Algebraic value editing works (for at least one "X vector") in language models

#AI Alignment #Mechanistic interpretability #Algebraic value edits

Ṁ36

Ṁ90

resolved May 13

Resolved

YES

ALL

https://www.lesswrong.com/posts/gRp6FAWcQiCWkouN5/maze-solving-agents-add-a-top-right-vector-make-the-agent-go

Resolves according to follow-up post.

Apr 2, 11:51am: ~~Algebraic value editing works (for at least one "X vector") in LMs~~ → Algebraic value editing works (for at least one "X vector") in language models

Get Ṁ200 play money

🏅 Top traders

#	Name	Total profit
1		Ṁ10
2		Ṁ3

Comments

3 Holders

5 Trades

Related in AI Alignment

By the end of 2026, will we have transparency into any useful internal pattern within a Large Language Model whose semantics would have been unfamiliar to AI and cognitive science in 2006?

49% chance

Will Eliezer Yudkowsky work for OpenAI OR Humane in any role before the end of 2025?

-34% 1d9% chance

See more AI Alignment questions

Related in Mechanistic interpretability

Is attention all you need? (transformers SOTA in 2027)

45% chance

Will OpenAI release weights to a model designed to be easily interpretable (2024)?

40% chance

See more Mechanistic interpretability questions

Related in Algebraic value edits

Algebraic value editing works better for larger language models, all else equal

48% chance

Will someone find a truth-telling vector which modifies completions in a range of situations by 2024-10-24?

58% chance

See more Algebraic value edits questions

AI Alignment questions