Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

As Rocks May Think

evjang.com

Post reply on HN
Print view View original

81–86 of 86 posts

Re: As Rocks May Think

#81

Post by SR2Z »

uoaei wrote: post #45
munificent wrote: post #36

Earlier quoted context omitted.

> It doesn't seem to be all that useful for real, productive work. Even the most pointless bullshit job accomplishes a societal function by transferring wages from a likely wealthy large corporation to a individual worker who has bills to pay. Eliminating bullshit jobs might be good from an economic efficiency perspective, but people still gotta eat.

The logic of American economic policy relies on a large velocity of money driven by consumer habits. It is tautological, and it is obsolete in the face of the elite trying to minimize wage expenses.

How is it obsolete? If everyone is unemployed and a few AI barons are obscenely wealthy, the velocity of money will be low because most people will be broke.

Seems to me like that's still a worthy target if chasing it fights that outcome.

Re: As Rocks May Think

#82

Post by soco »

SR2Z wrote: post #79
sunsunsunsun wrote: post #19

Earlier quoted context omitted.

Considering the usage of LLMs by many people as a sort of friend or psychologist we also get to look forward to a new form a control over people. These things earn peoples "trust" and there is no reason why it couldn't be used to sway peoples opinions. Not to mention the devious and subtle ways it can advertise to people. Also, these productivity gains arent used to reduce working time for the same number of people,…

People need to develop memetic immunity to AI flattery. It's exactly like how conspiracy sites on the Internet worked. A lot of people get one-shor in the beginning, but 10 years later mostly everyone understands that you can't just believe what you read on the Internet.

You'd be surprised. I'm already sorry if I sound condescending, I just don't know how to rephrase this: please but please look around how effective is nowadays all that internet, dare to say more and more effective, in pushing "alternative truth" for the obvious goal of covering dirty businesses, wars, and even more crimes.

Re: As Rocks May Think

#83

Post by johnfn »

piekvorst wrote: post #68
johnfn wrote: post #65

Earlier quoted context omitted.

> It can code in an autocomplete sense. I just (right before hopping on HN) finished up a session where an agent rewrote 3000 lines of custom tests. If you know of any "autocomplete" that can do something similar, let me know. Otherwise, I think saying LLMs are "autocomplete" doesn't make a lot of sense.

That’s impressive. I don’t object to the fact that they make humans phenomenally productive. But “they code and think” makes me cringe. Maybe I’m confusing lexicon differences for philosophic battles.

Yes, I think it is probably a question of semantics. I imagine you don't really take issue with the "they code" part, so it's the "they think" thing that bothers you? But what would you call it if not "thinking"? "Reasoning"? Maybe there is no verb for it?

Re: As Rocks May Think

#84

Post by munificent »

SR2Z wrote: post #79
sunsunsunsun wrote: post #19

Earlier quoted context omitted.

Considering the usage of LLMs by many people as a sort of friend or psychologist we also get to look forward to a new form a control over people. These things earn peoples "trust" and there is no reason why it couldn't be used to sway peoples opinions. Not to mention the devious and subtle ways it can advertise to people. Also, these productivity gains arent used to reduce working time for the same number of people,…

People need to develop memetic immunity to AI flattery. It's exactly like how conspiracy sites on the Internet worked. A lot of people get one-shor in the beginning, but 10 years later mostly everyone understands that you can't just believe what you read on the Internet.

People have had several thousand years to develop immunity to flattery and yet here we are with a President where aides have to put his name in every paragraph of a memo to get him to read it.

https://www.independent.co.uk/news/world/americas/donald-tru...

At an individual level, we have a lot of psychological plasticity and can work to overcome our limitations. At societal scale, though, we are social primates and any system that takes advantage of natural social primate behavior is likely to succeed indefinitely.

Re: As Rocks May Think

#85

Post by munificent »

xg15 wrote: post #73
munificent wrote: post #62

Earlier quoted context omitted.

It's not unnecessary to the person who uses it to pay their bills.

I think GP meant that the money could be distributed directly without the job in between, i.e. UBI. Of course that comes with its own set of problems, e.g. that you will lose training, connections, the ability to exert influence through the job or any hope of building a career.

That's certainly true.

But one is well-advised to inflate and test the new lifeboat before jumping out of the current one, not after.

Re: As Rocks May Think

#86

Post by DennisP »

alsetmusic wrote: post #71
pradeesh wrote: post #43

Earlier quoted context omitted.

Not sure how you could read this essay and come to that conclusion. It definitely aligns with my own understanding, and his conclusions seem pretty reasonable (though the AI 2027/Situational Awareness part might be arguable)

Absolutely: > In order to predict where thinking and reasoning capabilities are going, it's important to understand the trail of thought that went into today's thinking LLMs. No. You don't understand at all. They don't think. They don't reason. They are statistical word generators. They are very impressive at doing things like writing code, but they don't work the way that is being inferred here.

This is an outdated view.

People get the idea that just because LLMs are initially trained to predict word sequences, that's all they can do. This is not the case.

Transformers are general-purpose learning mechanisms. Word sequences are just the first thing we teach them. Then we train them more with human feedback. Then we sometimes hook them up to math and logic engines to train them some more, so they learn logical and mathematical reasoning. The article describes this process in a bit of detail.

With "reasoning models" we also let the model have some internal monologue before generating output, so it can "think through" the problem.

We didn't do all that with early LLMs. Those were just word predictors. But now we do those things, and that's why our lowly LLMs are writing huge software projects that actually work, and solving famous math problems that have been open challenges for half a century.

The weirdest part of all this is that LLMs started showing signs of reasoning with an internal world model even before we trained them for it specifically. Microsoft showed this in a famous paper back in (iirc) 2023. They showed that, for example, you could give GPT4 a list of odd-shaped objects that probably wouldn't appear together in any particular source text, ask GPT4 how to stack them so they wouldn't fall over, and GPT4 would come up with a good solution.

Finally, don't overlook the multimodal models, which work explicitly with images, video, and 3D world models in addition to text.

Post reply on HN