Jonathan Berant @jonathanberant.bsky.social · 06/03/2026Models also generalize to coefficients that never occurred at training time! 100
Jonathan Berant @jonathanberant.bsky.social · 06/03/2026We show prompted models fail to change their behavior for different cost coefficients. But training leads to steerable models that get high reward – when increasing the cost of turns, models ask less questions. When the price of tokens is high, models avoid multi. answers 100
Jonathan Berant @jonathanberant.bsky.social · 06/03/2026We test this in two question answering settings, (a) when the AI assistant uses parameterized knowledge and (b) When the AI assistant has access to privately-held information. 100
Jonathan Berant @jonathanberant.bsky.social · 06/03/2026In this work, we train steerable models – models take as input in their prompt scalar values that reflect different costs (cost of each token, cost of each turn) and are trained to maximize their cost-penalized accuracy through collaborative self-play with a user simulator. 110
Jonathan Berant @jonathanberant.bsky.social · 06/03/2026Newish work (arXived in December): Prompts can be ambig., but handling ambiguity is context/user dependent. Sometimes the right thing is to ask a clarifying question, sometimes to give multi. answers, and sometimes to just guess. Can we train steerable models that change their strategy per context? 121
