In case anyone follows what I mean by the word "distillation".
Remember I have this thing, a second brain? The one Claude, Codex, Cursor and the rest are wired into.
And since I cannot put an infinite amount of data into it, I try to distil the data and put the already distilled version in. That is, I put already distilled data into my second brain.
And I ran a study: Terra turned out to be much worse than Sonnet for distillation. Because I am saving tokens and trying to do it not with Sol and Opus, but with a cheaper model.
Well, yes, of course: on the merits Opus wins by a wider margin. But Terra turned out worse than Sonnet. Which, in principle, was rather what I suspected.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20261026.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20261026.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.
🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI
Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.
Top comments (0)