Turing Post
Deep dives on AI history, foundation models, agents and the people behind them — plus a weekly roundup of research worth reading.
Join us for a free conference
How Recursive’s Eureka Machine helped build Flint, what the evidence shows, and where it fits on the map of recursive self-improvement.
Anthropic’s safety disclosures, the cost of enforcement, and the privacy tradeoff – plus this week’s AI news and research
Explore 15 new System One models that rethink how AI scores choices, processes multimodal information, and decides when to think longer.
What 1830s physics can teach AI about predicting what happens next
A practical guide to storing visual evidence, searching its context, and retrieving the original image only when needed.
Why decision models are emerging, how AI can help create them, and where they fit in the bigger intelligence puzzle
How AI agent harnesses, modular tools, and memory help existing models do more. Plus this week’s AI news, research highlights, and recommended reading.
10 interesting Looped Transformer types to explore how advanced looping can improve reasoning, generation, and efficiency without making models bigger
What seven builders taught us about memory, organizational knowledge, and the evidence an agent needs to question its own conclusions.
3D Gaussian Splatting is hot! Let's learn how it turns captured scenes into spatial memory, how robots search and update it, and what it takes to move from remembering a world to understanding its physics
14 open-source tools for LLM monitoring, agent observability, evaluation, and bias detection—with licenses, replay capabilities, and best-fit guidance.
How EmbeddingGemma 2 connects text, images, audio, and video for on-device search—and what it means for agent memory and our own. Plus the week’s AI research.
A practical 2026 guide to LLM benchmarks for reasoning, coding, math, factuality, multimodal understanding, and agents—with papers and limitations.
How World Action Models connect predictions of the physical world with robot actions, how they differ from VLAs, and whether imagining the future actually makes robots better
SkillOpt, SkillAdam, SkillRefiner, skill-up: agents are getting tools to improve their own instructions. The interesting question is what they should learn, where that knowledge belongs, and whether another agent can…
Reflection’s Beam trails leading Chinese coding models. Explore the benchmark gaps, Ataraxos, Context Language Models, and this week’s AI news.
we discuss AI agents, world models, education, recursive self-improvement, and why he worries about trusting AI’s leaders with so much power.
What suppliers learn by building physical AI—and whether that knowledge becomes better products and useful customer results.
Your agent can have its own computer. OpenAI’s DevDay announcements show who gets access, what it can do, and where subscriptions set the limits.
Compare Wan2.2, Wan2.1, HunyuanVideo 1.5, LTX-2, CogVideoX, Mochi 1, Kandinsky 5.0, and Stable Video Diffusion.
Official plugins for Claude Code, Cursor, Gemini CLI, and VS Code bring MongoDB best practices into the workflow.
Let's learn the architectures that teach AI to imagine and act
Lessons from citizen diplomacy for the AI era, alongside this week’s AI news, robotics library, and research on world models and agents.
September 2026 robotics: Helix 2.5 in 30 unseen homes, Digit 5 lifting 22.7 kg, and why Skild's 66% is not a completion rate.
Why Jev – TypeSafe’s System One Model – caught developers’ attention, which older ideas it combines, and what you can already build without it
What to change in its memory, instructions, tools, or model – and how to test whether the fix holds. Enterprise lessons from Dreamforce for your own workflows. AI Builds AI, Episode 2.
How models can keep learning during inference – and why test-time training could be a key to overcome the limitations of agents and world models.
