Will we have any progress on the interpretability of State Space Model LLM’s in 2024? | Manifold

Will we have any progress on the interpretability of State Space Model LLM’s in 2024?

Plus

12

Ṁ312

Jan 1

71%

chance

1D

1W

1M

ALL

State Space Models like Mamba introduce new possibilities, as States are a new object type, a compressed snapshot of a mind at a point in time which can be saved, restored, and interpreted. But a cursory search didn’t turn up any work on interpreting either States or State Space Models.

This resolves Yes if research comes out that makes any significant interpretability progress into a state space large language model. I will not bet on this market.

This question is managed and resolved by Manifold.

#Mechanistic interpretability

Get

1,000

and

3.00

Sort by:

Putting a 1k limit order for "YES" at 50% if anyone wants to take it.

I'm watching this but the notion of "any significant ... progress" is too vague for me to trade.

Related questions

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?

Will an LLM Built on a State Space Model Architecture Have Been SOTA at any Point before EOY 2027? [READ DESCRIPTION]

Will there be a gpt-4 quality LLM with distributed inference by the end of 2024?

Will OpenAI release an LLM moderation tool in 2024?

Will mechanistic interpretability be essentially solved for GPT-2 before 2030?

Will the state-of-the-art AI model use latent space to reason by 2026?

Will there be an open source LLM as good as GPT4 by the end of 2024?

Will a lab train a >=1e26 FLOP state space model before the end of 2025?

Will the best LLM in 2024 have <1 trillion parameters?

Will a LLM/elicit be able to do proper causal modeling (identifying papers that didn't control for covariates) in 2024?

Related questions

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?

Will the state-of-the-art AI model use latent space to reason by 2026?

Will an LLM Built on a State Space Model Architecture Have Been SOTA at any Point before EOY 2027? [READ DESCRIPTION]

Will there be an open source LLM as good as GPT4 by the end of 2024?

Will there be a gpt-4 quality LLM with distributed inference by the end of 2024?

Will a lab train a >=1e26 FLOP state space model before the end of 2025?

Will OpenAI release an LLM moderation tool in 2024?

Will the best LLM in 2024 have <1 trillion parameters?

Will mechanistic interpretability be essentially solved for GPT-2 before 2030?

Will a LLM/elicit be able to do proper causal modeling (identifying papers that didn't control for covariates) in 2024?

© Manifold Markets, Inc.•Terms + Mana-only Terms•Privacy•Rules