Large Language Model News

Sakana AI CEO David Ha argues that orchestrating many small models will outperform frontier giants, unveiling Fugu, a model orchestration system that routes tasks across specialized models. The Tokyo-based startup, backe…

A patient used Claude Code with Opus 4.8 to analyze their shoulder MRI after receiving a diagnosis of a Grade III partial-thickness tear and extensive treatment recommendations. The AI found no tear, contradicting the hu…

Oded Rechavi of QED Science claims that papers not in the top 1% of their QED score 'suck', but a review of the company's white paper finds insufficient evidence that the AI-generated score accurately measures scientific…

A Hacker News user asked for recommendations for a local LLM under 2 billion parameters that consumes less than 3GB of RAM, seeking models suitable for a local project with limited resources.

A developer team accidentally built a multi-agent system when a single-agent monolith grew unmanageable. They now use patterns like supervisor-worker, pipeline, and event-driven coordination to manage complexity, with st…

A new platform called The Collaboration Layer for AI Intelligence has launched, offering tools for creating, sharing, and evolving AI prompts and skills with version control. The platform includes features for API docume…

A developer built Branch Agent, a system that applies Git-style branching to LLM conversations, enabling users to fork, branch, and merge AI conversations with different models, prompts, and providers in each parallel ti…

A 2026 guide compares AI coding subscriptions, highlighting a shift to credit-based pricing. Platforms like OpenCode, Kilo, and GitHub Copilot offer varying plans, with OpenCode Go at $10/month and Kilo's free BYOK optio…

A developer argues that programming is fundamentally broken due to complexity and boilerplate, and that large language models (LLMs) are not the solution because they produce inconsistent code and create dependency. Inst…

OpenAI released the system card for GPT-5.6, detailing three model sizes—Sol, Terra, and Luna—with Sol described as a 'step function better' than GPT-5.5 but still short of Mythos. The card highlights improved performanc…

Employees at fintech firm Slash and consulting giant Accenture are burning through expensive AI tokens on trivial tasks like creating a meme-themed video game and converting PDFs to PowerPoints, exposing the gap between …

Engineers at narrivo highlight the problem of overfitting in large language models, particularly in Retrieval-Augmented Generation (RAG) evaluation. They explain that overfitting causes models to perform well on test dat…

A developer building an Anime AI assistant for fans is seeking advice on missing features and implementation strategies, noting that existing chatbots on platforms like Character.AI attract global traffic while smaller s…

A developer discovered that using OpenAI's tiktoken tokenizer to count tokens for Claude models leads to a 15-20% undercount, causing inaccurate cost estimates and context budgets. The correct approach is to use Claude's…

A developer mapped ten distinct levels of AI skill construction, from a single prompt file to a full business closure system, based on a year of building AI agent skills for enterprise clients. The developer found that m…

The team Risers built MemoCode AI, an enterprise AI agent with persistent memory for software engineering assistance. The system remembers past conversations and project context to help developers write, debug, and impro…

Anthropic's growth team is hiring more product managers after Claude Code tripled engineering output, shifting the bottleneck from coding to deciding what to build. The AI coding tool compressed the traditional engineeri…

The US government ordered OpenAI to restrict the release of its new GPT-5.6 model, allowing access only to government-approved partners. The directive, issued by the Office of the National Cyber Director and the Office o…

Newer Claude models use more tokens but cost less per task solved, according to an instrumented evaluation of Claude Code on 10 Terminal-Bench tasks. The analysis found that while token usage increased, the overall cost …

Prosecutors used ChatGPT logs as evidence in the arson trial of Jonathan Rinderknecht, accused of starting a deadly wildfire in Los Angeles on New Year's Day 2025. The logs showed Rinderknecht asking the chatbot about fi…