Loading
90 days of the enterprise, pro and team plans free, no card needed. See the plans
Blog
Posts older than six months. They stay readable at their original address; models, prices and APIs they describe may have changed since.
AI1 min readarchive
Expanded preview access for Microsoft Discovery brings new enterprise-grade, agentic AI capabilities for research and development teams. The post Microsoft Discovery: Advancing agentic R&D at scale appeared first on Microsoft Azure Blog.
LLMs1 min readarchive
We’ve fundamentally transformed Facebook Groups Search to help people more reliably discover, sort through, and validate community content that’s most relevant to them. We’ve adopted a new hybrid retrieval architecture and implemented au...
Research1 min readarchive
Researchers released GRASP to fix fragile planning in large world models. They use gradient-based optimization with parallel time steps and added stochasticity.
LLMs1 min readarchive
A learning-oriented workflow for understanding new open-weight model releases
Research1 min readarchive
Introducing CRUX, a new project for evaluating AI on long, messy tasks
AI1 min readarchive
Discover how cloud cost optimization adapts in the age of AI, with best practices for managing spend, improving efficiency, and maximizing value. The post Cloud Cost Optimization: Principles that still matter appeared first on Microsoft ...
LLMs1 min readarchive
How coding agents use tools, memory, and repo context to make LLMs work better in practice
AI1 min readarchive
We often lack the tools for the job, even if the AI is capable enough
Models1 min readarchive
Mistral AI launched Spaces, a command-line tool that scaffolds projects and deploys them. It includes plugins and context files to help coding agents work without errors.
LLMs1 min readarchive
A Round Up And Comparison of 10 Open-Weight LLM Releases in Spring 2026
Research1 min readarchive
Quantifying the capability-reliability gap
LLMs1 min readarchive
And an Overview of Recent Inference-Scaling Papers
LLMs1 min readarchive
LLMs1 min readarchive
A 2025 review of large language models, from DeepSeek R1 and RLVR to inference-time scaling, benchmarks, architectures, and predictions for 2026.
LLMs1 min readarchive
A curated list of bookmarked LLM research papers from July to December 2025, covering evaluation and reinforcement learning methods.
LLMs1 min readarchive
Label some data, align LLM-evaluators, and run the eval harness with each change.
AI1 min readarchive
Photo: Ketut Subiyanto
Today, we're previewing the fastest way to run Ollama on Apple silicon, powered by MLX, Apple's machine learning framework.
Models1 min readarchive
Mistral AI released Voxtral TTS, a 4B-parameter speech model with nine languages. It offers low latency and high-quality voice generation via API.
LLMs1 min readarchive
From MHA and GQA to MLA, sparse attention, and hybrid architectures
Models1 min readarchive
Mistral AI announced Forge to train models on proprietary data. Partners include ASML and the European Space Agency.
Models9 min readarchive
Mistral AI released Mistral Small 4 as the first single model combining reasoning, coding, and multimodal capabilities. It uses a hybrid architecture with Apache 2.0 licensing and NVIDIA H100 support.
Models9 min readarchive
Mistral AI became a founding member of the NVIDIA Nemotron Coalition to build open models. They are releasing Mistral Small 4 and co-developing new frontier models with NVIDIA.
Research1 min readarchive
Berkeley AI Research released SPEX and ProxySPEX to find influential interactions in large models.
Preface This essay argues that rational people don’t have goals, and that rational AIs shouldn’t have goals. Human actions are rational not because we direct them at some final ‘goals,’ but because we align actions to practices[1]: netwo...
LLMs1 min readarchive
Ollama now supports subagents and web search in Claude Code.
Research1 min readarchive
Applying the AI as Normal Technology framework to legal services
ollama launch is a new command which sets up and runs coding tools like Claude Code, OpenCode, and Codex with local or cloud models. No environment variables or config files needed.
LLMs1 min readarchive
Ollama versions 0.14.0 and later support the Anthropic Messages API, allowing local models to run with tools like Claude Code.
LLMs1 min readarchive
OpenAI's Codex CLI now supports open models via Ollama, defaulting to gpt-oss:20b. A minimum context window of 32K tokens is recommended for functionality.
Research1 min readarchive
Researchers created a method to measure information in imaging systems using only noisy measurements. Their approach predicts performance across color photography, radio astronomy, lensless imaging, and microscopy.
AI1 min readarchive
With the right tools, AI can accomplish impressive things
And why Nano Banana Pro is such a big deal
LLMs1 min readarchive
DeepSeek released V3.2 with a non-standard sparse attention variant and RL updates, following the earlier V3.1 model.
As AI advice becomes more important, we are going to need to get better at assessing it
LLMs1 min readarchive
Sebastian Raschka outlines alternatives to standard autoregressive transformers, including linear attention hybrids and text diffusion models. These approaches aim to improve efficiency or modeling performance beyond the classic multi-head attention mechanism.
Research1 min readarchive
Researchers introduced a reinforcement learning algorithm that avoids temporal difference learning. They scaled this approach to complex goal-conditioned tasks using a divide and conquer strategy.
