/Soli/how-many-stars-will-karpathys-new-l

@traders I still haven't decided how to resolve this market, but here are my thoughts. I admit that there is strong discrepancy between the title of the question and the resolution criteria/comments.

What are the undisputed facts?

1. Quitting Current Role: Karpathy quit his position at OpenAI 🟢

2. Leading the Open-Source AI Community 🟡

   1. Active Participation: Karpathy started two new open-source projects: nanoGPT (~30k ⭐️) and minbpe (~7k ⭐️) 🟢

   2. Continuous Effort: Karpathy's contributions were not a one-time effort, as evidenced by his GitHub commits and the fact that he started two new open-source repos. He also posted a video explaining tokenization in detail on his YouTube channel, prompting people to think of better methods. 🟢

What is the case for Yes?

• Resolution criteria 1 and 2.1 + 2.2 are fulfilled.

• Karpathy's social media accounts are some of the most popular in the machine-learning community; therefore, one should not underestimate the impact a video, repo, or even a single tweet could have on the whole community.

• Teaching is a form of leadership that involves guiding, influencing, and inspiring others to achieve knowledge, understanding, and/or skill. Karpathy started two new education repos and posted a two-hour video on YouTube.

What is the case for No?

• https://manifold.markets/Soli/will-karpathy-quit-and-lead-the-ope#xdu0u16ge3

• https://manifold.markets/Soli/will-karpathy-quit-and-lead-the-ope#wc2onufcard

• https://manifold.markets/Soli/will-karpathy-quit-and-lead-the-ope#jscd6ridmim

@Soli

2.1. Active Participation.

Are other repositories taking tricks out of nanoGPT/minbpe, or are those two simply projects albeit successful?

Karpathy started two new education repos and posted a two-hour video on YouTube

That's very small amount for teaching as leadership, isn't it? For instance, PyTorch, an open-source project techincally, has significantly more materials.

@Soli I sold my shares, so not really invested emotionally or otherwise. This is an outsider opinion to try and change your mind, just because I think it would be kind of laughable to conclude Karpathy is "leading the Open Source AI Community".

The people behind Llamma and Mistral have a much better claim to be "leading the community". Karpathy is just more famous, and the AI company he co-founded is notoriously not open, despite its name (nor did it become more open during his last stint there).

Use history and compare him to the impact of people like Linus Torvalds, Richard Stallman, Greg Kroah-Hartman, Eric S. Raymond, Larry Wall, Tim O'Reilly, John Carmack, Bryan Cantrill, Brendan Gregg, Kevin Mitnick, Mark Shuttleworth, Jamie Zawinsky, Ian Murdock, Bram Cohen, Sean Parker, Phil Zimmermann, Diffie, Helman, and Merkle, Aaron Swartz, Jacob Appelbaum, Rob Pike, Vitalik Buterin, among others.

All of those people listed above had more of a claim to have been among the leadership of the communities they were part of at one point or another, and Karpathy is simply not nearly as notorious or influential enough in AI compared to them at their respective fields and subfields.

You seem to be unsure yourself, so I would recommend committing to a poll or resolving as N/A and then maybe redoing it with better criteria if you want to.

@Soli If the quoted tweet https://twitter.com/chheplo/status/1725681364807340204 is considered to be a part of the market's description, then the meaning of the "unite the unorganized, scattered Open Source AI ecosystem" should be at least partially a description of "Personal Assessment of Leadership". Whether you consider his current role to be enough, is up for "Subjective Resolution".

I don't have much at stake so I'm not too unhappy about YES, but I'd say it isn't the leadership that was predicted

I don't think him co-founding OpenAI is not a point against him. he worked at OpenAI when it was actually open, and left around the time it stopped being so. He joined again recently, but left. It wouldn't be that surprising that it had to do with tensions around open-sourcing models. It's plausible that more people joined the opensource AI community because of how clear OpenAI's hypocrisy was, than would have if it had been called closedAI... LOL

I agree "leading the OS AI community" is somewhat of an ambiguous question with no "true answer", and if we went back in time maybe a better title would be "will Soli think that Karpathy is one of the leaders the OS AI community", or something, to clarify that it's a subjective market with a subjective opinion (which was clear in the comments and description).

I think Soli's cases for Yes above are strong, and I'll add that a leader must not act alone. i wouldn't find it reasonable for this to resolve No just because some other developer is argued to be a slightly more of a leader, somehow. I don't know who that is if there is one, but the idea that there can only be one leader of the OS AI community is something I disagree with.

Finally, I'd like to point to something I've mentioned before, being that today, with the size of the development in AI, and more and more development that is open source, it just isn't true that the few people behind the biggest current project are the main "leaders". I think the community does grow through their projects, but it arguably grows even more through knowledge dissemination, which is what karpathy's projects, and his youtube series, are doing best. Case in point: He has more stars on his own projects than Mistral AI does, and his most popular youtube video has 4 million views with 94K likes. Many people have gotten into open source AI because of him.

@Bayesian fair enough.

Although, if we had to choose one person, I would say Yann LeCun is much more of a leader of the community for open development of AI, both in terms of advocacy and actual impact and influence, however awesome Andrej's educational content may be.

@traders i slept over it and i am strongly leaning towards N/A at this point. I think judging by the title only the market should clearly resolve “No”. Judging by the resolution criteria and especially my comments the market should resolve “Probably Yes”. Apologies @Bayesian if you wasted time making very good points on an ill-defined market.

(personal opinion)

I disagree, I think it's NO, although N/A is obviously better than YES.

2.1 is fulfilled, but it's obviously a very low bar, so that shouldn't matter.

2.2 is not fulfilled. It says: "Continuous Effort: His participation should be consistent and not limited to a one-time contribution."

This has "continuous", "consistent", and "not limited to one-time". It's certainly not limited to one time, but 'continuous' and 'consistent' are both demonstrated false by his (iirc) lack of github contributions in March, the first calendar month after we learned he left OpenAI.

2.3 isn't fulfilled.

Karpathy's social media accounts are some of the most popular in the machine-learning community; therefore, one should not underestimate the impact a video, repo, or even a single tweet could have on the whole community.

There are a lot of people who are popular on twitter, in the machine learning community! Tweeting really is not the same thing as leadership, idk.

Teaching is a form of leadership that involves guiding, influencing, and inspiring others to achieve knowledge, understanding, and/or skill. Karpathy started two new education repos and posted a two-hour video on YouTube.

This feels like a huge stretch. A lot of other things could also be called leadership. Why isn't Sam Altman is leading the open-source AI community, OpenAI's product decisions set the direction for OS AI more than Karpathy ever could!

i wouldn't find it reasonable for this to resolve No just because some other developer is argued to be a slightly more of a leader

they are more leaders to a large extent, not just slightly

I think the community does grow through their projects, but it arguably grows even more through knowledge dissemination, which is what karpathy's projects, and his youtube series, are doing best

Marketing is not leadership! Making educational videos is not leadership! They really are just different!

and especially my comments

Your comments weren't clarifying rules, they were expressing opinions. A market creator shouldn't be bound by their past opinions if they realize they were incorrect!

I strongly support NO here, because, like, it's correct.

His contribution graph is just not that of someone who's "leading" the open-source AI community. https://github.com/karpathy I think there are dozens of people who are much closer to a "leader" than him

his overall impact on shaping the direction and focus of others in the community

he is not doing that

Like if you think Karpathy is leading the open-source AI community

Like if you think Karpathy is not leading the open-source AI community

@Soli any thoughts here? Look at his 'contributions', https://github.com/karpathy, he is not leading the community.

@jacksonpolack Leading a community has nothing to do with a contribution graph, it's more about influence and showing the path forward. And this specific market just requires "active participation", which he does.

@Weezing
> Continuous Effort: His participation should be consistent and not limited to a one-time contribution.

@jacksonpolack
and for 'Personal Assessment of Leadership' he just isn't leading the community by any common sense use of the word 'lead', and there are many other people who are better candidates than him, some named below.

"Active participation" should be a bare minimum, not a significant qualifier for leadership! Thousands of people actively participate, and during most months karpathy did not actively participate. I don't get the case for yes

@jacksonpolack Yes and as you can see he hasn't done just a single contribution. After he left OpenAI he is contributing every week. I am not saying how this market should resolve or if he is leading or not, just that his contribution graph is not a very good argument for NO in general sense and also based on this market specific resolution criteria.

the power of scale from sharing knowledge shouldn't make it surprising that someone with his influence is not spending all day pull requesting to 25 different projects. on the other hand,

Yes, he has a popular youtube channel, with a video 1 month ago and a video 4 months ago. The youtube videos aren't setting the priorities of the AI community, they're for beginners. He spent the entire month of March contributing to private repositories on github, as far as I can tell, no public ones. Wasn't that his first month not at openai? If we're going off of manifold perception, the 7 likes "way off" comment below seems the strongest signal there. The market mentions the github as a way to judge his leadership, not the youtube!

@jacksonpolack He made an entire video about tokenization, with the explicit hope that someone can figure out a better way. I wouldn't dismiss it as "it's for beginners".

The "way off" comment is probably popular because it ends with "getting out while I can". That's why I liked it, and why I wish I'd gotten out of this mess of a market.

"Manifold perception" is the market, not some number of likes on some comment.

@jacksonpolack you make valid points - i added a comment above where i explain why i think N/A is best

@Soli How close would you say his current opensource contributions are to counting? he's participating to a few continuously, but they all are created by him. I'm guessing that still counts but just making sure

@Bayesian i think based on the resolution criteria it should count but open for counter arguments

@Soli What’s the case for the subjective ‘leadership’ piece?

@Hedgehog

Karpathy is actively contributing to the following repos right now:

1. nanoGPT (~30k ⭐️)

2. minbpe (~7k ⭐️)

3. llama2.c (~14k ⭐️)

Most of these are very educational repos, and teaching is a form of leadership so 🤷🏻‍♂️

@Soli This seems way off to me: Leading the community really strongly implies something like being widely accepted as the leader, or at least one of a few leaders, not just making a few independent contributions. Getting out while I can.

@Hedgehog according to your definition, who is leading the AI open-source community right now?

@Soli I'm not sure—maybe nobody. I don't think that means the question is vacuous, though—Connor Leahy would clearly count if this were asked a couple of years ago, for example.

@Hedgehog On a similar note, maybe Stella Biderman?

@Hedgehog Or maybe some/all of Sasha Rush, Percy Liang, or Colin Raffel?

@Hedgehog super interesting, thank you for sharing. I see where you are coming from but I feel for Karpathy to be considered for your list he would have to put consistent effort for many years which would make it impossible for this question to resolve yes.

@Soli He’s visible enough that I think a few blog posts saying things like ‘let’s all do X, talk to me about it’ could go far pretty quickly.

@Soli this market is now a question of how you will define leadership ;)

@anonymous I sold my position to remove any biases since the market is highly subjective at this point

I don't think a blog post is necessarily needed, I think just sparking interest and encouraging others on twitter should count, especially given his popularity and the amount of respect people have for him and his teaching

also I think considering your previous comments soli, i would argue that "leading the open source community" doesn't require anything like "being the only person that leads it" or "being the principal leader", so I would disagree with hedgehog's comment that it would require "being seenn as the leader". He quit, leads the open-source ai community by example, but isn't the only one. I can't really judge as to whether he's the main one or not (I certainly follow him more than others but im probably not a super typical example), but imo it shouldn't matter because that didn't seem to be what you were trying to find out with the market, as judged by your comments. let me know if I judged it wrong

but I feel for Karpathy to be considered for your list he would have to put consistent effort for many years which would make it impossible for this question to resolve yes.

just based on open source contributions over the past month there are hundreds of people who are much better candidates