Are Mixture of Expert (MoE) transformer models generally more human interpretable than dense transformers?

1kṀ1318

Dec 31

45%

chance

ALL

Can resolve PROB in case of high uncertainty, resolves YES or NO based on the best understanding we have at the end of 2025.

If you'd like to lock in your priors or intuitions, without making trades, here's a free poll:

AI Safety

Technical AI Safety

Get

1,000

to start trading!

Comments

13 Holders

34 Trades

People are also trading

Is gpt-3.5-turbo a Mixture of Experts (MoE)?

84% chance

Will the most capable, public multimodal model at the end of 2027 in my judgement use a transformer-like architecture?

63% chance

Will mechanistic/transformer interpretability [eg Neel Nanda] end up affecting p(doom) more than 5%?

10% chance

By EOY 2025, will the model with the lowest perplexity on Common Crawl will not be based on transformers?

10% chance

People are also trading

Related questions