A distinct set of crafts for human-generative AI collaboration is beginning to emerge. This has me thinking again about supervisory craft and the distinction between generalists and specialists. Will we all become generalists, or will increasing AI capacity to model and

But it makes a difference if I talk about bacteria (or AI) in human terms — or if we develop new terms that cover humans, bacteria, AI, etc as variants of a general theme. these themes must be discovered, not merely abstracted from humans. otherwise we just have more of the same.

📢Thrilled to introduce ATLAS 🗺️: scaling laws beyond English, for pretraining, finetuning, and the curse of multilinguality. The largest public, multilingual scaling study to-date—we ran 774 exps (10M-8B params, 400+ languages) to answer: 🌍Are scaling laws different by

Received this milestone award from @OpenAI for passing 10 billion API tokens. This volume represents significant R&D investment in synthetic data generation, model drift detection and classifier testing. Looking forward to moving more of this work into production.

Time for Line 37?

Two things can be true at the same time: 1. Without additional advances, LLMs won't get us to general intelligence. 2. Even without additional advances, LLMs will radically transform the economy.

Fear-based marketing vs hope-based marketing. Which would you choose to build with, and which would you choose to live by?

Pattern: Major monetary transformations during financial crisis Historical precedent suggests major shifts in monetary systems often occur during crises when resistance is weakest. For example, gold confiscation in 1933 and the Nixon shock ending the gold standard in 1971 were

Chiral heterogeneity seems like nature's equivalent of a strange protocol. In the Protocol Field framework, a 'strange protocol' is a novel, emergent pattern of interaction arising from vast networks of AI agents. While initially opaque to humans, they can be discovered and

Open Protocols Can Prevent AI Monopolies https://t.co/lVUGLNf4ys

By 2028, AI Ethologist will emerge as a new aspirational job title. Not necessarily technical, but adjacent to AI in a way that feels thoughtful, human, and LinkedIn-friendly.

The commentarial protocol

I recently read that we make sense of time through emplotment by turning time into stories. Social media is deliberately designed to undermine this. The same perhaps applies to optimised workflows. Social media is particularly effective because "attention engineers" aim to

In this vein, the IMO results show how deep learning that uses symbolic tools (either in learning or solving) can achieve performance competitive with elite humans even in these formal domains.

I've been thinking about the implications of Cloudflare's 'pay per crawl' solution: Pattern: Resource Metering The allocation of resources through metering and rights systems is a recurring pattern. Control systems emerge to manage distribution when resources become scarce due

Concepts are to frameworks as precepts are to protocols.

Open-source implementation of AlphaEvolve https://t.co/XbGOocsSoc

Pattern: Variation, Selection and Retention Evolutionary systems generate variety through experimentation, apply selection pressure to filter what works, and retain successful adaptations. As open-ended AI systems emerge, their capacity to generate endless variety creates

This protocol field plots the same underlying protocol from two distinct observer positions. The protocol is the operational logic of a hypothetical open-ended AI system. From its own internal perspective, the AI operates in the top left of the field. Its processes are

The "Time to Answer" (ttA) is the time required to analyse a problem or system and find an answer to a specific question. Recent METR research finds the length of tasks AI can autonomously complete is doubling every 7 months. As AI-mediated ttA plummets, the source of value will