Blog
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
PetQA is a Korean QA benchmark with over 10,000 text and multimodal pairs, assessing veterinary knowledge and clinical reasoning in LLMs and LVLMs across various settings.
From arXiv cs.CL
LLMs1 min read
Research shows that overlap with internal circuits in LLMs can predict their ability to generalize arithmetic reasoning across different formats and languages.
From arXiv cs.CL
LLMs1 min read
A new framework combines structured reasoning and distance-aware calibration to enhance confidence estimation in LLMs, evaluated on diverse datasets.
From arXiv cs.CL
LLMs1 min read
The study examines internal changes in Audio LLMs when trained on data that requires audio for answer determination, highlighting how acoustic information impacts model representations and predictions.
From arXiv cs.CL
LLMs1 min read
Three open-weight LLMs were assessed for demographic misalignment across countries using Wasserstein distance; targeted fine-tuning reduced bias but redistributed it among personas.
From arXiv cs.CL
LLMs1 min read
Scale-QLoRA adapts only the native scale fields during merging, preserving the code plane and enabling accurate, fast, and lifecycle-friendly deployment of 4-bit models.
From arXiv cs.CL
Research1 min read
Enhanced capabilities in large language models may lead to more correlated behaviors, increasing systemic risk, especially when models share reasoning or misinformation environments.
From arXiv cs.AI
Research1 min read
A proposed approach estimates aleatoric uncertainty caused by input ambiguity directly from interpretation space, reducing computational costs and improving correlation with true uncertainty.
From arXiv cs.AI
Research1 min read
PerfReasoning assesses LLMs as performance reasoners and code generators, with models achieving up to 90% accuracy on reasoning tasks. Construction of models remains challenging, especially for smaller configurations.
From arXiv cs.AI
Research1 min read
MaxKernel introduces a multi-agent system for designing high-performance TPU kernels, leveraging LLMs and real-time compiler feedback. It includes human-in-the-loop, autonomous, and graph-based search paradigms.
From arXiv cs.AI
LLMs1 min read
LentEx introduces a framework for extracting implicit, contextually inferred entities using synthetic data and instruction fine-tuning of smaller LLMs, improving performance and domain generalization.
From arXiv cs.CL
LLMs1 min read
GPT-6 Astra demonstrates improved attention to detail, understanding, and output complexity, especially in 3D modeling tasks, compared to previous models.
From Simon Willison
LLMs1 min read
Blender can be used with coding agents on macOS by installing the full application and running prompts via ChatGPT Codex. This enables scene rendering with API-based control.
From Simon Willison
LLMs1 min read
Astra produces better visual outputs than Sol at similar or lower token costs, with lower input token counts and improved image quality across reasoning levels.
From Simon Willison
LLMs1 min read
GPT‑6 Astra is rolling out to select organizations and will be available via API at the same rate as Claude Fable 5.1. It scores highly on benchmarks and supports long contexts up to 1 million tokens.
From Simon Willison
LLMs1 min read
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
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
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
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
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.
