| ▲ | ThrowawayR2 7 hours ago |
| Nice. This demolishes the "LLMs can reason" (but not enough to avoid this sort of basic error) and "humans make mistakes too" (not like this) talking points from the LLM promoters. |
|
| ▲ | __MatrixMan__ 5 hours ago | parent | next [-] |
| There's a lot of space between: - LLM's can reason - everything an LLM does is the result of reasoning This demolishes only the latter point, which as far as I know has no supporters. |
| |
| ▲ | zahlman 5 hours ago | parent [-] | | A human wouldn't mindlessly reproduce the signature due to "not reasoning" and being intellectually lazy while doing the drawing. For the human, including the signature is more effort than omitting it; for the generative system, it appears the opposite is true. The point is to highlight that difference. | | |
| ▲ | __MatrixMan__ 5 hours ago | parent [-] | | Is that a useful point though? It's like accusing somebody of being a lousy chef because they have such terrible taste in takeout. There's just no connection between these things. | | |
|
|
|
| ▲ | runarberg 4 hours ago | parent | prev | next [-] |
| I don‘t think any of these pro-AI talking points are in good faith. I think these talking points are there to hype up the technology or otherwise excuse the mis-allignment (i.e. consumer hostility). |
|
| ▲ | wilg 5 hours ago | parent | prev | next [-] |
| Image generators aren't LLMs. |
| |
| ▲ | GeorgeWBasic 4 hours ago | parent [-] | | They are a related technology, though, using transformers, some with quite similar architectures to LLMs. |
|
|
| ▲ | dyauspitr 7 hours ago | parent | prev | next [-] |
| LLMs can very much reason and they do it very well. |
| |
| ▲ | bunderbunder 6 hours ago | parent [-] | | I remain unconvinced. The thing about a generative language model that’s trained from a massive but unknown corpus is, it’s practically (if not theoretically) impossible to evaluate the extent to which data leakage contributes to any particular output. But I would argue that, as things currently stand, “sophisticated engine for approximately querying a pastiche of the results of human reasoning that comprise its training corpus” remains a more parsimonious explanation than “it’s doing actual reasoning” for how this neural network architecture produces the phenomena we’ve been observing. | | |
| ▲ | Muromec 6 hours ago | parent [-] | | >sophisticated engine for approximately querying a pastiche of the results of human reasoning that comprise its training corpus Well if the thing can find and fix bugs in something that is using non-mainstream stuff that is surely not in it's training dataset, that's better than a rubber duck already. Whether it has soul is a different question of course. | | |
| ▲ | bunderbunder 6 hours ago | parent | next [-] | | “A soul”? As popular as I know the rhetorical tactic is on both sides of these discussions about LLMs, I’d still thank you not to strawman me. | |
| ▲ | phoghed 6 hours ago | parent | prev [-] | | Don’t even bother. These people almost always have some goofy ass, non standard, fluid definition of “thinking” or “reasoning” that cannot ever be met. | | |
| ▲ | bunderbunder 5 hours ago | parent [-] | | My opinion of their reasoning capability is based in part on (proprietary, non-published, only internally peer reviewed) experiments on GPT-series models’ ability to perform a suite of formal and informal inference and deduction tasks. Perhaps you could argue that “appropriately applies syllogism to arrive at correct conclusions” is too high a bar to set, but I don’t think it would be fair to call it a “goofy-ass”, “non-standard” or “fluid” element of a reasoning capacity assessment. | | |
| ▲ | chpatrick 5 hours ago | parent | next [-] | | Yeah I guess the Jacobian conjecture must have been disproved without reasoning. | | |
| ▲ | bunderbunder 5 hours ago | parent [-] | | It’s hard to say. But supposedly the counter example wasn’t found by an agent running in full auto; it came out of a bunch of back and forth with a human operator. Without, in addition to the aforementioned access to currently non-public information about these models, a detailed transcript of the chat sessions leading up to the discovery, it’s hard to ascribe the reasoning steps involved to any source in particular. Part of my concern here is that simply pointing out that LLMs appear to be performing tasks that can be done through reasoning, and using that in and of itself as evidence of reasoning, is affirming the consequent. |
| |
| ▲ | retsibsi 2 hours ago | parent | prev [-] | | But you're not just saying they are insufficiently good at reasoning, you're saying they're (probably) not "doing actual reasoning". So we need to know how you are defining "actual reasoning". I don't think the bar for an actual reasoner can possibly be 'always appropriately applies syllogism to arrive at correct conclusions', because in that case nobody in the world is an actual reasoner. And if your bar were 'sometimes appropriately applies syllogism to arrive at correct conclusions', it's hard to understand why the current generation of AIs doesn't meet it; they are clearly capable of doing so, at least to all outward appearances. (Maybe you think their apparently successful demonstrations of reasoning are illusions, but again, you would need to define what counts as "actual reasoning" vs. a superficially convincing simulation of it.) |
|
|
|
|
|
|
| ▲ | antonvs 7 hours ago | parent | prev [-] |
| > This demolishes the "LLMs can reason" This is completely silly. If you don’t think LLMs can reason, you’ve either never used them to do tasks that require reasoning, or you don’t understand enough to recognize what’s involved in the responses you get. In this case it’s clearly the latter, because you’re confusing image generation models with LLMs. There are very big differences between the two. No-one is claiming that image generation models are capable of reasoning. |
| |
| ▲ | rf33 7 hours ago | parent [-] | | [flagged] | | |
| ▲ | chpatrick 6 hours ago | parent | next [-] | | What's the argument here? An airplane, a bee and a bird all fly despite doing it totally differently. LLMs also reason despite being made out of matmuls instead of meat. | | |
| ▲ | JoshTriplett 6 hours ago | parent | next [-] | | > What's the argument here? The most common argument for this is some core unexamined axiom that only humans can reason by definition, and then working backwards to a justification for that. | | |
| ▲ | antonvs 3 hours ago | parent [-] | | Not only unexamined, but ineffable. I've yet to see anyone give a good definition of what they think "reasoning" is that applies to humans solving complex logical problems but not to machines. (I suppose a religious or otherwise superstitious person might introduce the soul into this, but I haven't come across anyone actually willing to defend that hill.) |
| |
| ▲ | zahlman 5 hours ago | parent | prev [-] | | This is just a trick of language. There's no rule that says a priori whether an English word created before the invention of machines mimicking the behaviour, should describe the mimicry. There's no contradiction between "an airplane can 'fly'" and "a computer cannot 'reason'" because there is no reason why the two claims should relate whatsoever. | | |
| ▲ | chpatrick 5 hours ago | parent | next [-] | | Well, but we do we have these words, and they are useful. An airplane and a bird both travel through the air. A human and an LLM both make logical deductions. | |
| ▲ | antonvs 3 hours ago | parent | prev [-] | | You can just look at the definitions and see whether they apply. The relevant MW definition for "reasoning" is: "the use of reason, especially: the drawing of inferences or conclusions through the use of reason." And "reason" is: "the power of comprehending, inferring, or thinking especially in orderly rational ways." Functionally speaking, i.e. in terms of observable behavior, LLMs exhibit comprehension, inferring, and reasoning. If someone wants to object to that, they'd need to explain what relevant property prevents a conclusion drawn by an LLM from being counted as involving reasoning. |
|
| |
| ▲ | harimau777 6 hours ago | parent | prev | next [-] | | Why do you believe that they aren't reasoning? What would convince you that they are? | |
| ▲ | hardbass 6 hours ago | parent | prev | next [-] | | What do you think they are doing, and do you think machines can reason (in general, not necessarily current systems)? If they can't, how do you explain humans being able to reason given that we are physical machines too? | | |
| ▲ | Muromec 6 hours ago | parent [-] | | We have a soul given by God himself. It's written in a book. Next question? | | |
| |
| ▲ | unrented7977 6 hours ago | parent | prev | next [-] | | Airplanes don't flap but they fly, just how LLMs don't think but do reason. | |
| ▲ | AspireOne 6 hours ago | parent | prev [-] | | You're out of line. I've spent a long long time thinking about the problem of reasoning and consciousness, and it's not nearly as simple as your confidence and feeling of intellectual superiority would indicate you think it is. First, you'd obviously have to define what you even mean by reasoning, precisely. Let's hear it. Then demonstrate that LLMs do not corresponds to that description. You are making absolute statements ("They are not reasoning", "It does nothing of the sort.", "FFS lmao"), and basing your insults on this premise ("Ai Psychosis", "Know the difference."), so surely you have an extraordinarily solid ground to support that - rather than just speculation, innuendo, and a lot of confidence. P.S. I don't see why you're equaling "reasoning" with "behaving like a human". P.P.S. We don't even know if LLMs are conscious (in any way, shape or form, however foreign). We simply do not know. They might. Minds far smarter than you and me tried to answer this question, and they couldn't prove nor disprove it. So feel free to speculate, but anytime anybody makes absolute statements regarding this, they are either overconfident, underinformed, or both. | | |
| ▲ | mcmcmc 6 hours ago | parent [-] | | I would argue reasoning implies agency, and LLMs have none. They only act on input. | | |
| ▲ | blackoil 5 hours ago | parent [-] | | That is a harness issue. LLM needs a body to provide sensory inputs plus an infinite loop of what's next. |
|
|
|
|
