Katherine Tian
@kattian_
cs/stat @harvard, working on calibration & factuality of LLMs, prev @GoogleAI tensorflow, golden state @warriors fan
ID:951151284735827968
https://kttian.github.io/ 10-01-2018 17:58:32
127 Tweets
712 Followers
494 Following
Come by the #NeurIPS2023 Instruction Following workshop (room 220-222) to see our work on:
*Emulated fine-tuning*: RLHF without fine-tuning!
*Fine-tuning for factuality*: how to fine-tune LLMs directly for factuality, reducing hallucination by >50%
RIGHT NOW!!!
DPO is a runner up for NeurIPS outstanding paper. π
Big congrats especially to the students Rafael Rafailov Archit Sharma Eric & the other awardees.
If you haven't learned about DPO already, check out the oral & poster π on Thurs afternoon.
Come see Katherine Tian @ ICLR π¦πΉ's work on the ability of RLHF'd LLMs to *directly verbalize* probabilities (yes that's right as tokens) that are actually pretty-well calibrated! (Usually than the log probs!!! π€―π€―π€―)
Poster 14B
R I G H T N O W until 3:30 SG time!!
LLMs fine-tuned with RLHF are known to be poorly calibrated.
We found that they can actually be quite good at *verbalizing* their confidence.
Led by Katherine Tian @ ICLR π¦πΉ and Eric, at #EMNLP2023 this week.
Paper: arxiv.org/abs/2305.14975