pizza

31 ms·

391.
by pizza 2y ago
It's definitely slept on. I do think it ought to be very powerful given enough compute to throw at it, hopefully. I think short description length algorithms such as simple compression algorithms or instant-ngp could be interesting to
392.
by pizza 2y ago
Do you mean something more than putting the current date / timestamp in the system prompt? Which is something I think most of the top chatbots are doing behind the scenes.
393.
Cyberpunk 2077: POV Photorealistic Motorbike Ride [video](youtube.com)
11 points by pizza 2y ago|0 comments
394.
by pizza 2y ago
For God knows why reason, the original PSPs used to come with an IR LED. I put a homebrew program on my PSP that let you control it, and fed it a txt file with thousands of TV IR codes. What a blast!
395.
by pizza 2y ago
Source?
396.
by pizza 2y ago
I remember Dario Amodei mentioned in a podcast once that most models won't tell you the practical lab skills you need. But that sufficiently-capable models would and do tell you the practical lab skills (without your needing to kno
397.
by pizza 2y ago
While we're on the topic of jupyter enhancements, would really love to be able to pop a cell off the pending execution stack if I realized running it would be a mistake and still have time before it gets there.. :^)
398.
by pizza 2y ago
In a way, some things like this are possible/have been explored. In Huffman coding, the code length is log2(length of path in binary tree to x) ~ log2(1/p(x)). BPE works by combining the most frequent pair of adjacent tokens into
399.
Freud at Home: At His Desk(freud.org.uk)
1 points by pizza 2y ago|0 comments
400.
Adam Optimizer Causes Privileged Basis in Transformer Language Models(lesswrong.com)
4 points by pizza 2y ago|0 comments
401.
Mouse study reveals new central action target of weight loss drug GLP-1 agonists(medicalxpress.com)
1 points by pizza 2y ago|0 comments
402.
by pizza 2y ago
I see, I had this notion that DirectStorage worked by somehow allowing the GPU cores to host multiple parallel threads each doing full requests on their own, but on further research I was mistaken - requests are still CPU-submitted
403.
by pizza 2y ago
Correct me if I'm wrong, but Microsoft's DirectStorage seems to me something like what the author is writing about. It lets you do eg massively parallel NVME file io ops from the GPU itself of lots of small files. This avoids the
404.
Combination Treatment Can Increase Human Insulin-Producing Cells in Vivo(mountsinai.org)
2 points by pizza 2y ago|0 comments
405.
The computational structure of consummatory anhedonia(cell.com)
4 points by pizza 2y ago|0 comments
406.
by pizza 2y ago
There's a quote I read somewhere once that I have struggled since to find the original source: "if something is 100% correct, it can't be called intelligence" edit: hope the source is not this one and thus originally fro
407.
by pizza 2y ago
off-topic: is there anything that aims to be a 'universal stdlib'? Something that lets you write eg some small spec about how your language operates and then it lets you add a fullblown polyglot batteries-included stdlib to your l
408.
by pizza 2y ago
From what I understand about ME/CFS, both post-exertional malaise and dysautonomia are symptoms- seems like stimulant use could make those things go either way
409.
by pizza 2y ago
It def is done (kv caching the system prompt prefix) - they (Anthropic) also just released a feature that lets the end-user do the same thing to reduce in-cache token cost by 90% https://docs.anthropic.com/en/docs/
410.
by pizza 2y ago
It applies to all JPEGs.
411.
by pizza 2y ago
They note that their discovery relating VAT activity increase to weight loss wasn't the original focus of their study, so they hadn't prepared an additional weight loss control group - I think they leave open both appetite suppres
412.
Microplastics are infiltrating brain tissue, studies show(theguardian.com)
26 points by pizza 2y ago|1 comments
413.
by pizza 2y ago
In practice there is a difference because FFT would have more edge artifacts at block boundaries - lowering visual quality - and DCT has better energy compaction into lower frequencies meaning longer runs of zeros of higher frequency coeffi
414.
by pizza 2y ago
FastAI course for 1, the Little Book of Deep Learning [0] / Deep Learning (Goodfellow et al) / the entire set of Kevin Murphy textbooks for 2 (in order of detail) [0] https://fleuret.org/francois/lbdl.html
415.
by pizza 2y ago
If the samples came from HN, I wonder how likely it is that the text is already a part of a dataset (ie common crawl snapshot) so that the LLMs have already seen them? edit: judging from the comments I saw, they were all quite recent, so I
416.
by pizza 2y ago
> Many of the teenage students I encountered seemed to be in a state of what I would call depressive hedonia. Depression is usually characterized as a state of anhedonia, but the condition I'm referring to is constituted not by an i
417.
Gravity, Entropy, and Life(rantonels.github.io)
3 points by pizza 2y ago|0 comments
418.
by pizza 2y ago
Depends on how fuzzy your definition of template is. If it's really structured, a simpler algorithm might work better. Otherwise maybe try https://github.com/Unstructured-IO/unstructured
419.
The LICORS Cabinet: Nonparametric Algorithms for Spatio-Temporal Prediction(arxiv.org)
1 points by pizza 2y ago|0 comments
420.
by pizza 2y ago
Tile a ton of these photos into one large image, grayscale it and threshold for high brightness, then apply a lossless image compression algorithm. Compare it with the combined size of the same process on each image individually. I bet (exc

More ›