Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Researchers from Aalto University and the University of Waterloo introduced Locket, a feature-locking technique for large language models that enables pay-to-unlock schemes by restricting specific model capabilities. The…
A developer argues that balanced ternary (-1, 0, +1) could replace binary for AI hardware, citing 20× model compression, 3× inference speedup, and 8× power reduction. Microsoft's BitNet b1.58 demonstrated ternary weights…
A developer released a GPT tool that provides free URLs for AI-generated content, aiming to simplify sharing and access to such materials.
A developer compares four dominant alignment methods—RLHF, DPO, IPO, and KTO—for fine-tuning large language models, detailing their mathematical formulations, data requirements, and practical tradeoffs. The analysis high…
A developer released an open-source CLI tool called whoburnedmore that aggregates token usage and cost estimates from AI coding tools like Claude Code, Codex, and Cursor into a single dashboard, and includes a public lea…
A developer built a multi-provider AI API fallback system after a single-provider outage caused an hour of downtime for their app. The system uses weighted random selection, circuit breakers, and exponential backoff to a…
A developer cut their LLM API costs by over 70% by building a thin proxy layer that routes requests to cheaper models based on complexity, without changing any application code. The proxy, written in Node.js, uses OpenAI…
Anthropic has indefinitely delayed subscription updates for Claude Agent SDK usage in third-party tools like Conductor, allowing users to continue using their Claude plans without additional limitations. The changes were…
Apple unveiled Siri AI, a major overhaul of its voice assistant, as part of iOS 27 on June 8. The new system adds personal context understanding, onscreen awareness, and third-party model extensions, with developer testi…
A developer argues that agent evaluations must run in CI to block regressions before deployment, not as post-hoc dashboards. The post introduces agent-eval for scoring outputs and AgentLens for execution traces, advocati…
A developer with ADHD describes using AI as an executive function prosthetic, compensating for deficits in working memory, task initiation, and other cognitive processes. The developer uses Claude Code to offload mental …
A developer with ADHD built 114 Claude Code skills, five of which directly target specific executive function deficits. The skills, including `/adhd-task-triage` and `/mirror`, provide deterministic workflows that bypass…
A developer with ADHD describes how CLAUDE.md, a Markdown file automatically read by Claude Code at the start of every session, serves as a cognitive prosthetic that preserves working memory across context switches. The …
Inithouse, a studio running parallel product experiments, built Be Recommended, a tool that checks brand visibility across AI models like ChatGPT and Perplexity. The team encountered four key technical mistakes: treating…
The US government treated Anthropic's LLM as a munition by ordering unprecedented restrictions on access to its Fable 5 model, effectively barring foreign nationals globally. The move, prompted by a report from Amazon, r…
Meta launched AI Mode on Facebook on June 15, replacing link-based search with AI-synthesized answers from public posts, Groups, and Reels, powered by its new Muse Spark model. The Muse Spark API is in private preview wi…
A Nature npj Digital Medicine paper published June 16, 2026 introduces MedSAFE, a decision-theoretic framework for evaluating LLM abstention in healthcare, identifying uncertainty-driven and safety-driven abstention. The…
Anthropic's Claude Fable 5 scored 161 on the Epoch Capabilities Index, surpassing OpenAI's GPT-5.5 Pro at 159, with dominant performances in math and software engineering benchmarks. The model achieved 88% on FrontierMat…
Large language models default to building custom implementations rather than using existing libraries, making AI-generated code over-engineered and expensive to review. Rewriting such code has become cheap, shifting deve…
A developer introduced DeepFork, a five-phase Claude Code skill that reverse-engineers open-source repositories to extract design intent rather than code. The tool uses an AST-based knowledge graph to surface core abstra…