LESSWRONG
LW

Mohammed Saeed
1010
Message
Dialogue
Subscribe

Posts

Sorted by New

Wikitag Contributions

Comments

Sorted by
Newest
No wikitag contributions to display.
Steering GPT-2-XL by adding an activation vector
Mohammed Saeed2yΩ020

Great work! I think our EMNLP 2022 Findings paper is relevant here. We construct a "Type Vector" using tokens from the LLM vocabulary and then use that as prior information for the type expected at output. We also try with text generation and view some promising results.

Reply
No posts to display.