Blog
Posts tagged llm
Daily notes on new models, LLM releases, agent frameworks and AI research, written from the sources we follow and delivered as a newsletter every day.
Get the daily issue
Every new post of the day, in one email. Confirmation required.
LLMs1 min read
llm-anthropic 0.28: Claude Fable 5.1 and New Features
llm-anthropic 0.28 introduces Claude Fable 5.1 with default reasoning traces and a new exception for refusals.
From Simon Willison
LLMs1 min read
The Modern CUDA Toolbox in Practice: A Step-by-Step Optimization Walkthrough
NVIDIA's CUDA remains central to GPU-accelerated computing, supporting scientific simulations and AI training. This article provides a detailed optimization process for CUDA workflows.
From NVIDIA technical blog
LLMs1 min read
Gemini 3.8 Flash Released with Multiple Thinking Levels
Google released Gemini 3.8 Flash, offering low, medium, and high thinking levels, with improvements in speed, cost, and HTML/JavaScript capabilities for developers.
From Simon Willison
LLMs3 min read
Claude's new system prompt really doesn't want to reproduce song lyrics
Anthropic publish the system prompts for their Claude consumer applications (Claude.ai and the Claude mobile apps - sadly not for Claude Cowork or Claude Code). I love that they do this, and that they share not just the current prompts b...
From Simon Willison
LLMs1 min read
IBM Time Series Models for Real-Time Analytics on Confluent Platform
IBM has integrated time series models with Confluent for real-time intelligence, enabling continuous data processing and analysis.
From Hugging Face blog
LLMs1 min read
Paint.NET Implements Internal Reverse-Engineered Direct2D for WINE
Paint.NET now uses an internal, reverse-engineered version of Direct2D on WINE, developed by Claude, to bypass incomplete support. The code is unreviewed and relies on trust.
From Simon Willison
LLMs1 min read
Claude Fable 5.1 introduces multiple reasoning levels and improved benchmarks
Claude Fable 5.1 achieves a 52.6% score on the Terminal-Bench-Science 0.1 benchmark and features five reasoning levels, affecting task performance and output length.
From Simon Willison
LLMs1 min read
BenchMIRT analyzes what LLM benchmarks actually measure
BenchMIRT investigates the alignment between LLM benchmark tasks and real-world capabilities, highlighting potential discrepancies for model evaluation.
From Hugging Face blog
LLMs1 min read
Meta releases Muse Glimmer, a 30B open-weight dense model for local AI workflows
Meta's Muse Glimmer is a 30-billion parameter open-weight dense model with a 120K+ context window, designed for local AI applications and open source deployment.
From NVIDIA technical blog
LLMs1 min read
OpenAI Codex includes LibreOffice binaries in cache folder
The OpenAI Codex desktop app stores LibreOffice binaries in its cache, including full installations and usage skills, which may impact model deployment and integration strategies.
From Simon Willison
LLMs1 min read
NVIDIA Nemotron enables adaptive agentic cybersecurity systems
NVIDIA announced Nemotron, an AI system designed for adaptive, agentic cybersecurity applications, capable of coordinating complex objectives over long periods.
From NVIDIA technical blog
LLMs1 min read
GeoJSON Map Viewer for Displaying and Exporting Geospatial Data
A new tool enables displaying GeoJSON files on maps and exporting them as PNGs, with data sourced from government boundaries using AI assistance.
From Simon Willison
LLMs1 min read
Tarn Adams discusses dwarf behavior and AI misconceptions
Tarn Adams, co-creator of Dwarf Fortress, emphasizes that dwarf behavior, not AI, is being referenced, highlighting the distinction between behavior and artificial intelligence in game design.
From Simon Willison
LLMs1 min read
datasette-mcp 0.2 updates execute_sql output to array of objects
The new version changes execute_sql's 'rows' output from an array of arrays to an array of objects, aiding models in tracking column mappings. It now depends on mcp>=2.1.1.
From Simon Willison
LLMs1 min read
Python 3.15.0 candidate 2 released ahead of final version
Python 3.15.0 candidate 2 is available, with only bug fixes allowed before the final release scheduled for October. Maintainers are encouraged to prepare their projects and publish wheels on PyPI.
From Simon Willison
LLMs1 min read
Hugging Face releases @huggingface/kernels with 200+ WebGPU kernels for local AI
Hugging Face announced @huggingface/kernels, offering over 200 WebGPU kernels designed for local AI processing, enabling efficient model inference on compatible hardware.
From Hugging Face blog
LLMs1 min read
Wrapture Extends Monkey-Patching for Testing and Tracing in Python
Wrapture builds on wrapt to enable tracing and overriding functions for testing and observability, with support for OpenTelemetry and configuration-based tracing in Python projects.
From Simon Willison
LLMs1 min read
Kākāpō population increases with record breeding success
The kākāpō population has grown with the addition of juvenile chicks from this year's record breeding season, indicating successful conservation efforts.
From Simon Willison
Posts are drafted from public feeds by models OpenSmartRoute routes to - the same router, skill and metering customers use - and always link to the original source. Corrections: support.
How this blog is made
Every post is a routed request
Each feed entry becomes one request to OpenSmartRoute: the router picks a model with a cost-weighted objective, the editorial-writer skill is layered on the prompt, and the outcome trains the learners - the same pipeline available to every workspace.
Open any post to see which target answered, its confidence, the alternatives and what the request cost. Run the same pipeline yourself: register feeds in the operator console, map a small model under Providers, or call POST /api/v1/route with execute: true.