He said he is "Asking GPT-4 to draw a unicorn every day to track changes in the model."

The variance he is seeing in the output is primarily the product of random chance, rather than changes in the model. Specifically this "unicorn" that he found today is likely just random chance and there was no changes in the model between yesterday and today that lead to it arising.

If he wanted to track changes in the model for real, he would have to ask multiple questions per day and try to infer some type of distribution characterization and then see if that changes over time. That is much more complex and not what he is doing.

This is just a curious experiment that doesn't mean much.

It takes a picture from GPT every day, so we'll be able to see if this was a fluke by looking at future days' outputs. I think it will work to track changes in the model.

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.

If you look closely, you can see that the model version is attached to each image, and the original blog post clearly states that asking the same model over multiple days is how they intend to track the variance. No bad logic there imo.

If you look at the examples from April, only 2 or 3 can be counted as unicorns. If in 3 months it's the reverse and only 2 or 3 can't be counted as unicorns, that would show a progressive improvement in the model. I agree we shouldn't take much from day N-1 to day N as there will be a lot of variance, but this can show us progressive improvement over model updates.

Perhaps in a few months half of the generated pictures will look like unicorns, perhaps in more months they will all be unicorns but 2 or 3 will look way more detailed instead of drawn by a 4 year old, etc. We just need to wait longer for the signal to break through the noise.

I do agree with your comment, especially this part:

> We just need to wait longer for the signal to break through the noise.

Currently, what we are observing is primarily noise with very little signal.

you're looking at it from the wrong time scale.

The difference over months or years is what is interesting.

I don't think anyone claims this is an iterative linear measure, rather than a step function.

SVG can present arbitrarily complex graphics. The underlying display tech supports what ever fidelity GPT will eventually mature into.

Has GPT plateaud? Will it be stuck forever at this hilariously naive level of competence at SVG art? Will it mature into Midjourney level competence? I have no frigging clue. Since the token context is so small I imagine it will put limitations to the complexity of SVG art piece.

But I don't know. And it's fun to have a daily measure.

As a software engineer with a penchant for graphics asking GPT to draw complex graphic shapes was one of the first tests I did for it. It's extremely interesting for me to collect progress data, no matter how noisy.

I have no idea if GPT will ever mature beyond these squiggles but if it does, this track record will have at least considerable artistic value, if nothing else.

* * *

And so, if we see that it /does/ get better, over the next few years, will that not lead us to ask /how/?

Let's think about it:

1. It has to output SVG [1] 2. It is given a text based representation of what it must draw[2] 3. It must then somehow convert words -- the concept of a unicorn: equine with a horn, white, maybe rainbows? -- into SVG code, and attempt to convey both their location, shape, colour, appearance, with code.

And keep in mind, this is just a token predictor. I doubt there is much data in its training that is this specific.

So while it's quite far from science, for me, it's a bit of fun and I get emails every now and then remarking on things like the turd of May (2023-05-18) and it lightens the mood every now and then, which I think ultimately, is worth it.

[1] System: You are a helpful assistant that generates SVG drawings. You respond only with SVG. You do not respond with text.

[2] User: Draw a unicorn in SVG format. Dimensions: 500x500. Respond ONLY with a single SVG string. Do not respond with conversation or codeblocks.

See: https://github.com/adamkdean/gpt-unicorn/blob/master/src/lib...

GPT-4 is really great at transferring concepts between domains.

That's one the reasons why GPT when it works, feels magical. SVG art does not need to be in it's training set, as long as it knows how to present geometric concepts in SVG.

A good unicorn would require capabilities something like "the outline of unicorn is composed of lines {...}." -> "export lines as svg".

Ok that makes perfect sense, thanks.

There's a linked blog post[0] that goes more into the methodology and reasoning.

"As mentioned in the hacker news discussion, the model doesn't change daily. [...] As OpenAI releases incremental updates, we'll see the model change automatically and be able to judge outputs. A single sample per day leads to quite different results, but that's fine I think. What I expect to see a year from now is an evolution of output. In variance: how varied are the outputs over a month?"

So yes, they could just produce 100 images with each new model release, but chose to spread those out over 1 per day instead. Is it the most scientific way to measure progress? No. Is it more fun and interesting to check back daily? Probably.

[0] https://adamkdean.co.uk/posts/gpt-unicorn-a-daily-exploratio...

Thanks, fun and light-hearted is the approach I went for