christian
@curious_vii
driven by compression progress
Technology is good so long as it allows you to go deeper into reality, not flee from it
We're in this interesting liminal space right now with agents where the complexity of the tasks and the test time compute necessary to tackle them isn't quite long enough to really spin up initiatives in parallel. You still more or less have to tackle problems in serial, but…
AI as a means of decentralizing verified expertise: How value is captured through the process (and by whom) is as interesting as the means by which this process happens (generation of new info through manual experimentation and measurement, RL, distillation) — digital property?
Aren't tools just ultimately CRUD actions and incremental verification? Complicated business logic maybe for the latter to prevent the model from going off the rails, and the former for continuous context curation (both will give you a slight edge even as models get better).