pizza

29 ms·

451.
by pizza 2y ago
Very cool. Also related, the Riffusion project, which is a finetuned text-to-audio model, based on the CLIP vision model! It also outputs spectrogram images (but moreso music-inclined).
452.
by pizza 2y ago
This style of embeddings could be quite lightweight/cheap/efficient https://github.com/cohere-ai/BinaryVectorDB
453.
by pizza 2y ago
Largest solar storm in 20 years
454.
by pizza 2y ago
It keeps the relevant information local to the site of execution and fairly explicit. These operations can have quite a few inputs where you need to keep track of, e.g. the size of every axis of every input, that you would otherwise have to
455.
by pizza 2y ago
Simplicity depends on the observer
456.
by pizza 2y ago
Wouldn’t WWDC-related endeavors be more product-facing? I’m not so sure this has to do with their efforts to incorporate ai into products, and tbh I would say their ai research has been pretty strong generally speaking.
457.
Personal Saving Rate (PSAVERT)(fred.stlouisfed.org)
3 points by pizza 2y ago|0 comments
458.
by pizza 3y ago
I remember pytorch has some pytree capability, no? So is it safe to say that the any-pytree-compatible modules here are already compatible w/ pytorch?
459.
by pizza 3y ago
Not exactly the same game but you might be interested in Mathematical Structure of Syntactic Merge, Marcolli, Chomsky, Berwick (2023) . When we speak we give a string. When we think we don't have to use a string. But we do have to hav
460.
by pizza 3y ago
I'm just speculating here since I don't know what or where the code is but since inference is still autoregressive; given [a b c] sample [d] distribution of [d] could be over [reasoning token] | [vocab token] then at next step you
461.
by pizza 3y ago
Think of it like an edge flow matrix
462.
by pizza 3y ago
You could just learn the right estimated number of recursions, also passing 'backtracking'/'state' information at the next nested level. Kind of like how state space models encode extractible information via a basis
463.
Macron says France and its allies 'could have stopped' the 1994 Rwanda genocide(foxnews.com)
7 points by pizza 3y ago|3 comments
464.
by pizza 3y ago
You could under-volt or watt-limit a bit and lose just a fraction of FLOPS for much less heat/power though, depending on the workload
465.
by pizza 3y ago
I think the main benefit is a better decorrelation transform so the compression is higher at the same quality parameter. So you could choose whether you want better accuracy for the same quality parameter, or lower the quality parameter and
466.
by pizza 3y ago
If you want a good read that (to summarize it quite tersely) takes the idea of 3- or n-body simulations and goes very far indeed with it (i.e. why there is something rather than nothing), I highly recommend checking out Julian Barbour'
467.
Reverse Mode Differentiation Is Kind of Like a Lens(philipzucker.com)
1 points by pizza 3y ago|0 comments
468.
by pizza 3y ago
I think maybe another way to put this is that Alice's number is in a typical set [0] of the distribution of bitstrings whereas Bob's might not be. Depending on the tolerance, the typical set can have near-total coverage of the dis
469.
UK court denies bid to extradite Assange to the US(mercurynews.com)
29 points by pizza 3y ago|0 comments
470.
by pizza 3y ago
tree-sitter?
471.
by pizza 3y ago
Depends on the complexity of your task, the quality and breadth of your data, and how aligned the task is with what the pretrained model was pretrained for. Anywhere from very cheap and quick to very expensive and tricky.
472.
The Clock (2010 Film)(en.wikipedia.org)
1 points by pizza 3y ago|0 comments
473.
by pizza 3y ago
Reminds me a lot of David Spivak’s “ologs”, down to the example of how dollars are a vector space
474.
by pizza 3y ago
4090 or 3090 are reasonable choices. You want the GPUs with a lot of VRAM. Seen a lot of people running 2-3x used 3090s for reasonable-ish prices (ie <2k usd). If you want higher speed go for 4090s if you can afford it and your computer
475.
by pizza 3y ago
Maybe prolog (or other logic programming language) or z3 (or other SAT solver)?
476.
by pizza 3y ago
The mixing is the generalization
477.
Krautrock singer Damo Suzuki dies aged 74(theguardian.com)
3 points by pizza 3y ago|0 comments
478.
by pizza 3y ago
Can you add the cost of the minimum reservation inline too? E.g. if it were $1/gpu/hr and "Minimum 1 GPU for 1 week" then make that text "Minimum 1 GPU for 1 week ($168)". This would allow faster comparison bet
479.
FDA says 561 deaths tied to recalled Philips sleep apnea machines(cbsnews.com)
418 points by pizza 3y ago|297 comments
480.
by pizza 3y ago
- too much frantic hyperactivity for too long - too short deadlines so you have to dip into 'personal' energy to meet them in time - too much concurrent spillover of past work and getting new/constantly shifting objectives so

More ›