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

Study mode

openai.com

Post reply on HN
Print view View original

801–810 of 828 posts

Re: Study mode

#801

Post by chpatrick »

bscphil wrote: post #565
chpatrick wrote: post #427

Earlier quoted context omitted.

Humans who have heard of Monty Hall might also say you should always switch without noticing that the situation is different. That's not evidence that they can't think, just that they're fallible. People on here always assert LLMs don't "really" think or don't "really" know without defining what all that even means, and to me it's getting pretty old. It feels like an escape hatch so we don't feel like our human speci…

> Humans who have heard of Monty Hall might also say you should always switch without noticing that the situation is different. That's not evidence that they can't think, just that they're fallible. At some point we start playing a semantics game over the meaning of "thinking", right? Because if a human makes this mistake because they jumped to an already-known answer without noticing a changed detail, it's because (…

I'm not sure what your argument is. The common claim that annoys me about LLMs on here is that they're not "really" coming up with ideas but that they're cheating and just repeating something they read on the internet somewhere that was written by a human who can "really" think. To me this is obviously false if you've talked to a SOTA LLM or know a little about how they work.

Re: Study mode

#802

Post by Timwi »

johnnyanmac wrote: post #630
chpatrick wrote: post #427

Earlier quoted context omitted.

Humans who have heard of Monty Hall might also say you should always switch without noticing that the situation is different. That's not evidence that they can't think, just that they're fallible. People on here always assert LLMs don't "really" think or don't "really" know without defining what all that even means, and to me it's getting pretty old. It feels like an escape hatch so we don't feel like our human speci…

>People on here always assert LLMs don't "really" think or don't "really" know without defining what all that even means, Sure. To Think: able to process information in a given context and arrive at an answer or analysis. an LLM only simulates this with pattern matching. It didn't really consider the problem, it did the equivalent of googling a lot of terms and then spat something that sounded like an answer To Know:…

I have to push back on this. It's the people who constantly assert that LLMs “don't think” who are not engaging in a conversation. It's a thought-terminating cliché.

Unfortunately, even those willing to engage in this conversation still don't have much to converse about, because we simply don't know what thinking actually is, how the brain works, how LLMs work, and to what extent they are similar or different. That makes it all the more vexing to me when people say this, because the only thing I can say in response is “you don't know that (and neither does anyone else)”.

Re: Study mode

#803

Post by Timwi »

no_wizard wrote: post #440
chpatrick wrote: post #435

Earlier quoted context omitted.

Computers might be accurate but statistical models never were 100% accurate. That doesn't imply that no reasoning is happening. Humans get stuff wrong too but they certainly think and reason. "Pattern matching" to me is another one of those vague terms like "thinking" and "knowing" that people decide LLMs do or don't do based on vibes.

Pattern matching has a definition in this field, it does mean specific things. We know machine learning has excelled at this in greater and greater capacities over the last decade The other part of this is weighted filtering given a set of rules, which is a simple analogy to how AlphaGo did its thing. Dismissing all this as vague is effectively doing the same thing as you are saying others do. This technology has lim…

You're derailing the conversation. The discussion was about thinking, and now you're arguing about something entirely different and didn't even mention the word “think” a single time.

If you genuinely believe that anyone knows how LLMs work, how brains work, and/or how or why the latter does “thinking” while the former does not, you're just simply wrong. AI researchers fully acknowledge ignorance in this matter.

Re: Study mode

#804

Post by Timwi »

phailhaus wrote: post #742
Sherveen wrote: post #585

Earlier quoted context omitted.

LLMs are vulnerable to your input because they are still computers, but you're setting it up to fail with how you've given it the problems. Humans would fail in similar ways. The only thing you've proven with this reply is that you think you're clever, but really, you are not thinking, period.

And if a human failed on this question, that's because they weren't paying attention and made the same pattern matching mistake. But we're not paying the LLM to pattern match, we're paying them to answer correctly. Humans can think.

“paying the LLM”??

Re: Study mode

#805

Post by johnnyanmac »

Timwi wrote: post #802
johnnyanmac wrote: post #630

Earlier quoted context omitted.

>People on here always assert LLMs don't "really" think or don't "really" know without defining what all that even means, Sure. To Think: able to process information in a given context and arrive at an answer or analysis. an LLM only simulates this with pattern matching. It didn't really consider the problem, it did the equivalent of googling a lot of terms and then spat something that sounded like an answer To Know:…

I have to push back on this. It's the people who constantly assert that LLMs “don't think” who are not engaging in a conversation. It's a thought-terminating cliché. Unfortunately, even those willing to engage in this conversation still don't have much to converse about, because we simply don't know what thinking actually is, how the brain works, how LLMs work, and to what extent they are similar or different. That m…

>It's the people who constantly assert that LLMs “don't think” who are not engaging in a conversation.

I'm responding to the conversation. Oftentimes it's engaged on "AI is smarter than me/other people". It's in the name, but "intelligence" is a facade put on by the machine to begin with.

>because we simply don't know what thinking actually is

I described my definition. You can disagree or make your own interpretation, but to dismiss my conversation and simply say "no one knows" is a bit ironic for a person accusing me of not engaging in a conversation.

Philosophy spent centuries trying to answer that question. Mine is a simple, pragmatic approach. Just because there's no objective answer doesn't mean we can't converse about it.

Re: Study mode

#806

Post by johnnyanmac »

chpatrick wrote: post #800
johnnyanmac wrote: post #630

Earlier quoted context omitted.

>People on here always assert LLMs don't "really" think or don't "really" know without defining what all that even means, Sure. To Think: able to process information in a given context and arrive at an answer or analysis. an LLM only simulates this with pattern matching. It didn't really consider the problem, it did the equivalent of googling a lot of terms and then spat something that sounded like an answer To Know:…

You're just deferring to another vague term "pattern matching". If I think back to something I was taught in primary school and conclude that 1+1=2 is that pattern matching? Therefore I don't really "know" or "think"? People pretend like LLMs are like some 80s markov chain model or nearest neighbor search, which is just uninformed.

Do you want to shift the discussion to the definition of a "pattern" or are we going to continue to move the goalpost? I'm trying to respond to your inquiry and instead we're just stuck in minutia.

Yes, to make an apple pie from scratch, we need to first invent the universe. Is that productive conversation to fall into or can we just admit that your dismissing any opinion that goes against your purview?

>If I think back to something I was taught in primary school and conclude that 1+1=2 is that pattern matching?

Yes. That is an example of pattern matching. Let me know when you want to go back to talking about LLMs.

Re: Study mode

#808

Post by QuantumGood »

johnnyanmac wrote: post #631
maerch wrote: post #579

Earlier quoted context omitted.

I am not sure how good your test really is. Or at least how high your bar is. Paul Erdös was told about this problem with multiple explanations and just rejected the answer. He could not believe it until they ran a simulation.

I don't know who Paul Erdös is, so this isn't useful information without considering why they rejected the answer and what counterarguments were provided. It is an unintuitive problem space to consider when approaching it as a simple probability problem, and not one where revealing new context changes the odds.

Erdös published more papers than any other mathematician in history—and collaborated with more than 500 coauthors, giving rise to the concept of the "Erdős number," a (playful) measure of collaborative proximity among mathematicians

Re: Study mode

#809

Post by Zambyte »

Anamon wrote: post #770
Zambyte wrote: post #706

Earlier quoted context omitted.

It certainly should be able to tell you it doesn't know. Until it can though, a trick that I have learned is to try to frame the question in different ways that suggest contradictory answers. For example, I'd ask something like these, in a fresh context for each: - Why does Duckduckgo change it's logo based on what you've searched? - Why doesn't Duckduckgo change it's logo based on what you've searched? - When did Du…

I see these approaches a lot when I look over the shoulders of LLM users, and find it very funny :D you're spending the time, effort, bandwidth and energy for four carefully worded questions to try and get a sense of the likelihood of the LLM's output resembling facts, when just a single, basic query with simple terms in any traditional search engine would give you a much more reliable, more easily verifable/falsifia…

The nice thing about using a language model over using a traditional search engine is being able to provide specific context (ie disambiguate where keyword searches would be ambiguous) and to correlate unrelated information that would require multiple traditional searches using a single LLM query. I use Kagi, which provides interfaces for both traditional keyword searches, and for LLM chats. I use whichever is more appropriate for any given query.

Re: Study mode

#810

Post by Wilder7977 »

wing-_-nuts wrote: post #758
Wilder7977 wrote: post #656

Earlier quoted context omitted.

Lets not forget also the ecological impact and energy consumption.

Honestly, I think AI will eventually be a good thing for the environment. If ai companies are trying to expand renewables and nuclear to power their datacenters for training, well, that massive amount of renewables and battery storage becomes available when training is done and the main workload is inference. I know they are consistently training new stuff on small scale but from what I've read the big training batch…

I find this take overly optimistic. First, it's bases on the assumption that the training will stop, and that energy will be available for other, more useful, purposes. This is not guarantees. Besides this, it completely disregards the fact that today, tomorrow, energy will be utilized. We will keep emitting co2 for sure, and maybe, in the future, this will cause a surplus of energy? It's a bet I wouldn't take, even because LLMs need lots of energy to run as well as for training.

But in any case, I wouldn't want Microsoft, Google, Amazon and OpenAI to be the ones owning the energetic infrastructure in the future, and if we realize, collectively, that building renewable sources is what er need, we should simply tax them and use that wealth to build collective resources.

Post reply on HN