ai-discussions

[fiery tartan]

for instance even with our text language model, there are customers writing in every day who have different theories like “it’s better in between 3 am and 5 am”

[sturdy terrace]

Lol

[fiery tartan]

there is clip guidance but we don’t fidget w that — and def not for perf reasons

[sturdy terrace]

Thank you for the clarification

🙏

[fiery tartan]

my pleasure!

[sturdy terrace]

Ok, just did a variation of an old prompt and it produced great results. Can confirm.

That, or joanne just put me back on the old model

(kidding)

[sick quartz]

[fiery tartan]

LOOL

no i bet it was because it was after 8:30 pm

[sturdy terrace]

[fiery tartan]

(for those reading, i am kidding and this is a reference to a message i wrote above)

[sturdy terrace]

All is well, I have went through 10+ prompts and had some great results. Randomness and bias.

[tranquil plover]

Nice, good to know it can create the points required to form a pentagon - but it still can't do pentagons.

[cunning creek]

i really have to resist the temptation to do a deep dive on pentagons, like we did with the glass rings 😅

[surreal holly]

Hi folks,

Following up on what @thick storm & @cunning creek were saying, would it be possible to add a setting for people to return either 4 or 8 results?

Returning 4 is a far slower/less fun creative experience for big projects than 8 was

It also seems like you could easily show 8 without actually changing anything on your front-end AND would result in people like me using my credits twice as fast!

[thick storm]

I actually don't think it works how people are presuming it all works. I think once you make you submit your prompt, it's done. There's no precedent for scaling or "lower res" image previw stuff that we're seen. Even still, that isn't very useful, is it? At a low resolution, you really can't tell what the final quality of the image will be, imho. Hypothetically.

I also think that the reason we have 4 prompt images now is because OAI want to scale to a million users without bankrupting themselves. So, they can't "just let it show 8 images" because they're only letting each user ask for 4, so to speak, each time we submit a prompt we're saying to dalle "four images please!" they're not just "not showing you the rest"

So, more users can ask dall-e at the same time, rather than half the users being able to generate 8 images and a bunch of people being unable to use the service.

I know nothing, really, about the technology or their process, though. This is just my instinct. I think they're learning what is best for everyone, and it's all still a work in progress.

[fallow grail]

So Dalle-3, give or take a year for it to come out?