Weekly Links 42

Tangle

This short but practice-based article on agentic loop design is about methods that can be followed in order to avoid unintended consequences—agentic systems act, so we need to take some measures like sandboxing in order to avoid their possible flawed outcome. A must-read from someone who is knowledgeable in the field.

https://simonwillison.net/2025/Sep/30/designing-agentic-loops/

What a better approach to reveal the way models are trained and work. I’ve recently read Karen Hao’s book “Empire of AI” and in the references I found this link. Many misconceptions can be revealed while browsing the project, but I think that the main gain from this examination is that we should stand very critically against the methods that are used in order foundational models to be trained, as well as to request accountability both from the AI researchers and also from “empires of AI”.

https://knowingmachines.org/models-all-the-way

A really interesting interview of Brian Eno talking about generative music - something that he’s been working on well before the advent of generative AI, as well as the value of art in general. It struck me the way he explained his experiments with music, especially when he explained how he tried to generate music that would be slightly different every time it was played - a way to transform music into theater—a unique experience. It also reminded me of the way that cyberneticians view the world: not through the final goal, but instead by means that act as enablers of different and not always predictable final results.

https://www.nytimes.com/2025/10/03/opinion/ezra-klein-podcast-brian-eno.html