🏅 Top traders

So, this was 4o ?

@JamesBaker3 yeah it seems like it, I’m resolving to 80%

It did not turn out to be at all like what was described. Arrakis was a (failed) attempt to make a sparsely activating model (which would be cheaper to run). It was supposed to be at the level of GPT-4. I see no evidence it was SOTA (much less "really close to AGI") in any way (it was scrapped).

https://www.businessinsider.com/openai-model-arrakis-dystopian-desert-world-dune-2023-10

OpenAI tried to develop an AI model named after a dystopian hellscape in 'Dune,' but the project failed to land, report says

The ChatGPT developer tried to build a more efficient AI model codenamed Arrakis, a reference to "Dune's" hellish planet, The Information reported.

@RobertCousineau This adds credence to the Jim Apple is a homeless man overhearing conversations in the restaurant near the OA office theory.

https://twitter.com/jon_victor_/status/1714366376209760627

It may be a ruse to avoid triggering the Assist Clause and merge away, which is worded "before we do". And Google may proclaim AGI anytime soon.
https://manifold.markets/MetaculusBot/will-openai-have-triggered-its-assi-ffd373939394

Will OpenAI have triggered its 'Assist Clause' in order to support another lab, before 2031?

46% chance. https://www.metaculus.com/questions/16187/openai-triggers-assist-clause/

How does this resolve if the name is correct but details are off? Or versa visa? Or small details are off?

@ElliotDavies resolves yes if the details are broadly correct and it turns out OpenAI had the model in Sept 2023, even if the name or some small details are incorrect

@ahalekelly Can we get anything more precise than "broadly correct", maybe a few examples of which details being wrong would resolve it NO or which ones being correct would be enough to resolve it YES? I'm asking because I looked into the rumours today and I'm pretty sure that there is such a model but also that at least some of the details described there are wrong.