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

Claude's Cycles [pdf]

www-cs-faculty.stanford.edu

Post reply on HN
Print view View original

361–370 of 376 posts

Re: Claude's Cycles [pdf]

#361

Post by qsera »

empath75 wrote: post #346
qsera wrote: post #327

Earlier quoted context omitted.

People don't hallucinate. That is they can pretty reliably assess if they know or don't know something.

Your comment is a perfect example of a human hallucinating something and not knowing they are wrong about it. People are confidently wrong about things _all the time_.

No no, you don't understand. People can misunderstand. But they will not, for example, proceed to drive a car as if they have attended driving lessons when they have not.

They might misremember, but they can know, for sure, if they have NOT come across some information. So if you ask someone if they know where `x` is, they might have came across that info, and still be wrong. But they will know if they have never come across it.

A neural network will happily produce an output when when the input is completely out of range of the training data.

Re: Claude's Cycles [pdf]

#362

Post by marcus_holmes »

bitexploder wrote: post #343
marcus_holmes wrote: post #340

Earlier quoted context omitted.

The dolphin question, for me, is about what we'd even communicate with a creature that lives in such a different world. Humans mostly live in a 2D environment, for instance - we walk on flat planes, rarely looking up. We always have the ground beneath us, the unattainable sky above. Dolphins live in a 3D space, visiting the air above regularly to breathe, the "ground" below a varying distance away. I have no idea how…

They still have mammalian wet-ware. The dolphin has a relatively advanced neocortex which means they likely have some relatively advanced processing. They also have an interesting part of their brain that we don't have and it is likely for social and emotional information based on their behavior. We suspect they may even have a model of the self. They still have roughly the same kind of hardware as we do. Their diffe…

Good points, interesting.

I think this is probably neuroscience vs psychology. We can explain a lot with neuroscience, but two people with essentially identical brain chemistry can have very different psychology. There are people out there who have beliefs and cognition processes that I find completely incomprehensible despite having the same brain and even sharing a language.

I'm not sure how I'd have a meaningful conversation with an animal that has such a different worldview. I guess there's a simple level of conversation, like that which we have with dogs - fetch the stick, good boy, food, need to wee, love the human, etc. But if that's the limit of what we can discuss with dolphins (or an actual AI) then I'd be disappointed.

Re: Claude's Cycles [pdf]

#363

Post by throw310822 »

qsera wrote: post #361
empath75 wrote: post #346

Earlier quoted context omitted.

Your comment is a perfect example of a human hallucinating something and not knowing they are wrong about it. People are confidently wrong about things _all the time_.

No no, you don't understand. People can misunderstand. But they will not, for example, proceed to drive a car as if they have attended driving lessons when they have not. They might misremember, but they can know, for sure, if they have NOT come across some information. So if you ask someone if they know where `x` is, they might have came across that info, and still be wrong. But they will know if they have never com…

> if you ask someone if they know where `x` is, they might have came across that info, and still be wrong. But they will know if they have never come across it.

False memories are super common. "I thought I had seen this thing there, but turns out it's not" is a perfectly normal, very frequent occurrence.

Re: Claude's Cycles [pdf]

#365

Post by bitexploder »

marcus_holmes wrote: post #362
bitexploder wrote: post #343

Earlier quoted context omitted.

They still have mammalian wet-ware. The dolphin has a relatively advanced neocortex which means they likely have some relatively advanced processing. They also have an interesting part of their brain that we don't have and it is likely for social and emotional information based on their behavior. We suspect they may even have a model of the self. They still have roughly the same kind of hardware as we do. Their diffe…

Good points, interesting. I think this is probably neuroscience vs psychology. We can explain a lot with neuroscience, but two people with essentially identical brain chemistry can have very different psychology. There are people out there who have beliefs and cognition processes that I find completely incomprehensible despite having the same brain and even sharing a language. I'm not sure how I'd have a meaningful c…

But at some level, you can "just be" with the other organism. Eat some food. Make some dopamine. Hang out. I feed my dog. I exercise my dog. I exercise myself. I eat. We sit down together, I pet her. We both create oxytocin and perceive that positively when I pet her. Most animals map that to "safety" or "contentment". Survival needs satisfied for now. Who knows what that maps to for my dog, but we exist in a pretty similar state in that moment of being. That very desire to try and map the dolphin is our "I" narrative that /constantly/ wants to map things out and figure the patterns out.

Dolphin has concept maps between objects and semantic meaning and an "I" narrative. Dog is almost fully present with no narrative constantly mapping past to future. We probably have a lot more in common with dolphin, if we can map that somehow.

https://www.frontiersin.org/journals/psychology/articles/10....

This article is right up this conversation's alley; about chimps being fascinated with crystals. And I am not saying it is wrong to map and communicate, communication means cooperation, deeper connection and meaning, discovering boundaries of if we can socially coordinate and form new and exciting groups and collaboration, etc.

Re: Claude's Cycles [pdf]

#366

Post by klooney »

mccoyb wrote: post #2

It's fascinating to think about the space of problems which are amenable to RL scaling of these probability distributions. Before, we didn't have a fast (we had to rely on human cognition) way to try problems - even if the techniques and workflows were known by someone. Now, we've baked these patterns into probability distributions - anyone can access them with the correct "summoning spell". Experts will naturally us…

> Experts will naturally use these systems more productively, because they know how to coerce models into the correct conditional distributions which light up the right techniques.

How much can you patch over with the models doing their own metacognition?

Re: Claude's Cycles [pdf]

#367

Post by qsera »

throw310822 wrote: post #363
qsera wrote: post #361

Earlier quoted context omitted.

No no, you don't understand. People can misunderstand. But they will not, for example, proceed to drive a car as if they have attended driving lessons when they have not. They might misremember, but they can know, for sure, if they have NOT come across some information. So if you ask someone if they know where `x` is, they might have came across that info, and still be wrong. But they will know if they have never com…

> if you ask someone if they know where `x` is, they might have came across that info, and still be wrong. But they will know if they have never come across it. False memories are super common. "I thought I had seen this thing there, but turns out it's not" is a perfectly normal, very frequent occurrence.

If someone ask "Hey how do I do this thing in python programming language" what are the chances that you will try to make up a solution, if you have never tried to learn Python?

Just tell me..

Re: Claude's Cycles [pdf]

#368

Post by UltraSane »

qsera wrote: post #367
throw310822 wrote: post #363

Earlier quoted context omitted.

> if you ask someone if they know where `x` is, they might have came across that info, and still be wrong. But they will know if they have never come across it. False memories are super common. "I thought I had seen this thing there, but turns out it's not" is a perfectly normal, very frequent occurrence.

If someone ask "Hey how do I do this thing in python programming language" what are the chances that you will try to make up a solution, if you have never tried to learn Python? Just tell me..

Models ARE slowly improving at this

https://artificialanalysis.ai/evaluations/omniscience

Re: Claude's Cycles [pdf]

#369

Post by lhl »

pushedx wrote: post #307
lhl wrote: post #305

I was a bit interested to do a replication and see if better harness could avoid some of the problems they ran w/ context management, poor instruction following, etc and it looks like yes, it's definitely possible. Here's my repo: https://github.com/lhl/claudecycles-revisited I used Codex w/ 5.2 xhigh and a relatively simple AGENTS.md - I have some session-analysis as well. The original replication was 47 minutes, th…

As described in the readme of your repo (did you read it?) your agent found the Knuth paper located one directory level above its working directory. So, you didn't produce a replication in 47 minutes, it just took around 30 minutes for your agent to find that you had the answer in a PDF in a nearby directory.

Yes, I read it and specifically pointed it out (that's why there are 3 hours of interactive logs). There are 4 other runs pushed now so you can see what actual clean room runs for 5.2 xhigh, 5.3-Codex xhigh, 5.4 xhigh, and Opus 4.6 ultrathink look like: https://github.com/lhl/claudecycles-revisited/blob/main/COMP... as well as the baseline.

Re: Claude's Cycles [pdf]

#370

Post by throw310822 »

qsera wrote: post #367
throw310822 wrote: post #363

Earlier quoted context omitted.

> if you ask someone if they know where `x` is, they might have came across that info, and still be wrong. But they will know if they have never come across it. False memories are super common. "I thought I had seen this thing there, but turns out it's not" is a perfectly normal, very frequent occurrence.

If someone ask "Hey how do I do this thing in python programming language" what are the chances that you will try to make up a solution, if you have never tried to learn Python? Just tell me..

Have you tried this with Claude?

---

Q: How do I reverse an array in the Navajo programming language?

A: I'm not familiar with a programming language called "Navajo." It's possible you might be thinking of a different language, or it could be something very niche that I don't have information about.

---

As for your question, the chances go from 0 to 100% depending on how many languages I already know and whether I have an idea (or I think I have an idea) of how python looks like. And LLMs have seen (and tried to "learn") pretty much everything.

Post reply on HN