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

61–70 of 86 posts

Re: As Rocks May Think

#61

Post by piekvorst »

johnfn wrote: post #49
piekvorst wrote: post #48

> Chief among all changes is that machines can code and think quite well now. They can’t and never will.

Are you really claiming that there isn't a machine in existence that can code? And that that is never possible?

It can code in an autocomplete sense. In the serious sense, if we don’t distinguish between code and thought, it can’t.

Observe that modern coding agents rely heavily on heuristics. LLM excels at training datasets, at analyzing existing knowledge, but it can’t generate new knowledge on the same scale, its thinking (a process of identification and integration) is severely limited on the conscious level (context window), where being rational is most valuable.

Because it doesn’t have volition, it cannot choose to be logical and not irrational, it cannot commit to attaining the full non-contradictory awareness of reality. That’s why I said “never.”

Re: As Rocks May Think

#62

Post by munificent »

DennisP wrote: post #47
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.

If the only point is distributing money, then the pointless bullshit job is an unnecessary complication.

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

Re: As Rocks May Think

#63

Post by libraryofbabel »

piekvorst wrote: post #61
johnfn wrote: post #49

Earlier quoted context omitted.

Are you really claiming that there isn't a machine in existence that can code? And that that is never possible?

It can code in an autocomplete sense. In the serious sense, if we don’t distinguish between code and thought, it can’t. Observe that modern coding agents rely heavily on heuristics. LLM excels at training datasets, at analyzing existing knowledge, but it can’t generate new knowledge on the same scale, its thinking (a process of identification and integration) is severely limited on the conscious level (context window…

Some of that is true, sure, but nobody who claims LLMs can code and reason about problems is claiming that they operate like humans. Can you give concrete examples of actual specific coding tasks that LLMs can’t do and never will be able to do as a consequence of all that?

Re: As Rocks May Think

#64

Post by piekvorst »

libraryofbabel wrote: post #63
piekvorst wrote: post #61

Earlier quoted context omitted.

It can code in an autocomplete sense. In the serious sense, if we don’t distinguish between code and thought, it can’t. Observe that modern coding agents rely heavily on heuristics. LLM excels at training datasets, at analyzing existing knowledge, but it can’t generate new knowledge on the same scale, its thinking (a process of identification and integration) is severely limited on the conscious level (context window…

Some of that is true, sure, but nobody who claims LLMs can code and reason about problems is claiming that they operate like humans. Can you give concrete examples of actual specific coding tasks that LLMs can’t do and never will be able to do as a consequence of all that?

I think it can solve about any leetcode problem. I don’t think it can build an enterprise-grade system. It can be trained on an exiting one but these systems are not closed and no past knowledge seems to predict the future.

That’s not very specific but I don’t have another answer.

Re: As Rocks May Think

#65

Post by johnfn »

piekvorst wrote: post #61
johnfn wrote: post #49

Earlier quoted context omitted.

Are you really claiming that there isn't a machine in existence that can code? And that that is never possible?

It can code in an autocomplete sense. In the serious sense, if we don’t distinguish between code and thought, it can’t. Observe that modern coding agents rely heavily on heuristics. LLM excels at training datasets, at analyzing existing knowledge, but it can’t generate new knowledge on the same scale, its thinking (a process of identification and integration) is severely limited on the conscious level (context window…

> 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.

Re: As Rocks May Think

#66

Post by jrmg »

RealityVoid wrote: post #50
johnfn wrote: post #46

Earlier quoted context omitted.

The previous post in this blog is titled "Leaving 1X". So your wait is over!

Unrelated, but it seems his previous company, 1x, was initially named Halodi and was located in Norway. And eventually, it was moved with all employees in Silicon Valley. How the hell does that work? That sounds like a logistical nightmare.Do you upend all those people's lives? Do you fire those who refuse? How many norwegians even want to go to the US? Sounds crazy to me.

How do they all get work visas?

Re: As Rocks May Think

#67

Post by Paracompact »

johnfn wrote: post #46
mynameisjody wrote: post #29

I'm still waiting for one of these articles to be written by someone without something to be directly gained by the hype. Eric Jang, VP of AI at 1X.

The previous post in this blog is titled "Leaving 1X". So your wait is over!

Very, very likely he is remaining in the AI space.

Re: As Rocks May Think

#68

Post by piekvorst »

johnfn wrote: post #65
piekvorst wrote: post #61

Earlier quoted context omitted.

It can code in an autocomplete sense. In the serious sense, if we don’t distinguish between code and thought, it can’t. Observe that modern coding agents rely heavily on heuristics. LLM excels at training datasets, at analyzing existing knowledge, but it can’t generate new knowledge on the same scale, its thinking (a process of identification and integration) is severely limited on the conscious level (context window…

> 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.

Re: As Rocks May Think

#69

Post by skybrian »

munificent wrote: post #60
skybrian wrote: post #52

Earlier quoted context omitted.

I don't think you've read those quotes very closely? He's writing about all computer science problems. And "just about any short digital problem" is not the same as solving the world's problems. AI ghosts can do a lot of things, but they're limited by being non-physical.

He does also say: > The entire global economy is re-organizing around the scale-up of AI models. > Software engineering is just the beginning; ... > Air conditioning currently consumes 10% of global electricity production, while datacenter compute less than 1%. We will have rocks thinking all the time to further the interests of their owners. Every corporation with GPUs to spare will have ambient thinkers constantly…

Yeah, I think that's magical thinking about how much better war planning will help.

Re: As Rocks May Think

#70

Post by keeda »

munificent wrote: post #4

> We are entering a golden age in which all computer science problems seem to be tractable, insomuch as we can get very useful approximations of any computable function. Alternatively, we are entering a dark age where the billionaires who control most of the world's capital will no longer need to suffer the indignity of paying wages to humans in order to generate more revenue from information products and all of the…

Here are my thoughts, which are not fully formed because AI is still so new. But taking this line of thought reductio ad absurdum, it becomes apparent that the elites have a critical dependency on us plebs:

Almost all of their wealth is ultimately derived from people.

The rich get richer by taking a massive cut of the economy, and the economy is basically people providing and paying for services and goods. If all the employees are replaced and can earn no money, there is no economy. Now the elite have two major problems:

a) What do they take a cut of to keep getting richer?

b) How long will they be safe when the resentment eventually boils over? (There's a reason the doomsday bunker industry is booming.)

My hunch is, after a period of turmoil, we'll end up in the usual equilibrium where the rest of the world is kept economically growing just enough to keep (a) us stable enough not to revolt and (b) them getting richer. I don't know what that looks like, could be UBI or something. But we'll figure it out because our incentives are aligned: we all want to stay alive and get richer (for varying definitions of "richer" of course.)

However, I suspect a lot will change quickly, because a ton of things that made up the old world order is going to be upended. Like, you'd need millions in funding to hire a team to launch any major software product; this ultimately kept the power in the hands of those with capital. Now a single person with an AI agent and a cloud platform can do it themselves for pocket change. This pattern will repeat across industries.

The power of capital is being disintermediated, and it's not clear what the repercussions will be.

Post reply on HN