pizza

47 ms·

691.
C++ is a superset of a subset of C(blog.joren.ga)
4 points by pizza 3y ago|1 comments
692.
40 Years of Koyaanisqatsi(thecurb.com.au)
347 points by pizza 3y ago|118 comments
693.
by pizza 3y ago
Here’s my one-liner explanation: attention is the sum of a dict of functions evaluated on input data. But the data is matrices, the dict keys are matrices, and the functions that are found in the dict are also matrices.
694.
by pizza 4y ago
I guess wrt ieee floats, it depends on how often these cases happen and their consequences in the model: - rounding errors in arithmetic between small and large numbers, large and large numbers, or small and small numbers - unrepresentable
695.
by pizza 4y ago
Interesting- my vision of future LLM interface also envisions one where more bits of information per second are required per interaction to operate it to exactly the spec that you want. But for exactly because it’ll just be a plain old engi
696.
by pizza 4y ago
RLHF seems to suggest that human feedback to tune the model after plain textual data pretraining is quite potent per sample. There might be some optimal ratio of data+model size:rlhf size that works quite favorably for us in getting halluci
697.
by pizza 4y ago
as I understand it a cellular sheaf complex assigns vector spaces to nodes and linear transformations to edges
698.
by pizza 4y ago
Nice, I think you've just invented the first cellular sheaf db that I'm aware of
699.
by pizza 4y ago
functional Near Infrared Spectroscopy, like fMRI
700.
by pizza 4y ago
The owl of minerva always flies at dusk. Or so I’m told.. /s
701.
Why Is Working Memory Performance Unstable? A Review of 21 Factors (2018)(ncbi.nlm.nih.gov)
1 points by pizza 4y ago|0 comments
702.
Rocket Launch Alert Newsletter Mailing List(spacearchive.info)
1 points by pizza 4y ago|0 comments
703.
EditAnything: Segment Anything + ControlNet + BLIP2 + Stable Diffusion(github.com)
214 points by pizza 4y ago|24 comments
704.
by pizza 4y ago
One thing I've thought about is a static typed compiled neural network framework. To do backprop generally you create a computation graph and find the gradients from e.g. reverse-mode automatic differentiation. Just tracing along opera
705.
by pizza 4y ago
Ah, was just a flicker in my mind. Thanks - though I will ask, is there anything like the law of sines/cosines (in high dimensional statistics) that could get you to at least the first step of the proof?
706.
by pizza 4y ago
Huh, wonder how hard it'd be to get automatic differentiation working with this graph notation type system... groovy stuff
707.
by pizza 4y ago
It would be really interesting to see if this could be applied to other analogous scenarios that are sometimes called Pythagorean theorems, in particular I’m thinking of the Pythagorean Theorem of Information Geometry- p* = argmin p o
708.
First Names and Bayes’ Theorem(johndcook.com)
2 points by pizza 4y ago|0 comments
709.
by pizza 4y ago
Well, plenty of compression algorithms have been patented or attempted to be patented - e.g. arithmetic codes and Microsoft's rANS patent
710.
Zoom.earth – global live satellite and radar weather map(zoom.earth)
1 points by pizza 4y ago|0 comments
711.
Celebrities Die 2.7183 at a Time(ssp.impulsetrain.com)
2 points by pizza 4y ago|0 comments
712.
by pizza 4y ago
I think RWKV ameliorates this to some degree: How it works: RWKV gathers information to a number of channels, which are also decaying with different speeds as you move to the next token. It's very simple once you understand it. RWKV
713.
by pizza 4y ago
Linux thinclient as my dormroom desktop: some of my best grades semesters Windows gaming machine: some of my flakiest semesters
714.
by pizza 4y ago
- vim users who like configuring their vim. goes double for people who like scripting nvim with lua - python coders who keep trying to add haskell features to python - phil wang (@lucidrains on github) - people who have a strong stated pref
715.
by pizza 4y ago
It’s worth pointing out that there can be a distinction between the type of information in signals and the type of information in filters. I think people feel disturbed because they think the filters here are overfitted, therefore are jus
716.
by pizza 4y ago
My mistake it looks as though the desiderata was single client/server db backend in the first place, and remote db was just an added bonus of that - https://github.com/pgrok/pgrok/pull/11
717.
by pizza 4y ago
Hang on a second- this project looks to be assuming that the db will be remote and over the internet. Even the SQLite official document recommends people to use PostresSQL in that scenario [0]: >Generally, if your data is separated from
718.
by pizza 4y ago
Also edit to add: we have pretty magnificent tools for the mapping to a shared semantics of syntax- tree-sitter :: any syntax -> 1 semantics - we have tree-sitter, lsp, etc; any language can be added with an easy description. Bu
719.
by pizza 4y ago
unrelated but still interesting fact is Varoufakis used to be the "official Team Fortress 2 economist" for Valve, long ago - ie pre Diem-25 and pre- his government post [0] https://www.businessinsider.com/yanis-var
720.
Former Finance Minister Yanis Varoufakis Attacked and Beaten in Athens(greekreporter.com)
3 points by pizza 4y ago|1 comments

More ›