SIGNAL//SYNTH
Ai

Designing How AI Grows — Tom McGrath

aired Sep 02, 2026 · 100.0m
Signal
79.0/ 100
High signal
confidence 0.99
Orig48.7
Actn100.0
Dens100.0
Dpth100.0
Clty71.2
Summary

which is like Neil Nander says, SAEs are dead. I think interpretability is, I think of it as a natural science, you know, like physics, biology, chemistry, but it's a natural science that you do completely on the computer.

Why listen

It goes beyond the title with direct discussion of like, know, sort, including: You can use that for the intro.

Key takeaways
  1. 01And so this means that like, we should be able to kind of, again, speed run science once we have agents that can do kind of experimental work for us and the ability to do that expe
  2. 02Like real scientific work to do no barrier to research scientific work to be done and it's sort of gated on both empirical data collection and theory building dario amodi so he had
  3. 03and where on that spectrum do we want the foundation models to sit and when we build applications, where do we want those to sit?
Best for
research-minded practitioners comparing model behavior