Full-text search across 26454 articles. Combine with topic and date filters; results sorted by relevance.
A developer who spent six months building production features across ChatGPT, Claude, and Gemini reports that Claude 3.5 Sonnet wins on reasoning, coding, and creative writing, while Gemini 1.5 Pro's 2M-token context dom…
A developer argues that many LLM calls are effectively multiple-choice or extraction tasks and should be handled by pointer-style models that score spans or labels already present in the input rather than generating text…
Grok Bot's chat agent is built on standard distributed-systems primitives — queues, background workers, durable workflows, idempotency, caching and eventual consistency — with the model as the only new component, accordi…
AegisGate Security has built tool poisoning defense into its open-source MCP server framework, screening every tool description at registration time before it is exposed to an LLM. The scanner runs 34 detection patterns …
Goodfire researcher Nick Cammarata published an essay claiming that roughly 90% of neurons in modern models including Qwen3-32B perform a single interpretable job, and that the canonical "cat-and-car neuron" in Inception…
The maintainers of ARTEX, an autonomous penetration-testing system described as the champion entry in Baidu's "agent+" attack-and-defense competition, have preserved the project's final version as a pure-source backup af…
A developer built Memory Lane AI, a local-first web application that captures, transcribes, and organizes family oral histories and photographs into downloadable archival storybooks without sending personal data to cloud…
A developer who has been building AI-powered tools since 2019 argues that AI workflows should be designed to surface and learn from mistakes rather than demand flawless output, citing documented design choices in the Oke…
A developer built a recommendation app called "Because I liked" that matches films, shows, books and games to the reasons users give for loving something, then measured its quality with a recall@12 test set drawn partly …
Meta Platforms released Muse Glimmer, an open-weight AI model with approximately 30 billion parameters that runs locally on a Mac or a PC with a single consumer graphics card, on August 10, 2026, while keeping its more a…
A developer who used GrokBot to orchestrate 16 always-on agents merged 78 pull requests in a week, more than double their previous best of 35, but the setup collapsed after the bots burned $100 over a weekend and hit the…
Reddit is rolling out AI-powered moderation tools built on large language models, opening its Rules Hub to all newly created communities after testing with over 700 online communities and members of Reddit's Mod Council …
Nvidia released two open-weight multimodal models on Hugging Face on October 13: Nemotron 3.5 Super VL, a vision-language model with roughly 121 billion total parameters and about 13 billion active per token across 512 r…
A developer building the FAF project concluded that AGENTS.md and similar markdown context files are meant for instructions, not project definitions, after a 2026 study (Gloaguen et al., arXiv 2602.11988) found that repo…
A developer published a guide showing how to add live web search and page-reading tools to a LangChain 1.x agent in Python without an API key, using the AgentSearch services on the Pocket Agentic Portal. The tools cost $…
Google's Gemini 4 Argon returned to the top spot on Roboflow's overall vision benchmark, scoring 82.5% and 85.1% mAP@50 on object detection at low and high effort settings, respectively, and beating GPT-6 Astra at both e…
A developer guide details migrating from OpenAI's API to RollTok, a multi-model gateway, by changing only the API key and base URL in existing OpenAI SDK code. The guide claims RollTok provides access to 49+ models—inclu…
Epoch AI's Capabilities Index, updated October 1, shows Claude Opus 5.5 leading at 167.35, with GPT-6 Astra at 166.51, Claude Sonnet 5.5 at 165.2 and Claude Fable 5.1 at 164.82, while Google's Gemini 4 Argon does not app…
Frontier AI models can now beat Slay the Spire 1 at its base Ascension 0 difficulty at least 90% of the time, according to a new Epoch AI benchmark called SpireClimb, up from struggling to win at that level six months ea…
Kilo Code founding developer relations engineer Brendan O'Leary presented real-world gateway data from roughly 25 trillion tokens showing that developers adopt AI coding agents along a "trust ladder" rather than by bench…