zone411

27 ms·

361.
by zone411 4y ago
I think OpenAI made two strategic blunders: 1. Publishing the Dall-E papers and pre-trained CLIP weights. This inspired Stable Diffusion/MidJourney, reducing the amount of time OpenAI had for their Dall-E service to get established by
362.
by zone411 4y ago
Wow, that's one of the best misspellings you could have.
363.
by zone411 4y ago
I predict that this will a research area that advances hugely after we solve text-to-video and text-to-3D (so maybe in 2 years?): learning these types of heuristics from billions of hours of videos. Implementing this in robots is another ma
364.
by zone411 4y ago
It's an interesting question. Solving extremely difficult math problems, e.g. the Riemann hypothesis or P vs NP, would be very impressive. Or coming up with algorithms given many examples of desired results. There should be many benefi
365.
by zone411 4y ago
Still funny how it loves six fingers per hand
366.
by zone411 4y ago
Hmm that seems excessively slow. I'm assuming you're using multithreaded compiling and SSDs? Any huge libraries? I'd try Build Insights. It seems worth investigating because 800 source files is not really that many.
367.
by zone411 4y ago
My experience (C++): - 3 bugs marked as "Our teams prioritize action on product issues with broad customer impact" and never fixed - 1 bug fixed - 1 bug merged with another bug and fixed
368.
by zone411 4y ago
/permissive- is now the default.
369.
by zone411 4y ago
This has been going for many years already, with Google giving answers right in the search results and just using websites as their private content factories. The relationship has become more and more uneven every year, with no pushback fro
370.
by zone411 4y ago
You can read all the gory details here: https://arxiv.org/pdf/2207.00032.pdf
371.
by zone411 4y ago
Chinese is well-represented among ML researchers.
372.
by zone411 4y ago
I think you will see an approach like mine with a heavy use of AI assistants in creating various elements of the songs first (I will compare 10 melodies I've created with an AI assistant I made to human hit melodies: https:/&#x2F
373.
by zone411 4y ago
I haven't seen a paper on this either, but I think that if it does work, somebody should definitely write one about it because it seems quite consequential. I am not sure if the underlying reason would be regularization though, seems m
374.
by zone411 4y ago
I've seen that happen before, but it was always a temporary artifact of the chosen architecture, optimizer, loss function, or training details that would disappear later on once they were improved and then offer no help. I have little
375.
by zone411 4y ago
Nice write-up. From my experience, the finding that "predicting additional details about each car (e.g. make and model) improved the accuracy of price predictions" is indeed surprising and possibly a red flag that something is not
376.
SatelliteXplorer(geoxc-apps.bd.esri.com)
2 points by zone411 4y ago|0 comments
377.
by zone411 4y ago
It is true that ChatGPT's answers do not have a deeper underlying meaning because it does not have a consistent model of the world. Many humans don't either. However, it is possible for a chat model to be biased in one way or anot
378.
by zone411 4y ago
I had a recent example of ChatGPT clearly beating Google: "how to debug a Chrome extension in VSCode?" After 10 minutes of reformulating this, I couldn't get a good answer from Google, but ChatGPT got it on the first try. Man
379.
by zone411 4y ago
Speaking of drug-inspired ideas (though it was just boring old weed), I recently had the idea that AI is the ultimate medium that will finally enable effective many-to-many communication that the Internet promised but hasn't quite deli
380.
by zone411 4y ago
Direct link to the web demo: https://znah.net/lenia/
381.
by zone411 4y ago
I would like to see the top list of most upvoted users but only counting their comment scores, not submissions.
382.
by zone411 4y ago
True. I think they're counting duplicated code though. I don't see any mention of de-duplication in their paper.
383.
by zone411 4y ago
No. It shows the opposite. All model sizes converged to a similar loss as the compute increased towards maximum. But larger models had larger loss for a given compute budget. Their text about Figure 3 confirms what I'm saying: "We
384.
by zone411 4y ago
I've actually seen this paper before, but I don't think it's helpful. If the entire GitHub is 100B tokens and your prune it down properly, then fine, you can get equal performance with fewer tokes. However, if you want improv
385.
by zone411 4y ago
It doesn't work like this. A 1T model without architectural changes would not perform substantially better unless it has been trained on a lot more code. The original Codex was trained on 100B tokens, so you could possibly get some gai
386.
by zone411 4y ago
Interesting. I experimented a bit with the approach of using diffusion on whole audio files, but I ultimately discarded it in favor of generating various elements of music separately. I'm happy with the results of my project of composi
387.
by zone411 4y ago
I've been following this case and I'm definitely on the no-cheating side, but it should be mentioned that what made this case weird and polarizing was that the employee of the production team was caught stealing $15k of her chips
388.
by zone411 4y ago
They're using a CNN for the decoder, not a transformer.
389.
SBF Arrested by Bahamian Authorities(twitter.com)
1308 points by zone411 4y ago|812 comments
390.
by zone411 4y ago
One connection I've seen is energy-based models. I suppose you could try applying ML to things like fluid dynamics? In general, it is thought that physics knowledge transfers better to ML than math because it's less abstract and p

More ›