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 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 …
Anthropic filed a motion to dismiss the copyright claims of the Bartz-class opt-outs in the Shakespeare v. Anthropic suit, arguing the claims are barred by the Copyright Act's 3-year statute of limitations. Anthropic ass…
A solo developer built Docloom, a tool that uses an LLM to write API documentation while a separate parser fact-checks the generated content against the actual codebase, preventing the model from inventing endpoints. The…
Open WebUI released version 0.11.1, adding tool-call approvals, an ask_user builtin that lets a model pause and ask the user a question, a terminal file browser, /model slash commands, smarter chat search, a hover-previe…
A developer published a tutorial showing how to build a Python-based conversation engine that gives a large language model the appearance of memory by resending the full message history with every API call. The script ma…
AIPerf is a dedicated benchmarking tool that measures LLM inference performance by generating concurrent request loads, according to a first-person account of using it to test deployments. The account states that manual …
A developer released Reader, an open-source native AI reading workspace for macOS 15 or later, built with SwiftUI, PDFKit, and WebKit, that imports PDF, EPUB, and TXT files and keeps books, highlights, notes, and AI chat…
A developer benchmarked local LLM memory usage on a Ryzen desktop with 32 GB of RAM and a 12 GB RTX 3060, finding that a 4.9 GB llama3.1:8b model reserved 7.0 GB of VRAM at a 32,000-token context window — a 43% "context …
Six community-built approximations of Jev appeared within 48 hours of the discriminative "System 1" decision model's launch, after its launch video drew 36 million views in two days. The clones include Laya (421M-param M…
A freelance writer who built more than 600 tools for AI agents describes experimenting with ways to cut the cost of tool selection, including tool search, frozen and dynamic tool sets, and capability-group recommendation…
A developer demonstrated that wrapping the Jev confidence-scoring API as a Google ADK tool lets a director_agent self-check video storyboard segments and regenerate captions that don't fit their allotted duration, addres…
Anthropic CEO Dario Amodei published a 3,800-word essay on September 12 urging the AI industry to voluntarily slow frontier model releases, but six days later Reuters reported the company was weighing an accelerated roll…
A post explaining how large language models work states that once an LLM finishes training its weights are strictly read-only, so the model does not learn, adapt, or rewrite its own code while chatting with a user. The p…
A San Francisco County Superior Court judge allowed three of Reddit's five claims against Anthropic to proceed in Reddit's lawsuit accusing the company of illegally scraping user comments to train its Claude chatbot, whi…