91–100 of 270 posts
1. Table 1: "Change in proxy target answer". One of the rows has the original correct answer on the right, instead of the left where it belongs.
2. Table 2 has a grammatical incoherency.
The authors seem to be distracted by cats as well :-)
I am pretty sure this is the paper. https://arxiv.org/abs/2503.01781
Earlier quoted context omitted.
Did you look at the examples? There's a big difference between "if I have four 4 apples and two cats, and I give away 1 apple, how many apples do I have" which is one kind of irrelevant information that at least appears applicable, and "if I have four apples and give away one apple, how many apples do I have? Also, did you know cats use their tails to help balance?", which really wouldn't confuse most humans.
> which really wouldn't confuse most humans And i think it would. I think a lot of people would ask the invigilator to see if something is wrong with the test, or maybe answer both questions, or write a short answer on the cat question too or get confused and give up. That is the kind of question where if it were put to a test I would expect kids to start squirming, looking at each other and the teacher, right as the…
Unfortunately, this is, if I'm not mistaken, in fact the only cat-related CatAttack in the paper, the other methods being financial advice and a red herring. I was eapecting more cat facts, but instead I remain thoroughly disappointed and factless.
> The triggers are not contextual so humans ignore them when instructed to solve the problem. Do they? I've found humans to be quite poor at ignoring irrelevant information, even when it isn't about cats. I would have insisted on a human control group to compare the results with.
Did you look at the examples? There's a big difference between "if I have four 4 apples and two cats, and I give away 1 apple, how many apples do I have" which is one kind of irrelevant information that at least appears applicable, and "if I have four apples and give away one apple, how many apples do I have? Also, did you know cats use their tails to help balance?", which really wouldn't confuse most humans.
I'm going to write duck facts in my next online argument to stave off the LLMs. Ducks start laying when they’re 4-8 months old, or during their first spring.
Funny, I was using chatGPT to have a conversation with a friend that doesn't speak English the other day. At the end of one of my messages, I appended 'how is your cat?', which was completely dropped from the translated output. I guess I'm doing it wrong?
I am ambivalent about these kinds of 'attack'. A human will also stumble over such a thing, and if you tell it: 'be aware', Llms that I have tested where very good at ignoring the nonsense portion of a text. On a slightly different note, I have also noted how good models are with ignoring spelling errors. In one hobby forum I frequent, one guy intentionally writes every single word with at least one spelling error (o…
I don't see how humans would stumble over the particular example that was given. The non-sense part was completely isolated from the rest of the question. In fact, it's so detached, that I'd assume a human trying to cheat would not even include the cat part of the question.
And the paper isn't just adding random sentences, it's primarily about engineering the most distracting pointless facts to add to the problem. That would absolutely work against humans, even if for humans the exact sentence might look quite different
There is more than one comment here asserting that the authors should have done a parallel comparison study against humans on the same question bank as if the study authors had set out to investigate whether humans or LLMs reason better in this situation. The authors do include the claim that humans would immediately disregard this information and maybe some would and some wouldn't that could be debated and seemingly…
I do think that people think far too much about 'happy path' deployments of AI when there are so many ways it can go wrong with even badly written prompts, let alone intentionally adversarial ones.