Closed afiaka87 closed 3 years ago
As it stands, it is perhaps a bit misleading as to the current state of things. Unless there's a snail checkpoint out there I'm unaware of...
They are taken from OpenAI's blog post afaik.
Why are they on here then? That's incredibly misleading.
Sorry - I come more from the perspective that not everyone here is a researcher (or maybe it's just me?).
For non-researchers, the distinction between the name of the trained system/model and the name of the architecture used to train that system is confusing. I'm honestly not even sure I got the wording on that sentence right! That we choose to name these efforts after the architecture while OpenAI is happy to use the architecture's name to refer to just, sort of, either one of them, is misleading. We can't do anything to change the way OpenAI behaves in this regard, so it's on us to be responsible about it.
It's a distinction that caused much confusion when GPT-Neo hit the front page of hacker news (largely software engineers, not researchers, in my experience). In terms of open source "PR", or whatever, I assume they lost at least one potential open source contributor that day because of how misleading it felt. The top comment was a disclaimer.
@afiaka87 hmm, i had no intent on deceiving, its just the image being used by media outlets for DALL-E
I've updated it with Kobiso's results, which I find equally impressive
The snails on the README have always been curious to me. I wasn't around when that was produced - but it's definitely not a result any of us have been even close to producing yet. What's the story there? Was it made with this repository?