Till Saenger

PhD student @ Princeton

[Zachary Bamberger @ZacharyBamberg1]

Pre-filling is the driving force behind STATe-of-Thouguts, (https://t.co/kCdGggcV0I), my new paper with @trsaenger , @GiladMorad12 , @ofraam , @b_m_stewart , and @amir_feder .

How is it useful beyond its typical application to Red-Teaming LLMs?

🧵👇

#NLProc #reasoning

[Zachary Bamberger @ZacharyBamberg1]

Tree-of-Thoughts & Best-of-N use high-temperature sampling to elicit diverse responses, but they're often very similar.
STATe-of-Thoughts replaces high-temperature sampling with "branching" through pre-filled textual interventions
🔼Diversity
🔼Control
🔼Interpretability
1/11🧵👇