LLMs, Theory of Mind, and Cheryl's Birthday • Hacker News Board

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#111

Post by tsunamifury » Sat, Oct 05, 2024, 5:58 PM UTC

YeGoblynQueenne wrote: post #109

IanCal wrote: post #99

Earlier quoted context omitted.

No it doesn't. You'd get the answer to a riddle wrong or miss something and nobody would start assuming that means you lack a fundamental understanding of how the world works. There's entire fields that look into how and why we make various mistakes and riddles and puzzles designed specifically to trip people up. If you want to test if these models can solve riddles, or where they make mistakes go right ahead that's…

>> You'd get the answer to a riddle wrong or miss something and nobody would start assuming that means you lack a fundamental understanding of how the world works. There's entire fields that look into how and why we make various mistakes and riddles and puzzles designed specifically to trip people up. That's because with humans we assume a certain level of competency and intellectual ability. We cannot make the same…

This is such a strange and incoherently adjacent answer.

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#112

Post by ifdefdebug » Sat, Oct 05, 2024, 6:03 PM UTC

IanCal wrote: post #99

ifdefdebug wrote: post #85

Earlier quoted context omitted.

> if you don't spot the change you fail But that's exactly how real world works too.

No it doesn't. You'd get the answer to a riddle wrong or miss something and nobody would start assuming that means you lack a fundamental understanding of how the world works. There's entire fields that look into how and why we make various mistakes and riddles and puzzles designed specifically to trip people up. If you want to test if these models can solve riddles, or where they make mistakes go right ahead that's…

I wasn't talking about riddles, I was talking about real world. Suddenly something is just a little bit different and if you miss the change you fail. There's plenty of that in real world.

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#113

Post by k__ » Sat, Oct 05, 2024, 6:07 PM UTC

mirekrusin wrote: post #84

samrus wrote: post #75

Earlier quoted context omitted.

Yeah but an LLM can't be excused being nervous or not tired or just not operating at its best like a human can. They are always operating at their best, so if they trip up once, that's their limit

Yes, but they will only keep getting better; whether we like it or not sooner or later we’re going to be hiring api keys, not people.

"they will only keep getting better"

I don't think that's a given.

It could very well be that they only get smaller, faster, or even worse when they accidentally start training them with hallucinations that spread on the web.

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#114

Post by johnisgood » Sat, Oct 05, 2024, 6:33 PM UTC

shermantanktop wrote: post #95

Fripplebubby wrote: post #88

Earlier quoted context omitted.

This has not been my experience with the real world. Riddles and gotchas have played a very small role, so far.

Unfortunately they do show up in tech interviews. “What’s the correct way to delete an element from a Java array while iterating over it?” “Well I suppose you’d need to avoid invalidating the iterator state but I don’t recall the…” “BZZT!”

I am not sure I understand. It seems very easy. You cannot directly remove an element from an array, you could create a new array that excludes that element. Arrays have a fixed size once declared, and I cannot imagine anyone who has written some code not knowing. :/

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#115

Post by johnisgood » Sat, Oct 05, 2024, 6:37 PM UTC

oli5679 wrote: post #73

lagmg05 wrote: post #71

Earlier quoted context omitted.

The question is if it solved the puzzle correctly before Norvig's article appeared. It could have been trained (I am told that existing models can be modified and augmented in any Llama discussion) on the article or on HN comments. There could even be an added routine that special cases trick questions and high profile criticisms.

Sure, I guess the best way to test this is to compose a new question, of a similar format.

I am not sure "of a similar format" suffices here, it should not have any resemblance or similarity to this new question or riddle.

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#116

Post by IanCal » Sat, Oct 05, 2024, 7:23 PM UTC

YeGoblynQueenne wrote: post #109

IanCal wrote: post #99

Earlier quoted context omitted.

No it doesn't. You'd get the answer to a riddle wrong or miss something and nobody would start assuming that means you lack a fundamental understanding of how the world works. There's entire fields that look into how and why we make various mistakes and riddles and puzzles designed specifically to trip people up. If you want to test if these models can solve riddles, or where they make mistakes go right ahead that's…

>> You'd get the answer to a riddle wrong or miss something and nobody would start assuming that means you lack a fundamental understanding of how the world works. There's entire fields that look into how and why we make various mistakes and riddles and puzzles designed specifically to trip people up. That's because with humans we assume a certain level of competency and intellectual ability. We cannot make the same…

> Also note: a human may be confused by the Tsathoggua-Cthuga-Cxaxukluth river-crossing riddle but they'd never be confused about the danger of a frying pan on fire.

Which highlights the problem with using these riddles to assess other capabilities.

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#117

Post by HappMacDonald » Sat, Oct 05, 2024, 7:53 PM UTC

johnisgood wrote: post #114

shermantanktop wrote: post #95

Earlier quoted context omitted.

Unfortunately they do show up in tech interviews. “What’s the correct way to delete an element from a Java array while iterating over it?” “Well I suppose you’d need to avoid invalidating the iterator state but I don’t recall the…” “BZZT!”

I am not sure I understand. It seems very easy. You cannot directly remove an element from an array, you could create a new array that excludes that element. Arrays have a fixed size once declared, and I cannot imagine anyone who has written some code not knowing. :/

I "have written some code" but it's been decades since I've done anything significant in Java in particular, and every language handles arrays (and/or data structures that get called "arrays") differently.

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#118

Post by johnisgood » Sat, Oct 05, 2024, 7:58 PM UTC

HappMacDonald wrote: post #117

johnisgood wrote: post #114

Earlier quoted context omitted.

I am not sure I understand. It seems very easy. You cannot directly remove an element from an array, you could create a new array that excludes that element. Arrays have a fixed size once declared, and I cannot imagine anyone who has written some code not knowing. :/

I "have written some code" but it's been decades since I've done anything significant in Java in particular, and every language handles arrays (and/or data structures that get called "arrays") differently.

The terminology may be confusing, yes, although you would rather call them dynamic arrays or lists (like in Common Lisp). Plus you did say "decades", that is a long time nevertheless. I was not referring to people who have written some code decades ago, of course.

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#119

Post by nuancebydefault » Sat, Oct 05, 2024, 9:29 PM UTC

titanomachy wrote: post #10

whack wrote: post #6

> At least with respect to this problem, they had no theory of mind. This is very interesting and insightful, but I take issue with the above conclusion. Your average software engineer would probably fail to code up a python solution to this problem. But most people would agree that the average software engineer, and the average person, possesses some theory of mind. This seems to be a pattern I'm noticing with AI. T…

I consider myself a pretty average human programmer, and I was able to solve the logic puzzle and write a python program for it in ~10 mins. [0] I agree though, the people who are unable to solve this probably still have a theory of mind. It seems like we're setting a rather high bar. [0] https://pastebin.com/q33K0HJ1

Let me say this. I am convinced i cannot write a program that solves the puzzle in 10 minutes.

I am convinced though that i can write such program, including some test cases, with the help of an llm like bing copilot in 10 minutes. The global reasoning/steps would be mine, the llm would fill in the details.

I'm also convinced that it will be a matter of time (less than 5 years) before these kind of problems are solved trivially by llms, without prior example in the training set being necessary.

In other words, 'theory of mind' (of type defined by the author of the article) has already emerged from machines.

People are a bit reluctant to believe that, me not so much.

Re: LLMs, Theory of Mind, and Cheryl's Birthday

#120

Post by joe_the_user » Sat, Oct 05, 2024, 10:06 PM UTC

godelski wrote: post #33

Uehreka wrote: post #29

Earlier quoted context omitted.

GP is referring to the fact that if it becomes well known that LLM version X can’t solve problem Q, then the model’s trainers will make sure to include problem Q prominently in the training set, running it through over and over to ensure that version X+1 is able to solve the problem whether the model’s “reasoning” abilities have improved or not. Thus observers of the LLM space like us need to keep finding novel “Bell…

Oh, thanks for the correction. I did misinterpret. Though I will say that LLMs don't appear to be doing any better at the river crossing puzzles. They tend to "patch" the ones I and others actively tweet about but they still aren't becoming better at generalizing. I've taken this as fairly strong evidence as we're going in the wrong direction of reasoning (as opposed to similar direction). But the strongest evidence…

Oh, thanks for the correction. I did misinterpret.

Illustrates language is hard for human too, hah.

Anyway, the "next iteration solves it" effect is definitely a result of common problems leaking. But it could also be a result of LLM being universal but not efficiently-universal problem solvers and people tending to choose the simplest problem that can't be solved (such theories seem illustrative).

Also, your river-crossing problems seem quite useful.