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

Notes on OpenAI's new o1 chain-of-thought models

simonwillison.net

Post reply on HN
Print view View original

341–350 of 659 posts

Re: Notes on OpenAI's new o1 chain-of-thought models

#341

Post by dailykoder »

simonw wrote: post #340
dailykoder wrote: post #318

I am mostly only an LLM user with technical background. I don't have much in-depth knowledge. So I have questions about this take: >the output token allowance has been increased dramatically—to 32,768 for o1-preview and 65,536 for the supposedly smaller o1-mini! So the text says reasoning and output tokens are the same, as in you pay for both. But does the increase say that it can actually do more, or does it just me…

I included that note because output limits are a personal interest of mine. Until recently most models capped out at around 4,000 tokens of output, even as they grew to handle 100,000 or even a million input tokens. For most use-cases this is completely fine - but there are some edge-cases that I care about. One is translation - if you feed in a 100,000 token document in English and ask for it to be translated to Ger…

Sure, your cases are perfectly reasonable. I just wish the LLMs had a "feel" about when to output long or short text. Always thinking about adding something like "be as concise as possible" is kinda tedious

Re: Notes on OpenAI's new o1 chain-of-thought models

#342

Post by kristianp »

_fs wrote: post #95
layer8 wrote: post #18

The o1-preview model still hallucinates non-existing libraries and functions for me, and is quickly wrong about facts that aren't well-represented on the web. It's the usual string of "You're absolutely correct, and I apologize for the oversight in my previous response. [Let me make another guess.]" While the reasoning may have been improved, this doesn't solve the problem of the model having no way to assess if what…

o1-preview != o1. In public coding AI comparison tests, results showed 4o scoring around 35%, o1-preview scoring ~50% and o1 scoring ~85%. o1 is not yet released, but has been run through many comparison tests with public results posted.

Good reminder. Why did OpenAI talk about o1 and not release it? o1-preview must be a stripped down version: cheaper to run somehow?

Re: Notes on OpenAI's new o1 chain-of-thought models

#343

Post by K0balt »

multjoy wrote: post #282
ramraj07 wrote: post #112

Earlier quoted context omitted.

Many of my PhD and post doc colleagues who emigrated from Korea, China and India who didn’t have English as the medium of instruction would struggle with this question. They only recover when you give them a hint. They’re some of the smartest people in general. If you try to stop stumping these models with trick questions and ask it straightforward reasoning systems it is extremely performant (O1 is definitely a step…

“Don’t be mean to LLMs, it isn’t their fault that they’re not actually intelligent”

In general LLMs seem to function more reliably when you use pleasant language and good manners with them. I assume this is because because the same bias also shows up in the training data.

Re: Notes on OpenAI's new o1 chain-of-thought models

#344

Post by z7 »

energy123 wrote: post #173
ironhaven wrote: post #133

So is o1 nicknamed “strawberry” because it was designed to solve the “how many many times does the letter R appear in strawberry” problem.

No, that was a coincidence according to an employee there

Source: https://x.com/polynoamial/status/1834312400419652079

Re: Notes on OpenAI's new o1 chain-of-thought models

#345

Post by rtakha »

fragmede wrote: post #151
pedrosorio wrote: post #110

Earlier quoted context omitted.

> It knows english at or above a level equal to most fluent speakers, and it also can produce output that is not just a likely output, but is a logical output This is not an apt description of the system that insists the doctor is the mother of the boy involved in a car accident when elementary understanding of English and very little logic show that answer to be obviously wrong. https://x.com/colin_fraser/status/183…

what's weird is it gets it right when I try it. https://chatgpt.com/share/66e3601f-4bec-8009-ac0c-57bfa4f059...

Perhaps OpenAI hot-patches the model for HN complaints:

  def intercept_hn_complaints(prompt):
    if is_hn_trick_prompt(prompt):
       # special_case for known trick questions.

Re: Notes on OpenAI's new o1 chain-of-thought models

#346

Post by bonoboTP »

maeil wrote: post #177
hmottestad wrote: post #161

Earlier quoted context omitted.

Reminds me of a trick question about Schrödinger's cat. “I’ve put a dead cat in a box with a poison and an isotope that will trigger the poison at a random point in time. Right now, is the cat dead or alive?” The answer is that the cat is dead, because it was dead to begin with. Understanding this doesn’t mean that you are good at deductive reasoning. It just means that I didn’t manage to trick you. Same goes for an…

There is no "trick" in the linked question, unlike the question you posed. The trick in yours also isn't a logic trick, it's a redirection, like a sleight of hand in a card trick.

Yes there is. The trick is that the more common variant of this riddle says that a boy and his father are in the car accident. That variant of the riddle certainly comes up a lot in the training data, which is directly analogous to the Schrödinger case from above where smuggling in the word "dead" is analogous to swapping father to mother in the car accident riddle.

I think many here are not aware that the car accident riddle is well known with the father dying where the real solution is indeed that the doctor is the mother.

Re: Notes on OpenAI's new o1 chain-of-thought models

#347

Post by slig »

andrew_eu wrote: post #280

I thought with this chain-of-thought approach the model might be better suited to solve a logic puzzle, e.g. ZebraPuzzles [0]. It produced a ton of "reasoning" tokens but hallucinated more than half of the solution with names/fields that weren't available. Not a systematic evaluation, but it seems like a degradation from 4o-mini. Perhaps it does better with code reasoning problems though -- these logic puzzles are es…

Hey, I run ZebraPuzzles.com, thanks for mentioning it! Right now I'm trying to improve the puzzles so that people can't "cheat" using LLMs so easily ;-).

Re: Notes on OpenAI's new o1 chain-of-thought models

#348

Post by edouard-harris »

bambax wrote: post #281

Near the end, the quote from OpenAI researcher Jason Wei seems damning to me: > Results on AIME and GPQA are really strong, but that doesn’t necessarily translate to something that a user can feel. Even as someone working in science, it’s not easy to find the slice of prompts where GPT-4o fails, o1 does well, and I can grade the answer. But when you do find such prompts, o1 feels totally magical. We all need to find…

> Results are "strong" but can't be felt by the user? What does that even mean?

Not every conversation you have with a PhD will make it obvious that that person is a PhD. Someone can be really smart, but if you don't see them in a setting where they can express it, then you'll have no way of fully assessing their intelligence. Similarly, if you only use OAI models with low-demand prompts, you may not be able to tell the difference between a good model and a great one.

Re: Notes on OpenAI's new o1 chain-of-thought models

#349

Post by ImHereToVote »

bonoboTP wrote: post #322
pedrosorio wrote: post #110

Earlier quoted context omitted.

> It knows english at or above a level equal to most fluent speakers, and it also can produce output that is not just a likely output, but is a logical output This is not an apt description of the system that insists the doctor is the mother of the boy involved in a car accident when elementary understanding of English and very little logic show that answer to be obviously wrong. https://x.com/colin_fraser/status/183…

This illustrates a different point. This is a variation on a well known riddle that definitely comes up in the training corpus many times. In the original riddle a father and his son die in the car accident and the idea of the original riddle is that people will be confused how the boy can be the doctor's son if the boy's father just died, not realizing that women can be doctors too and so the doctor is the boy's mot…

My codebases are riddled with these gotchas. For instance, I sometimes write Python for the Blender rendering engine. This requires highly non-idiomatic Python. Whenever something complex comes up, LLM's just degenerate to cookie cutter basic bitch Python code. There is simply no "there" there. They are very useful to help you reason about unfamiliar codebases though.

Re: Notes on OpenAI's new o1 chain-of-thought models

#350

Post by j_maffe »

accountnum wrote: post #325
rdtsc wrote: post #253

Earlier quoted context omitted.

Fluency is a necessary but not the only prerequisite. To be able to answer a trick question, it’s first necessary to understand the question.

No, it's necessary to either know that it's a trick question or to have a feeling that it is based on context. The entire point of a question like that is to trick your understanding. You're tricking the model because it has seen this specific trick question a million times and shortcuts to its memorized solution. Ask it literally any other question, it can be as subtle as you want it to be, and the model will pick u…

It does mean something. It means that the model is still more on the memorization side than being able to independently evaluate a question separate from the body of knowledge it has amassed.
Post reply on HN