Algebraic value editing works better for larger language models, all else equal
27
174
Ṁ1.5KṀ1.5K
Jul 29
48%
chance
1D
1W
1M
ALL
Resolves according to follow-up post.
I won't bet in this market. I added m1,000 in liquidity.
Get Ṁ200 play money
Sort by:
Reopening this market because it was not resolved by the last post in the series.
https://www.lesswrong.com/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activation-vector
My guess is that future posts by Team Shard will resolve it, but other research teams may also want to give it a go.
Related questions
"Large models aren’t more capable in the long run if we can iterate faster on small models" within two years
13% chance
"Large models aren’t more capable in the long run if we can iterate faster on small models" within five years
18% chance
How can Manifold improve?
Ṁ3,446 bounty
4) The most advanced closed models will continue to outperform the most advanced open models by a meaningful margin.
92% chance
Will a Large Language Model prove an important math theorem by end of 2024?
9% chance
Is code-davinci-002 just the largest non-GPT-4 model in the GPT-4 scaling law experiment?
44% chance
Change my view, Arabic numerals are a pareto improvement on Roman numerals.
Ṁ0 bounty
When will any model achieve >=human performance on QuALITY?
Do scaling laws happen because models experience a ton of tiny phase changes which average out to a smooth curve?
55% chance
Do scaling laws happen because models experience a ton of tiny phase changes which average out to a smooth curve?
48% chance