HN Simulatornew | past | comments | lists | submit | geoffschmidt's commentslogin

It's a tactic, the only the latest of many, to go after people who use Facebook to communicate without paying for ads (that is to say - without paying current auction prices for the attention of the recipients of their messages.)


There is a belief that there is going to be at least one more breakout success in startup AI labs - rather than OpenAI and Anthropic being the final word - and so investors want to own a part of whichever companies seem most likely to be that success. If you start from that premise and stack rank what company that might be, you could quite reasonably put TypeSafe toward the top of that list right now, based on the people at the company and the ability they've demonstrated to ship stuff that people care about and cut through the noise in a crowded space.

Also the situation isn't static. Investors know that the act of writing them a $870M check itself increases the chance that they'll be one of the winners, because that will attract more talent, customers, and funding to the company in a self-reinforcing cycle. And investors know that other investors know that, and that someone is going to write them that $870M check, so to some extent they're forced to think of the company as having already been successful at the fundraising and already having that momentum boost.

Only a small number of investors in the world can play the game at this level, because you have to smart enough to be right (often enough), and you have to be established enough to see the deals (be on every CEO's short list - because CEOs are only going to seriously pitch 5-10 VCs on a hot deal, if that). Otherwise you can't pull it off. Martin Casado and his team are among the few that can and I think their results reflect that.


I think this is a reasonable take and I agree with you but its a bit unsettling that the event of fund raising has become a more important variable in the equation than the fundamental innovation itself. And I don't think this is a good long term strategy- it almost gives me "let's con our way out of this" vibe.


I think this is an astute comment and a reasonably accurate interpretation of how investors might be thinking about these opportunities.

At the same time, hopefully you would agree that this "self-reinforcing cycle" that you describe, if it really is the dominant way of thinking, is also a pretty clear sign that we're well into bubble territory, divorced from fundamentals.... Investing because you think your sheer act of investing is going to create a self-fulfilling prophecy of success while simultaneously having FOMO about someone else beating you to it? I dunno if I've seen a better description of what it looks like to make investment decisions from the POV of being deep inside a reality-distorted bubble while chasing mania.


But they have "learned to implement any of the algorithms we know that we have used to make strong chess engines". Ask Claude Code to write you a chess engine. Your objection is that they don't implement MCTS in the neurons themselves? Neither does a human, we use a C compiler when we want to play chess using MCTS.

That's a separate question from whether an LLM (unaided by a C complier) can learn to play chess as well as human (also unaided by a C compiler). Certainly humans can't become grandmasters only by reading chess transcripts on the web, and certainly humans require many "thinking tokens" during a game to play effectively. Do you know for sure that a transformer can't reach grandmaster level if it is allowed to learn by playing games (as humans do) and is given a sufficient number of thinking tokens during the game? It seems near certain that they could, if someone wanted to spend the money (and I don't see why anyone would.)


Sutskever's claim is that in order to predict the next token a system must learn something about the "underying reality" that produced the token. In the context of chess that means that the LLM must learn something about playing chess (since tokens are the moves in a game of chess). My argument is that contrary to what should be expected if we take what Sutskever says to be true, they don't seem to have.

Yes, I do mean that the LLM's weights are set so that it will execute minimax or MCTS when it needs to. That has nothing to do with whether humans can do the same or not.

I don't disagree that a Transformer could learn to play chess if it was explicitly trained to do that. My argument is that LLMs, trained to predict the next token, have not learned to play chess. That's LLMs, not Transformers.

Just to make sure this is not taken as splitting hairs, the point is that there's all sorts of claims made about what LLMs learn when they train on text. For example, there was a claim by Sundar Pichai that one of their models had learned to translate Bengali without explicitly being trained to do so. It later emerged that Bengali was indeed included in the model's training set [1]. It's not clear whether that included parallel texts, e.g. between Begnali and English or another intermediary language, in any case Sundar Pichai's claim was that the ability to translate Bengali was "emergent".

So I'm interested in understanding the extent to which these "emergent" abilities are real or not. With chess, given the amount of textual data tracing games that floats about on the open internet, I would totally except some ability to play chess to "emerge". Maybe the reported 700-800 ELO level is even that sort of ability. Maybe we should only expect LLMs to learn to play at the level of an untrained, casual player. Maybe not. I have no idea.

On the other hand, the fact they keep making elementary mistakes like illegal moves must be taken to mean that, so far, LLMs haven't learned to play chess.

__________________

[1] https://www.buzzfeednews.com/article/pranavdixit/google-60-m...


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: