from Hacker News

Outcome-Based Reinforcement Learning to Predict the Future

by bturtel on 5/27/25, 1:33 PM with 15 comments

• by ctoth on 5/27/25, 4:14 PM

  Do you want paperclips? Because this is how you get paperclips!

  Eliminate all agents, all sources of change, all complexity - anything that could introduce unpredictability, and it suddenly becomes far easier to predict the future, no?

• by valine on 5/27/25, 6:26 PM

  So instead of next token prediction its next event prediction. At some point this just loops around and we're back to teaching models to predict the next token in the sequence.

• by jldugger on 5/27/25, 10:25 PM

  From the abstract

  > A simple trading rule turns this calibration edge into $127 of hypothetical profit versus $92 for o1 (p = 0.037).

  I'm lazy: is this hypothetical shooting fish in a barrel, or is it a real edge?

• by amelius on 5/27/25, 9:56 PM

  Why would you use RL if you're not going to control the environment, but just predict it?

• by garbagecoder on 5/28/25, 7:45 PM

  "a couple of wavy lines"

  bzzzzz "sorry this isn't your lucky day"