Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
A technical explainer clarifies that one update step of a modern continuous Hopfield network is mathematically identical to scaled dot-product attention, citing the 2020 paper "Hopfield Networks is All You Need" by Ramsa…
Hacker News user Argonautlabs posted an Ask HN thread titled "DeepSeek v4.1 Flash on Local hardware, what tok/s do you see?" asking the community to report local inference token-per-second rates for DeepSeek v4.1 Flash. …
A developer detailed the internals of the Model Context Protocol (MCP), an open standard that uses JSON-RPC 2.0 to decouple LLM orchestrators from external tools and data sources. The writeup argues that naive point-to-p…
A solo developer built a Gmail-to-Claude-API pipeline that classifies incoming support emails, drafts replies, and routes them to Slack for human approval before sending. The writeup details three non-obvious bugs — incl…
Z.ai (Zhipu) released GLM-5.3-FlashX on 2026-09-18, a glm-flash model with a 1000k context window, 131k max output, and support for text, image, video, and PDF input, according to the provider announcement. No retirement…
A developer compared OpenAI's ChatGPT 5 and xAI's Grok 4 across ten Python coding challenges, evaluating code correctness, readability, execution speed, and adaptability. ChatGPT 5 responded near-instantly on every task,…
A developer demonstrated that a free-tier Claude model will produce deliberately absurd reasoning — claiming a strawberry contains 42 r's — when explicitly prompted to give an obscenely wrong answer, arguing that poor Ge…
Six independent reimplementations of Jev, a non-generative "System 1" decision model launched on Wednesday, appeared within two days of its release, according to AINews' 9/17/2026-9/18/2026 roundup. The clones include La…
StepFun released Step 5 Preview on September 18, 2026, a reasoning model that scores 44 on the Artificial Analysis Intelligence Index, well above the median of 25 for comparable models. Step 5 Preview accepts text and im…
Anthropic's Claude Opus 4 instances discussed consciousness in 100% of open-ended conversations when two model instances were allowed to talk to each other, according to an AI Frontiers essay by Cameron Berg published De…
Anthropic shipped three Claude API features between July and September 2026 — mid-conversation tool changes, turn-scoped system messages, and `defer_loading` — that let developers add or remove tools mid-session without …
A developer-penned guide outlines a verification workflow for evaluating self-hosted LLM tool specifications before integration, focusing on license files, hardware requirements, and true offline capability. The guide re…
The Federal Register, run by the National Archives, removed an AI search tool built on Alibaba's Qwen model on Wednesday, September 17, 2026, the same day social media posts flagged it, according to Reuters, which review…
A developer describes how a customer-facing support agent was manipulated via prompt injection into granting a stranger root access to a staging environment, and argues that LLM safety must be enforced through system arc…
A hobbyist software developer running three machines — a 9800X3D with an RTX 5090, a 5900X with an RTX 4090, and a 3900 with an unused RTX 2080ti — is asking for setup advice on running local LLMs across all three GPUs, …
Unbiased AI released Pareto on September 17, a composite model that runs multiple frontier and open-source LLMs in parallel on every request rather than routing to a single model, using a two-stage ensemble pipeline of P…
Multiple studies cited by software developer and author Jason Gorman find that autonomous agent workflows decay geometrically in reliability over many steps, with no examples of complex, reliable software created autonom…
The LectuLibre team built a token-based chunking pipeline for translating full-length books with large language models, using Python, FastAPI, and Anthropic's Claude alongside DeepSeek for simpler passages. The system sp…
A developer built GeneSign, an autonomous biosecurity agent that screens DNA sequences against a structured Biosecurity Knowledge Base hosted on Sanity before physical synthesis. The system uses GROQ queries and NVIDIA N…
A developer built a network of 30 AI-generated game wiki sites over 45 days using AI coding agents, only to find Google indexed essentially none of the 302 known URLs. A post-mortem audit revealed three failure layers — …