How to build a working AI only using synthetic data in just 5 minutes

Re:How to build a working AI only using synthetic data in ju... by lumost on Monday June 13, 2022 @11:41AM (#31727090)

I’ve heard such claims for a long time, I can likewise create a classifier out of a simple dice role. It doesn’t say anything about how good it is.

Most software applications have low tolerance for error rates, the ones that do have big money being spent on ensuring their accuracy is better than everybody else’s.

So while you can make a classifier out of anything in Y time, that doesn’t say anything about whether it’s of any practical use.

[ Parent | on HN ]

Re:How to build a working AI only using synthetic data in ju... by minimaxir on Monday June 13, 2022 @11:44AM (#31727152)

No, CLIP is indeed that good. The robustness of its embeddings is the entire reason why VQGAN+CLIP works and can stablely generate images close to the text prompt.

[ Parent | on HN ]

Re:How to build a working AI only using synthetic data in ju... by lumost on Monday June 13, 2022 @12:06PM (#31727416)

I'm sure it's better than a dice role, but does it beat a modern classifier trained on the domain specific data?

EDIT: I raise this issue as over-promises are the death nell for software. Overpromising capability leads to disappointment.

[ Parent | on HN ]

Re:How to build a working AI only using synthetic data in ju... by minimaxir on Monday June 13, 2022 @12:10PM (#31727486)

See the zero-shot performance of CLIP: https://openai.com/blog/clip/

It's definitely better performance than what you'd get working in 5 minutes from more conventional approaches.

[ Parent | on HN ]

Re:How to build a working AI only using synthetic data in ju... by ShamelessC on Monday June 13, 2022 @05:51PM (#31732245)

It tends to, yes. I suggest reading the paper as they discuss this very thing in detail.

[ Parent | on HN ]