sdenton4

32 ms·

571.
by sdenton4 2y ago
I always turn off self view because it is incredibly distracting.
572.
by sdenton4 2y ago
Fusion is only eight minutes out, and always has been...
573.
by sdenton4 2y ago
If you think they weren't trying to replicate sj's voice, I can only assume you've got open ai stock clouding your judgement... It looks like they caaaaarefully avoided writing down 'sound like sj' while choosing+di
574.
by sdenton4 2y ago
Black widow spiders last something like a thousand eggs, and the hatchlings murder and easy one another until only a couple survive... Who go in to live for as long as black widows live for. The average lifespan is likely quite low, however
575.
Bento3D(polar-tadpole-97b.notion.site)
238 points by sdenton4 2y ago|74 comments
576.
by sdenton4 2y ago
This strikes me as equally "AI complete" as drawing hands, which is now essentially a solved problem... No one test is sufficient, because you can add enough training data to address it.
577.
by sdenton4 2y ago
That doesn't seem in any way useful, though... To use a very blunt analogy, are color blind people intelligent/sentient/whatever? Obviously, yes: differences in perceptual apparatus aren't useful indicators of intelligen
578.
by sdenton4 2y ago
Amen, brother.
579.
by sdenton4 2y ago
No, the corpus is not the source. It's data. So we can have concepts of open models, open source, and open data. Any combination of these can be chosen independently. (Open data and open model but not open source is a bit weird, but no
580.
by sdenton4 2y ago
Are humans able to destroy all this habitat because they've got a better brain architecture, because they are able to achieve higher brain mass (because they don't need to fly to survive), or because they have opposable thumbs? Th
581.
by sdenton4 2y ago
Humans are an n=1 example, is my point. And there's no direct competition between bird brain architecture and mammalian brain architecture, so there's no reason for one architecture to 'win' over the other - they may bot
582.
by sdenton4 2y ago
Birds are under a different set of constraints than non-bat mammals, of course... They're very different. Songbirds have ~4x finer time Perception of audio than humans do, for example, which is exemplified by taking complex sparrow son
583.
by sdenton4 2y ago
User: which is the larger number, 3 or 2? ChatGPT: The larger number is 3. User: If I have one apple, pick another apple and then get an apple from a friend, and Alice has two apples, who has more apples? ChatGPT: You have more apples. You
584.
by sdenton4 2y ago
Think about what industrial farming says about humans, and be horrified.
585.
by sdenton4 2y ago
DiskANN is worth taking a look at; their game is fast graph based search using an index on disk.
586.
by sdenton4 3y ago
Found that paper on frontends: https://www.sciencedirect.com/science/article/pii/S157495412...
587.
BirdCLEF 2024 – Birdsong ID Competition(kaggle.com)
1 points by sdenton4 3y ago|1 comments
588.
by sdenton4 3y ago
This is a competition I have been co-organizing for a number of years now. The goal is to use 'focal' bird recordings to identify species in passively recorded soundscapes. We focus on a different understudied region each year, wh
589.
by sdenton4 3y ago
For bioacoustics there are significant problems with domain shift, label shift, label imbalance, and sample bias in training data. You need models to generalize to new data with very different noise profiles than the available training data
590.
by sdenton4 3y ago
For those who want to try it a real world version of this, the BirdClef 2024 competition is currently running: https://www.kaggle.com/competitions/birdclef-2024 Lots more I can say here - I've been working on pro
591.
by sdenton4 3y ago
That's the point of this paper: investigating whether 'chain of thought' promoting kinda-works because it actually induces reasoning, or whether it's just that more verbose answers give the model more tokens to work with
592.
by sdenton4 3y ago
Of course, it would be easy to put such questions together, and I'm sure the LLM would do fine with them - there's a massive amount of human text about human relationships. One difference, as in all ml training, is interactivity.
593.
by sdenton4 3y ago
"Consider this: can an mlp approximate a sine wave?" Well, yes - we have neutral speech and music synthesis and compression algorithms which do this exceedingly well...
594.
by sdenton4 3y ago
"The models, in other words, do not well generalise to understand the relationships between people." Curiously, the need to do this well - handling the quadratic complexity of a shifting set of human relationships, grudges, and al
595.
by sdenton4 3y ago
You get embeddings at every activation layer of the network, at every token. That's extra state accessible to the network when running in recurrent 'generate the next token' mode.
596.
by sdenton4 3y ago
Wellll.... Watching that video now, it looks like the most-selected number in their poll of 200k people was 69, followed by 42. Once /those/ are eliminated, 37 and 73 stand out. https://youtu.be/d6iQrh2TK98?si=SOpc
597.
by sdenton4 3y ago
Humans are also notoriously bad at choosing random numbers... Does that mean they are undependable and inherently untrustworthy?
598.
by sdenton4 3y ago
Very reminiscent of parallel wavenet, which speed things up by generating multiple audio samples at a time.
599.
by sdenton4 3y ago
An important piece here is that there's still a training signal making it to the makes weights. See SimSiam for a similar example.
600.
by sdenton4 3y ago
Sadly, pitchfork has recently been gutted. Wouldn't be surprised if the new masters are doing this purely for PR...

More ›