Who we follow in AI

Model launches, safety work and company news, straight from the lab.

How Anthropic builds with Claude: agent design, tool use, evaluation.

Releases, system cards and product notes from OpenAI.

Open-weight and commercial model releases out of Paris.

Google's own account of what it shipped across its AI products.

Research announcements and model releases from DeepMind.

The LFM family: small models built to run on the device in front of you.

Open models and datasets trained on permissively licensed text.

Low-latency speech models, and the latency work that makes them usable.

Task-specific small models, and what they cost to run.

Fastino's papers on small-model training and extraction.

Chatbot Arena, SGLang, and the evaluation work around both.

Distilling large models into small ones you can own outright.

Open-weight small models and the merging techniques behind them.

Olmo, Tulu, and the fully open research stack underneath them.

Open research on interpretability, training and evaluation.

Practical research from the lab Jeremy Howard co-founded.

Apple's published research, much of it about running models on device.

Independent evaluation of frontier model capability and autonomy.

Benchmarks for guardrails and automated red-teaming of deployed models.

Berkeley's AI research blog, written by the authors themselves.

Research announcements from across Microsoft's labs.

Context engineering and harness design for coding agents.

Building and measuring autonomous coding agents, from the Devin team.

Memory for agents that learn, from the team behind MemGPT.

Open reinforcement learning at scale: environments, verifiers, prime-rl.

Developer-facing posts on Gemini APIs, tooling and agent frameworks.

The open-source serving engine, and the release notes that come with it.

Inference performance engineering, written up in public.

Serving, fine-tuning and function calling on hosted open models.

Sizing, benchmarking and running inference across mixed GPU fleets.

Running models on your own machine, and which ones just landed.

Post-training with reinforcement learning, written for people doing it.

MAX, Mojo, and the argument for rebuilding the inference stack.

Structured outputs, schema-aligned parsing and function calling.

Search and retrieval engineering, down to the object store.

Pydantic AI and Logfire: typed agents, and evaluating them.

Agent framework releases and the evaluation practice around LangSmith.

Embedding and reranking models, plus how they measure retrieval.

Kernel-level and deployment notes from the people who make the hardware.

Open-model serving and fine-tuning, with the inference research behind it.

The community blog: releases, recipes and model cards worth reading.

Nathan Lambert on post-training, RLHF and the open model ecosystem.

Sebastian Raschka on LLM architecture and training, from first principles.

Lilian Weng's long-form explainers on agents, safety and reinforcement learning.

New cs.AI and cs.CL preprints, filtered down to agent work.

What engineers are arguing about, filtered to agents and LLMs.