Pavel @spavel.bsky.social · 30/05/2025The fact of the matter is that you cannot eliminate *measuring the trailing indicator* from your approach to measuring success, if your success is defined by trailing indicators. Your hypothesis is not enough. 2934
Pavel @spavel.bsky.social · 30/05/2025This process is modeled with "learning loops." Single loop learning is merely, is the task Done? If not, keep doing it. Double-loop learning checks, is the output of the task helpful? If not, choose a different task. Triple-loop learning asks, are we framing "helpful" right? What do we WANT?Pavel@spavel.bsky.socialSingle-loop processes are not able to ask "is the thing we are doing working" but perhaps more importantly, they cannot ask "is the thing we are working towards worthwhile?" In all 3 cases, the answer of AI boosters is the same: we are building AI so that we can have AI. 312320
Pavel @spavel.bsky.social · 30/05/2025There's a fourth loop that was modeled in 2020 based on responses to COVID - "are we learning right?" In other words, is our approach for modeling the world and bringing in data allowing us to define what our goal should be effectively? 1956
Torbjörn Björkman @torbjornbjorkman.eurosky.social · 30/05/2025As a Scientist Proper I should probably write a lot of snark here, but the sad truth is that research funding is governed by depressingly similar dynamics. 110
@spavel.bsky.social
