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

Language models still struggle with the concept of negation

quantamagazine.org

Post reply on HN
Print view View original

61–70 of 172 posts

Re: Language models still struggle with the concept of negation

#61

Post by furyofantares »

akavi wrote: post #14

Does this increase anyone else's sense that LLMs are actually modeling something similar to the way humans think? It certainly does for me. I find that negations/conjugations of negations are surprisingly difficult for to parse. Beyond two negations at most, I can feel myself having to switch into "code" mode, where I'm explicitly casing out the booleans, to understand them. Obviously the examples given in the articl…

"Don't think of a pink elephant" comes to mind with this stuff as well, where telling someone not to think about something causes them to think about it.

Re: Language models still struggle with the concept of negation

#62

Post by foobarbecue »

byyyy wrote: post #29
hn_throwaway_99 wrote: post #11

Earlier quoted context omitted.

While the answer landed on bats, ChatGPT kind of trips itself up in its explanations: 1. "Most winged creatures either have paws (like bats) or lay eggs (like birds and insects)" - but bats don't have paws. 2. "However, if we consider the term "wings" more broadly" - why do we need to consider the term "wings" more broadly? Nobody is arguing there is some definition of "wing" that would exclude bats. Bat wings evolve…

> but bats don't have paws. Bat paw: https://upload.wikimedia.org/wikipedia/commons/e/e1/Bat_in_a... ChatGPT was totally in tune with this concept. The appendage on the elbow of the wing can be a paw or not a paw but it certainly evolved from paws. The reasoning was spot on and amazingly nuanced.

What? No. You're rationalizing what is clearly an error. Truth still exists, and bats do not have paws, any more than humans do.

Re: Language models still struggle with the concept of negation

#63

Post by seanmcdirmid »

scheme271 wrote: post #38
ordu wrote: post #18

Earlier quoted context omitted.

> Bats have wings and no paws. Can hind limbs of bats be classified as paws? English is not my native language, so I'm just being curious.

No, not really. Paws kind of implies something that's fairly fleshy and soft and which bats don't have. Claws would be appropriate description of what bats have.

I saw a huge bat in Ubud Bali who was definitely eating a peach with their paws. I guess they were closer to hands? But definitely not claws.

Re: Language models still struggle with the concept of negation

#64

Post by canjobear »

akavi wrote: post #14

Does this increase anyone else's sense that LLMs are actually modeling something similar to the way humans think? It certainly does for me. I find that negations/conjugations of negations are surprisingly difficult for to parse. Beyond two negations at most, I can feel myself having to switch into "code" mode, where I'm explicitly casing out the booleans, to understand them. Obviously the examples given in the articl…

Here's a good example of the difficulty of processing negation for humans. "No head injury is too trivial to ignore" -- What does this mean?

Re: Language models still struggle with the concept of negation

#65

Post by usaar333 »

jostmey wrote: post #6
williamcotton wrote: post #5

What animals don’t have paws or lay eggs, but have wings? Work it out, step by step. ChatGPT (GPT-4): Sure, let's go through this step by step: First, you asked for animals that don't have paws. This would exclude mammals such as dogs, cats, bears, and so on, as these species have paws. Second, you specified animals that don't lay eggs. This eliminates a wide range of animals including all birds, most reptiles, and s…

From what I’ve seen, there’s a huge difference in chat GTP 4 from other LLM

Ya, it doesn't really struggle with the examples in this article to the point it feels like it invalidates the article.

Re: Language models still struggle with the concept of negation

#66

Post by benreesman »

williamcotton wrote: post #5

What animals don’t have paws or lay eggs, but have wings? Work it out, step by step. ChatGPT (GPT-4): Sure, let's go through this step by step: First, you asked for animals that don't have paws. This would exclude mammals such as dogs, cats, bears, and so on, as these species have paws. Second, you specified animals that don't lay eggs. This eliminates a wide range of animals including all birds, most reptiles, and s…

And if you don’t force it with the “step-by-step” cheat code?

I know what both 3.5-turbo and 4 say on my account, but I think I might be in some cost-effective inference holdout group because it’s always so dumb.

Re: Language models still struggle with the concept of negation

#67

Post by chaxor »

benreesman wrote: post #66
williamcotton wrote: post #5

What animals don’t have paws or lay eggs, but have wings? Work it out, step by step. ChatGPT (GPT-4): Sure, let's go through this step by step: First, you asked for animals that don't have paws. This would exclude mammals such as dogs, cats, bears, and so on, as these species have paws. Second, you specified animals that don't lay eggs. This eliminates a wide range of animals including all birds, most reptiles, and s…

And if you don’t force it with the “step-by-step” cheat code? I know what both 3.5-turbo and 4 say on my account, but I think I might be in some cost-effective inference holdout group because it’s always so dumb.

What's 375792*7385482? You can't use paper, or think about it more than a second. If you get it wrong, it's clear that you are not intelligent.

See how ridiculous that sounds?

CoT or tokens are not 'cheating' - they allow the system to have (as an analogy) a sheet of paper to work out the problem, instead of relying on a single step to jump through thousands of steps of reasoning. It's trained on the algorithmic reasoning embedded within natural language (TransformersGATs+PE⊂GNNs, which are dynamic programmers, and align well with algorithmic reasoning), so the steps are necessary to lay out in as much detail.

Re: Language models still struggle with the concept of negation

#68

Post by IanCal »

gregsadetsky wrote: post #8

This reminded me of https://github.com/elsamuko/Shirt-without-Stripes Whose days may be soon over as gpt4 gets this overall right..? As noted by many folks here

Gpt4 + klarna plugin does that just fine. Gpt4 decided to search for plain shirts.

Re: Language models still struggle with the concept of negation

#69

Post by hammyhavoc »

hn_throwaway_99 wrote: post #11
williamcotton wrote: post #5

What animals don’t have paws or lay eggs, but have wings? Work it out, step by step. ChatGPT (GPT-4): Sure, let's go through this step by step: First, you asked for animals that don't have paws. This would exclude mammals such as dogs, cats, bears, and so on, as these species have paws. Second, you specified animals that don't lay eggs. This eliminates a wide range of animals including all birds, most reptiles, and s…

While the answer landed on bats, ChatGPT kind of trips itself up in its explanations: 1. "Most winged creatures either have paws (like bats) or lay eggs (like birds and insects)" - but bats don't have paws. 2. "However, if we consider the term "wings" more broadly" - why do we need to consider the term "wings" more broadly? Nobody is arguing there is some definition of "wing" that would exclude bats. Bat wings evolve…

I was sat here WTFing that some bat species laid eggs until I looked it up. This is why you can't trust the output.

Re: Language models still struggle with the concept of negation

#70

Post by hammyhavoc »

jiggawatts wrote: post #10
jostmey wrote: post #6

Earlier quoted context omitted.

From what I’ve seen, there’s a huge difference in chat GTP 4 from other LLM

Interesting that it contradicted itself by saying bats have paws and don’t have paws, but arrived at the correct conclusion anyway.

"A broken clock is right twice per day" as the saying goes.
Post reply on HN