Will an open-source LLM beat or match GPT-4 by the end of 2024?
➕
Plus
32
Ṁ3247
Jan 1
81%
chance

If any open-source model eclipses or achieves GPT-4 quality within a 1% win rate on AlpacaEval using any evaluator (https://tatsu-lab.github.io/alpaca_eval/) by January 1st, 2025, this question resolves to "YES." Otherwise, it resolves to "NO." If GPT-4 is open-sourced by January 1st, 2025, the question resolves to "YES."

Get
Ṁ1,000
and
S3.00
Sort by:

Will the newest definition of open source AI (https://opensource.org/ai/open-source-ai-definition) be used to judge whether a model is, in fact, open source?

@Primer > I think that has always been the prevailing definition, actually: by any sensible measure, weighths and training data are as much part of an LLM model as its computer code. It was "Open"AI and its ilk that had introduced their twisted explanations on why their semi-closed models should be considered open.

@Zozo001CoN But some models, including everything from Meta, won't be considered open source under this definition. How will "being open source" be judged for this market?

bought Ṁ500 YES

Should resolve on nemotron, assuming this means original GPT-4

Does this include 4o?

predicts NO

Which GPT-4?

© Manifold Markets, Inc.Terms + Mana-only TermsPrivacyRules