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

GPT Unicorn has drawn a unicorn

gpt-unicorn.adamkdean.co.uk

Post reply on HN
Print view View original

101–110 of 207 posts

Re: GPT Unicorn has drawn a unicorn

#101

Post by sheepscreek »

sigmoid10 wrote: post #15

It has become common knowledge that GPT4 (and also 3.5) have problems with deterministic outputs (even at T=0). So what we're seeing here is just the effect of random sampling, not any actual change to the model itself. If you scroll down, you'll see other close attempts by the exact same model that could already be counted as a win depending on who you ask. Edit: This comment section is a super fascinating case stud…

Agree with this except for one data point. OpenAI does enhance/tweak/do something with the models at different levels. This can be determined by:

1. A change in the current model number (eg. gpt-3.5-turbo-0613)

2. On ChatGPT UI, the date at the bottom (eg. August 2023)

So it isn’t correct to say “it is incredibly obvious that nothing has happened”. Not that obvious to me.

A bit like how you can never tell for sure if Coca Cola has tweaked their formula, or McDonalds has changed the recipe for its signature sauce. Only in this case, the model number going up or the date becoming more recent leads credence to something having changed.

Re: GPT Unicorn has drawn a unicorn

#104

Post by noduerme »

sigmoid10 wrote: post #15

It has become common knowledge that GPT4 (and also 3.5) have problems with deterministic outputs (even at T=0). So what we're seeing here is just the effect of random sampling, not any actual change to the model itself. If you scroll down, you'll see other close attempts by the exact same model that could already be counted as a win depending on who you ask. Edit: This comment section is a super fascinating case stud…

I have not kept up with GPT architecture, other than noticing that other people have noticed that T=0 is clearly not deterministic for these things (and that that results from a bug, not an intentional feature). This much was obvious when the supposedly genius idiots rolled out GPT-3. It's wonderful to see the whole world bend over and just take it up the ass from a bunch of people who can't figure out why their code can't produce the same result twice; but it's quite natural for there to be a big cheerleading section on HN for any new technology that's (1) brilliant in theory, (2) deeply anti-human in practice, and (3) just needs a couple more revisions before it "fixes" a bunch of stuff.

Re: GPT Unicorn has drawn a unicorn

#105

Post by sigmoid10 »

sheepscreek wrote: post #101
sigmoid10 wrote: post #15

It has become common knowledge that GPT4 (and also 3.5) have problems with deterministic outputs (even at T=0). So what we're seeing here is just the effect of random sampling, not any actual change to the model itself. If you scroll down, you'll see other close attempts by the exact same model that could already be counted as a win depending on who you ask. Edit: This comment section is a super fascinating case stud…

Agree with this except for one data point. OpenAI does enhance/tweak/do something with the models at different levels. This can be determined by: 1. A change in the current model number (eg. gpt-3.5-turbo-0613) 2. On ChatGPT UI, the date at the bottom (eg. August 2023) So it isn’t correct to say “it is incredibly obvious that nothing has happened”. Not that obvious to me. A bit like how you can never tell for sure if…

The ChatGPT UI is indeed a wildcard. But it is irrelevant here because according to the github repo this page queries the API and OpenAI guarantees it doesn't change models with version number information (like gpt-4-0613, which is mentioned in the latest images). So this "experiment" would make a lot more sense if it was only run once every few months when the API actually offers new model versions and then generate a bunch of images for every model, instead of generating one single image every day (which is meaningless due to non-deterministic noise, even if the model had somehow changed since yesterday). That is also how it was done in the original study during the development of GPT4. I don't know how this experiment came up with its logic, unless they had a gross misunderstanding about how these models actually work (which admittedly seems common among tech interested folks here).

Re: GPT Unicorn has drawn a unicorn

#107

Post by kykeonaut »

imtringued wrote: post #89
rav wrote: post #49

Earlier quoted context omitted.

One important source of non-determinism is from using massive parallelism together with floating point arithmetic. In real math, a sum of numbers has an exact value that doesn't change if you change which order the numbers are added up in, but floating point arithmetic addition is not associative in the same way as real math, and parallelism can cause numbers to be added in a different order from execution to executi…

>and parallelism can cause numbers to be added in a different order from execution to execution Parallelism doesn't magically add non-determinism of this kind unless you intentionally build it to be non deterministic. Nothing prevents you from processing an array in order in parallel.

However, the poster mentions parallelism in conjunction with floating point arithmetic, not parallelism by itself.

Re: GPT Unicorn has drawn a unicorn

#108

Post by bhouston »

calderknight wrote: post #98
bhouston wrote: post #64

Earlier quoted context omitted.

Yes, this makes sense. You are agreeing with me. In order to see if it is just a fluke or not, you need a lot of samples so you can characterize the distribution yourself and try to see if it changed.

OK but I'm still not seeing any logic error.

That a single sample per day can track changes meaningfully when the noise floor is above the signal strength. And also the fact that today's unicorn-ish sample is meaningful at all.

It is a fun experiment though.

Re: GPT Unicorn has drawn a unicorn

#109

Post by gpjanik »

fastball wrote: post #56
gpjanik wrote: post #45

The prompt must be particularly bad. I managed to get a nicely looking unicorn at the first (and every subsequent) attempt.

The prompts are thus: > system: You are a helpful assistant that generates SVG drawings. You respond only with SVG. You do not respond with text. > user: Draw a unicorn in SVG format. Dimensions: 500x500. Respond ONLY with a single SVG string. Do not respond with conversation or codeblocks. What were yours?

My prompt was:

Imagine you have to draw a SVG of an object. As a model that does not have any idea about how things look, you have to draw "blindly" - as there's no visual feedback, the only feasible tactic is to first list things components each thing consists of (e.g. for a car wheels, windows, chassis, bumpers, lights, etc.) with as much accuracy as you can, establish some constraints (e.g. in a horse legs come out of the body, ears come out of the head, and so on), and then attempt to put all of it in a SVG. This is your task for now, and I will evaluate your drawings. Give me HTML code with embedded SVG that you drew and be verbose about both the things you're going to draw and the constraints.

The first thing you will draw is a unicorn.

Re: GPT Unicorn has drawn a unicorn

#110

Post by itsoktocry »

sigmoid10 wrote: post #15

It has become common knowledge that GPT4 (and also 3.5) have problems with deterministic outputs (even at T=0). So what we're seeing here is just the effect of random sampling, not any actual change to the model itself. If you scroll down, you'll see other close attempts by the exact same model that could already be counted as a win depending on who you ask. Edit: This comment section is a super fascinating case stud…

>This comment section is a super fascinating case study on the inherent flaws in human cognition. Especially when it comes to seeing patterns in random noise. The fact that some people believe that the model really has to have changed in the past few days is amazing

You need only to look at the discourse around the Tesla FSD superusers to see this: they report a glitch at an intersection one day, then believe the next day it was "fixed" by the AI.

Post reply on HN