My homeostasis paper is now available. Beyond adding to discussions of the causal faithfulness condition, it gives the first complete DCM for a Watt governor and reinforces the need for causal modelers to pay more attention to modeling complex temporal dynamics. Comments welcome.

"We're in that 'Wile E. Coyote suspended in mid-air after running off a cliff' moment with AI (Agents)" - nikete during today's call

At 🇬🇧ARIA, we’re serious about catalysing a new paradigm for AI deployment—techniques to safely *contain* powerful AI (instead of “making it safe”), especially for improving the performance and resilience of critical infrastructure. This needs a new org. Want to be its founder?

The Belief State Transformer edwardshu.com/bst-website/ is at ICLR this week. The BST objective efficiently creates compact belief states: summaries of the past sufficient for all future predictions. See the short talk: microsoft.com/en-us/research… and mgostIH for further discussion.

nikete but I believe that George Salafatinos's carefully documented solution is quite instructive x.com/GSalafatinos/s…

I had a great time talking to Nathan Nunn about the backstory of his paper "Zero‑Sum Environments, the Evolution of Effort‑Suppressing Beliefs, and Economic Development." Definitely worth checking out!

You can both bootstrap insanely successful businesses, and not have a firm grasp of elasticity of demand, would not have expected. (in the example the implied absolute demand elasticity is 0.13)

Are world models necessary to achieve human-level agents, or is there a model-free short-cut? Our new #ICML2025 paper tackles this question from first principles, and finds a surprising answer, agents _are_ world models… 🧵

arxiv.org/abs/2506.10139 "models with superhuman capabilities, it is difficult or impossible to get high-quality human supervision. To address this challenge, we introduce a new unsupervised algorithm, Internal Coherence Maximization (ICM),"

I really like the term “context engineering” over prompt engineering. It describes the core skill better: the art of providing all the context for the task to be plausibly solvable by the LLM.

I know of several entrepreneurs trying to create schools that give kids the best education in the world. But who’s applying a similar level of ambition to creating communities that are the best places in the world (for people like me and my followers) to raise children?

Have friends who can help along in vibe coding something for this? Try vibetogether.dev

The prevailing wisdom is that compute is the most important factor for frontier AI training. We think this is wrong: data is the most costly and important component of AI training. We collected estimates of revenue for major data labeling companies and compared them with the

Whenever I see things like this I am always reminded of this quote about a Buddhist monk who gets cancer, watches *Ikiru*, & realizes he’s been selfish his whole life:

anyone done kaggle.com/game-arena but freestyle chess?

If this sounds like the future of governance, def get in touch with me about grouplang.ai

Live from Palo Alto! "The Economics of Transformative AI" NBER workshop. I will be trying to live tweet (mega jet lagged--flew from HKG--, so I really hope I can keep it up) the papers presented. Watch this thread Here is the fantastic line up nber.org/conferences/ec…

Aligning an AI with human preferences might be hard. But there is more than one AI out there, and users can choose which to use. Can we get the benefits of a fully aligned AI without solving the alignment problem? In a new paper we study a setting in which the answer is yes.

I also learned at NBER AI that hardly any of my colleagues do this. Codex and Claude Code can do more than write code! Folder with relevant content + long instructions file + first generate and edit to do and check with you + go do it for two hours works for *many* tasks now!

Matt Beane Me too and if I am being honest, university wide committee is the worst possible way to get around architectural constraints. Skunk works and/or shove it down throats from the top are the only way and even they have limits.