zone411

39 ms·

451.
by zone411 4y ago
The performance gap to the top-end Nvidia cards will get much larger as they release new cards later this year, though.
452.
by zone411 4y ago
Umm training is not the same as running it.
453.
by zone411 4y ago
Around 6 seconds.
454.
by zone411 4y ago
> "Stable Diffusion was trained primarily on LAION-Aesthetics, a collection of subsets of the LAION 5B dataset, which is itself a subset of LAION-400M" Err, no? As you can tell by the name itself, the 5B dataset is larger. >
455.
by zone411 4y ago
I played a bit with smoothly interpolating between Stable Diffusion prompts and the effect can be pretty cool but it's hard to avoid discontinuities (like the object changing its orientation), even when using some additional tricks lik
456.
by zone411 4y ago
You may want to try bfloat16 instead.
457.
by zone411 4y ago
It's not quite at that level yet. The paper introducing it recommends using only 5 images as the fine-tuning set so the results are not yet very accurate.
458.
by zone411 4y ago
It should be noted that the official repo now also supports Stable Diffusion: https://github.com/rinongal/textual_inversion .
459.
by zone411 4y ago
https://pubmed.ncbi.nlm.nih.gov/31830487/ "Rodent studies also demonstrate that ketamine rapidly increases BDNF and VEGF release and/or expression in the medial PFC (mPFC) and hippocampus, leading to increase
460.
by zone411 4y ago
Yes, it's inevitable. Some parts of this process are harder than than others, though. I think I've handled the hardest part, which is making catchy melodies ( https://www.youtube.com/playlist?list=PLoCzMRqh5SkFwkumE
461.
by zone411 4y ago
Maybe at some point for some cameras. But not soon after their release if they took steps to protect their pipeline with hardware.
462.
by zone411 4y ago
It'll be hard to deny crypto-signed photos https://petapixel.com/2022/08/08/sonys-forgery-proof-tech-ad... , especially if they include metadata that distinguish photos of AI generated images from normal
463.
by zone411 4y ago
Check out the melodies I made with an AI assistant I created (human-in-the-loop still but much quicker than if I tried to come up with them from the scratch): https://www.youtube.com/playlist?list=PLoCzMRqh5SkFwkumE578Y... .
464.
by zone411 4y ago
You don't need an A6000. You can run it on consumer GPUs.
465.
by zone411 4y ago
This is no longer true about Ferrari. "Ferrari’s first all-electric car will arrive in 2025." https://www.wsj.com/articles/inside-ferraris-plan-to-enter-t...
466.
by zone411 4y ago
I'm not saying that it does, just that like you definitely hear more about Blender than about Maya on Hacker News, you will hear more about an open source model than a commercial one.
467.
by zone411 4y ago
"DALL-E" in titles is about to be replaced with "Stable Diffusion." The beta website is already live but the interesting part will be specialized fine-tuned models based on public weights. There should be more technical
468.
by zone411 4y ago
They jumped the gun with this announcement. I get wanting to share the excitement of AI doing something cool with the world (I've been there) but they should've waited until it's accessible to the public.
469.
by zone411 4y ago
> You could probably invent a kind of ketamine that does absolutely nothing interesting and yet still has the benefits. Maybe it can be done but the real reason that the companies are working on this is that they can patent these new dru
470.
by zone411 4y ago
They actually show promise for stopping additions.
471.
by zone411 4y ago
Read a meta study like https://mindmedicineaustralia.org.au/wp-content/uploads/Long... . From section 3.3-3.4, you can see that psychedelic use is beneficial for anxiety and depression and results in improved well-
472.
by zone411 4y ago
I have some first-hand experience about how the copyright office views these works from creating an AI assistant to help me write these melodies: https://www.youtube.com/playlist?list=PLoCzMRqh5SkFwkumE578Y... . Here is a qu
473.
by zone411 4y ago
Foreign companies can trade on the U.S. public markets. Baidu, Sony, and Telefônica Brasil do, it doesn't make them American companies.
474.
by zone411 4y ago
Spotify.
475.
by zone411 4y ago
I got them recently as the side monitors for a 16:9 43'' 4k in the middle. The height matches almost perfectly. I love this layout so far, so much desktop space and such high resolution! And it's nice to have a big 16:9 scree
476.
by zone411 4y ago
1. Try to get more data 2. Augmentation 3. Heavy regularization 4. Cross-validation 5. Transfer learning/fine-tuning from related data 6. Sharpness aware optimization 7. Ensembles 8. Any domain knowledge/feature engineering
477.
by zone411 4y ago
I shorted this Nasdaq-listed fraud: https://www.sec.gov/litigation/litreleases/2020/lr24706.htm . It was a wild story that included a 2000% price increase after they bought a blockchain company, Russell 2000 i
478.
by zone411 4y ago
I meant for the AI. I do currently need to be involved creatively to create these melodies but I figure that it might not even be necessary with some more advances.
479.
by zone411 4y ago
I've mentioned this before - I'm using the melodies I've created together with my AI songwriting assistant as input for further training. This requires human judgement and I only have around 100 melodies that I've judged
480.
by zone411 4y ago
To each their own but I'll take Ave Maria a capella any time: https://www.youtube.com/watch?v=su0JxmJGPpU . I think there is a good reason that it's the melody that gets stuck in our heads for earworms.

More ›