[dkislyuk]

> New techniques, like diffusion models, shrink down the costs required to train and run inference.

Since the postscript mentions that the article was co-written by GPT-3, perhaps this is one of the generated lines? :)

(compared to the prior SOTA for generative ML in images i.e. GANs, diffusion models use far more compute-heavy and are much slower!)

[algo_trader]

More importantly, SD was built by a team with just 1 phd, and fully open sourced. What is the moat for all these great startups?

Sure, Sequoia wants to hear all about your great startup proposals.

And it will (rightly) reject 99%, and finally fund a "known team that can execute" kidding not kidding

[zone411]

This is only because Stable Diffusion was able to use techniques openly published by the likes of DeepMind, Google Research, and OpenAI. It even directly used some open source code from OpenAI. It wouldn't exist otherwise. DALL-E 2 was published in January 2021. I'm doing generative music AI (melody only) and there is quite a bit of real research required to make progress (I've decided not to publish for now).

[echelon]

> I'm doing generative music AI (melody only)

I'm doing vocals! Can't wait for the seams to come together.