SleekEagle

9 ms·

1. 61.▲
   by SleekEagle 4y ago

   Yeah, category theory isn't even broadly well studied amongst actual mathematicians lol. I'm all for learning for the sake of learning though!
2. 62.▲
   by SleekEagle 4y ago

   I had this thought earlier when thinking about negative prompts in SD. Also, it would be really cool to plug in a PFGM as the base in Imagen while maintaining Diffusion Models for the super-resolution chain As for the addition, in principle
3. 63.▲
   by SleekEagle 4y ago

   There is a stochastic element to diffusion models which is not present in PFGMs. Further, recall that Diffusion Models and PFGMs alike are "metamodels" - mathematical frameworks in which any function approximator can, in theory, b
4. 64.▲
   Image Generation with Electrostatics(assemblyai.com)

   53 points by SleekEagle 4y ago|10 comments
5. 65.▲
   Stable Diffusion faster in Keras thanks to XLA(assemblyai.com)

   4 points by SleekEagle 4y ago|0 comments
6. 66.▲
   by SleekEagle 4y ago

   You highlight that AlphaTensor can be used to optimize for a variety of distance metrics (speed, memory, numerical stability, energy usage, etc.). Would the distance metrics be what changed if you were using AlphaTensor to find better matri
7. 67.▲
   Generative deep learning that uses electrodynamics(assemblyai.com)

   1 points by SleekEagle 4y ago|0 comments
8. 68.▲
   Deep Learning with Light(news.mit.edu)

   2 points by SleekEagle 4y ago|0 comments
9. 69.▲
   by SleekEagle 4y ago

   Thanks! Yeah, I've felt for a while that the greatest strides in deep learning will come from novel approaches that pull from higher level math/physics (or from novel approaches to automating the data cleaning process)
10. 70.▲
    Poisson Flow Generative Models – new image generation model inspired by physics(assemblyai.com)

    8 points by SleekEagle 4y ago|2 comments
11. 71.▲
    OpenAI Whisper Benchmarks(assemblyai.com)

    3 points by SleekEagle 4y ago|0 comments
12. 72.▲
    Show HN: Audio Intelligence Dashboard(huggingface.co)

    3 points by SleekEagle 4y ago|0 comments
13. 73.▲
    How to Build an Audio Intelligence Dashboard(assemblyai.com)

    2 points by SleekEagle 4y ago|0 comments
14. 74.▲
    by SleekEagle 4y ago

    It can run on higher end consumer grade GPUs at a batch size of 1, but there are already forks making it more efficient, so the landscape is evolving very quickly right now!
15. 75.▲
    How to Run Stable Diffusion to Generate Images (Local and Cloud)(assemblyai.com)

    7 points by SleekEagle 4y ago|0 comments
16. 76.▲
    by SleekEagle 4y ago

    That's great! I'm glad you're enjoying them :)
17. 77.▲
    by SleekEagle 4y ago

    I'll be sure to post an update! It's not a Diffusion Model, but I believe this VQGAN-CLIP implementation has weights available - https://github.com/nerdyrodent/VQGAN-CLIP
18. 78.▲
    by SleekEagle 4y ago

    My pleasure! Thanks for the kind words :)
19. 79.▲
    by SleekEagle 4y ago

    Thanks so much! I learned a ton with this project so it's great to hear that I communicated it well :)
20. 80.▲
    by SleekEagle 4y ago

    Interesting, thanks for the link. It seems that CLIP encodings aren't as useful as frozen encoders from the textual domain, which is a little unintuitive imo. Can't keep up with all these advancements!
21. 81.▲
    by SleekEagle 4y ago

    Maybe I'm wrong that it was DeviantArt, but I read something about this. Sorry I don't have a source, I should've checked before mentioning it. The closest thing I could find were rules only for specific groups! https:/
22. 82.▲
    by SleekEagle 4y ago

    This is the first time I've seen Stable diffusion. Is it a new model or diffusion paradigm?
23. 83.▲
    by SleekEagle 4y ago

    Haha great point! AI always wins
24. 84.▲
    by SleekEagle 4y ago

    You don't think in a few years AI art will be indistinguishable for human generated? In 2012, self driving cars were a funny joke. A decade later, here they are. I think AI is somehow chronically both overestimated and underestimated
25. 85.▲
    by SleekEagle 4y ago

    Exactly, it's just not feasible. Maybe they will train a discriminator to determine which ones are generated, but I don't think that would work very well. Also, DALL-E 2 images come with a watermark but I'm pretty sure there
26. 86.▲
    by SleekEagle 4y ago

    And, of course, these models learn in part from the art generated by these artists. If we stop having humans create art, we're no longer generating data to train the next generation of models and so in some sense the "creativity&q
27. 87.▲
    by SleekEagle 4y ago

    I think for a creative website like that is intended to showcase artists it makes sense. Unfortunately I don't think there's much they can do about losing in the market to text-to-image models in the long run... the costs are esse
28. 88.▲
    by SleekEagle 4y ago

    We'll be training it over the coming weeks and releasing a checkpoint! Right now it is just the model source code
29. 89.▲
    by SleekEagle 4y ago

    Appreciate it! And I know ... I feel like we're really hitting a watershed moment with ML/DL. It looks like artists on DeviantArt have already objected to AI-generated art being allowed on the website or suggested a mandatory wate
30. 90.▲
    Build Your Own Imagen Text-to-Image Model(assemblyai.com)

    111 points by SleekEagle 4y ago|43 comments