🏅 Top traders

Here are the polls about the most "significant" announcements. I also included a question about Sora vs. o1 for my own curiosity, since those were about neck-and-neck throughout the course of the market.

/CDBiddulph/which-was-more-significant-the-o1-a
/CDBiddulph/which-was-more-significant-the-sora-pLdhQuRyLZ
/CDBiddulph/which-was-more-significant-the-laun
Bonus:
/CDBiddulph/which-was-more-significant-the-sora

I forgot to mention the non-o3 announcements of the 12 days of OpenAI in my poll - oops! Hopefully no one thought 1-800-CHATGPT was more significant than ChatGPT itself 😄

The polls look statistically-significant enough to me - resolving now

EDIT: oops, I can't resolve them myself. @dominic please help?

@CDBiddulph o1 vs o3 poll is close

FWIW, o1 was ABSOLUTELY more significant than o3. It represented a paradigm shift in the industry, vs just iterating to acheive progressively more challenging benchmark scores

before o3 i thought o1 was basically the most you could train models in that new paradigm without running out of available compute. If that had been true it would have not been a very significant paradigm shift. The fact that o1 was orders of magnitude less scale than what was available is what made me update a lot. I learned that when o3 came out. o3 feels much more significant than o1 in that sense.

also where is the poll?

@Bayesian The polls are all linked in my top-level comment

@CDBiddulph can’t see them. Can you pin your comment or add the polls to the post?

@JaimeSantaCruz i also couldnt but turns out the bolded characters can hide a poll link! You night have some luck if you just click in bolder character in the first message in this comment thread

Marcus seems impressed: https://garymarcus.substack.com/p/o3-agi-the-art-of-the-demo-and-what

@Usaar33 I ran this article through ChatGPT and it said it was a negative sentiment

@dominic this is going to be prompt and interpretation specific. Entire article I can see being negative sentiment (and Chatgpt agrees) but the are positive reactions within it. Chatgpt acknowledges there is "something he reacted positively to" and recommends resolving yes

@Usaar33 Yeah, I might just resolve this one N/A

@Usaar33 I think we disagree on the domain for this question. I didn't bet on this, but I interpreted "something" as an actual product, like o3.

You could say that Gary Marcus reacted positively to "o3's performance on the ARC benchmark", but I wouldn't have counted that as one of the possible "somethings".

@TimothyJohnson5c16 This was also my intuition fwiw

“o1 and/or o1-mini fine tuning” is currently trading at 4%, but was revealed waaaay back on December 6: https://x.com/openai/status/1865136373491208674

Gary Marcus reacted negatively to the o3 demo according to ChatGPT (I pasted in his article). If anyone can find an example of a more positive reaction that he had I'll look at that too, but I haven't found anything yet

@dominic

@mathvc Even if that is a sentence he actually said, I am sure it's not very hard to find a positive sentence in his article. But I don't think it's fair to take it out of context and not analyze his full reaction to each thing.

@dominic it is not a sentence from his article. It is a separate tweet he made.

@mathvc Yeah, this particular tweet might get analyzed as positive. But it's a stretch to take one mildly positive tweet and reasonably say "this is his reaction to o3," unless he hadn't written anything else about it.

@dominic well i would like to disagree. He has a positive reaction to an announcement of model scores on FrontierMath

This is separate from his negative reaction to o3

I'll resolve the things that didn't happen NO next week, in case of any surprise announcements.

@JaundicedBaboon This should resolve YES: https://arcprize.org/blog/oai-o3-pub-breakthrough

Should o3 count as 1,000,000+ context length? According to the people who run ARC o3 was able to use almost 10 billion token chains of thought to solve just one problem.

https://www.reddit.com/r/singularity/comments/1hisp7o/o3_high_compute_costs_is_insane_3000_for_a_single/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

AGI achieved only at 1.6%? 👀