evolve both turn agent trajectories into reusable lessons. The difference is delivery: one injects the whole playbook every step, the other calibrates. On AppWorld, calibration wins with the same accuracy at a fraction of the tokens.

A Huawei-Queen's study finds open coding models fine-tuned under OpenHands degrade sharply under other scaffolds - SWE-Lego-Qwen3-32B drops from 52.6% to 8.4% Pass@1 on OpenCode. The fix: train planning as a model capability, not a scaffold artifact.

OpenJDK's interim policy bans AI-generated contributions in full or in part, while Oracle runs on AI-written code internally. What the policy actually says, how it compares to Rust and Debian, and what it means for Java contributors.

Automated skill optimizers write long SKILL.md files whose credit is a black box. SkillSV attributes value to rules, examples, and scripts inside a skill: pruning to 69% of tokens without significant loss on four benchmarks.

A Michigan team measures prose SKILL.md files against compiled harnesses: agents execute only 56% of the steps their own skill mandates. SIGIL compiles skills into typed graph harnesses, hitting 86% compliance with 0.58x the tokens.

An inside look at the gray-market relay economy that resells OpenAI, Anthropic, and Google API access at up to 97.8% off -- and what it means for developers building on AI APIs.

Debian is voting on four proposals to regulate LLM-generated contributions - from an outright ban to full acceptance. The HN discussion reveals the fault lines in open source's biggest AI policy debate yet.

A developer got Google's Gemma 4 26B running on 2013 Xeon hardware for under $300. The fix for a silent MoE bug is now upstream - here's what it means for local inference.

A new distributed inference system pools GPU resources across multiple machines and exposes them through a single OpenAI-compatible API. No RDMA, no NVLink - just QUIC and your existing hardware.

The Program-as-Weights paper is a useful signal for developers: some LLM calls may move from per-request API prompts into compact local artifacts that behave like reusable fuzzy functions.

Open-source LLM engineering platform: tracing, evals, prompt management, and datasets. Self-hostable, OpenTelemetry-native, with 50+ framework integrations.

Security researchers showed a €0.02 bank transfer could compromise a banking AI assistant. Here is the exact attack chain - and what every developer building agents needs to do differently.

A hands-on look at Mastra, the open source TypeScript framework for building production-ready AI agents and workflows -- with verified setup commands, honest tradeoffs, and current pricing.

OpenRouter gives you one API key for 300+ models, automatic fallbacks, and intelligent provider routing. Here is what it actually costs, how to set it up in five minutes, and when you should skip it entirely.

A practical comparison of LLM routing tools - LiteLLM, Portkey, and OpenRouter - covering cost management, fallbacks, caching, and when to use each for production AI applications.

How KV caching speeds up LLM inference - the math, the code, the memory tradeoffs, and when it stops helping. Every dev running local models hits this wall.

A hands-on developer guide to Mercury 2 from Inception Labs. OpenAI-compatible API, reasoning levels, tool use, structured outputs, and when a diffusion LLM beats an autoregressive one in real apps.

Promptlock gives every prompt a 12-char content-addressable id and a diff-able artifact, turning silent prompt drift into a reviewable change.

Self-healing browser automation harness that lets LLMs complete any browser task. 5,000+ stars in under a week.

The easiest way to run LLMs locally. One command to pull and run any model. OpenAI-compatible API. 52M+ monthly downloads. Supports GGUF, Safetensors, and custom Modelfiles.