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 developer released Reactive Agents, an open source MIT-licensed TypeScript framework now at v0.16, designed to make agent harness behavior explicit rather than hidden inside a black box. Each .with() call enables exact…
OpenAI's self-reported results show GPT-6 Astra reaching 95.9% mean voxel IoU on the University of Virginia's BenchCAD benchmark, against 83.3% for GPT-5.6 Sol and 84.3% for Claude Fable 5.1 under a modified evaluation p…
A developer building MCQplex, an exam-prep platform for Nepal's NEB curriculum, has open-sourced three small Node.js utilities extracted from that work. The packages include llm-free-cascade, which fails over across free…
Three independent security researchers at Hacktron compromised OpenAI employee GitHub accounts using Anthropic's Claude Opus 4.8 and Claude Opus 5 in under 72 hours, gaining access to OpenAI's Monorepo and internal devel…
Research reported by Ars Technica found that Google's SynthID-Text watermarking can alter LLM responses to adversarial prompts and increase the risk of safety guardrail bypasses, affecting tool invocation and model adher…
Newly unredacted Microsoft court filings in the New York Times lawsuit against OpenAI and Microsoft show a Microsoft executive privately described AI data scraping as "the largest theft of labor in human history," accord…
A developer has outlined five distinct failure modes of "context rot" in LLM agent harnesses, arguing that degradation in reasoning quality occurs as sessions grow even before a context window is exhausted. The writeup i…
A technical author who writes cookbooks for reporting tools tested hundreds of real developer questions against current chat models and found five recurring failure modes: models reproduce the most-repeated internet answ…
A developer released KillSwitch, an esoteric programming language explicitly designed to be difficult for large language models and coding agents to parse or generate. The project ships with an interpreter, a specificati…
A developer building a multi-agent drug-discovery system tested TypeSafe's Jev decision model as an independent validation gate for catching incorrect literature claims, molecular edits, and property records. Jev decline…
A user reported that the Chapter 11 Code Quiz in the LLM Course fails on submission with a MaxRetryError against api-inference.huggingface.co, where the endpoint /models/Qwen/Qwen2.5-Coder-32B-Instruct/v1/chat/completion…
TypeSafe released jev_jsonschema, a Python library that converts JSON Schema into questions for its Jev structured-output model and returns JSON validating against the original schema. The library maps boolean schemas to…
A developer's three-part series on token-efficient agentic development explains that AI coding agents consume far more tokens than a single prompt suggests, because each step of an agent loop — reading files, searching r…
A technical breakdown of Claude Code's session compaction explains that the runtime automatically summarizes conversation history once a session reaches roughly 160K tokens, about 80% of the 200K context window, silently…
A developer built an offline AI tool that converts static HTML sites into installable WordPress themes using a local language model paired with a retrieval-augmented generation layer, requiring no API keys, cloud service…
Xiaomi's MiMo team launched a public dashboard at mimo.xiaomi.com/rl on September 16 that streams live training telemetry for two trillion-parameter models, MiMo-V2.6-Pro and MiMo-V2.6-Flash, with the Pro run costing rou…
A developer published an engineering guide detailing how Retrieval-Augmented Generation (RAG) pipelines are vulnerable to knowledge injection and indirect prompt injection attacks, in which adversaries plant malicious in…
A developer built the Multilingual Code and Language Dataset, an open-source collection of nearly 150 GB of uncompressed text and source code covering Central Asian languages including Kyrgyz, Kazakh, Uzbek, and Tajik, n…
A U.S. military intelligence report this spring that a Chinese ship in the Middle East was carrying nuclear weapons components was generated with the help of AI and was "entirely false," according to CNN, prompting U.S. …
TypeSafe AI emerged from stealth with Jev, a model that returns typed decisions with calibrated probabilities instead of text, reporting 70 to 500 ms end-to-end latency at $0.042 per million input tokens and claiming up …