Number 3481 of things you should *not* use generative models for : forensic sketches 🙈
(I'm not sharing the direct link to the code for this, but it's not that hard to find, either..)
Trying to see something.
If you’re a man, and you can understand why a woman would be upset about deepfake porn being sold of her non-consensually, can you RT?
Restore my faith in humanity 🙏🏻
How is it possible that the company that made billions if not close to trillions off of SEARCHING AND ORGANIZING the web can’t make sensible, easy to follow instructions for their own products and produces as maze that puts the dark web to shame just to get to the right support.
Why didn't Tyre Nichols comply? Why did he run?
It appears clear that he was scared and did not trust these police officers with his life.
Their actions, as captured on video, proved him right.
"The supreme art of war is to subdue the enemy without fighting."
― Sun Tzu, The Art of War
context: Sun Tzu was realizing he could demoralize any enemy with a combination of dependency issues, missing libraries, build failures, then topping it off with a broken docker image
I wonder if Surge could get some love to support Tensors and match Numpy / PyTorch?
Having some parity with these API’s would do wonders for native data science tooling within native apps.
Curious if ya’ll have any thoughts:
The best thing Apple could do for the community is introduce Swift extensions for MLShapedArray that try to match semantics from Numpy / PyTorch to ease porting of code.
That would be so incredible. I imagine there has to be an internal code base for that at Apple !?
Has anyone written a MLShapedArray <-> Numpy / Torch cheat sheet or API reference?
The semantic differences in returned tensor ranks in operations is so difficult to grok so porting code to Swift is like a mine field.
It is only the 22nd day of the year and there have already been at least 36 mass shootings in the United States.
https://gunviolencearchive.org/reports/mass-shooting…
This is going great ya’ll. Good job. “Well regulated”
I'm thrilled to be joined on the Science Committee by my Republican colleague Dr. George Santos, winner of not only the Nobel Prize, but also the Fields Medal - the top prize in Mathematics - for his groundbreaking work with imaginary numbers.
It’s true—dolphins were spotted in the Bronx River this week! This is great news—it shows that the decades-long effort to restore the river as a healthy habitat is working. We believe these dolphins naturally found their way to the river in search of fish.
(Video: Nick Banco)
holy fucking shit, Al Franken posed for a stupid photo and Democrats forced him to resign. George Santos is a fucking felony factory and Republicans rewarded him with prime committee assignments. maybe it's just me but this is just fucking nuts
AFNetworking is now officially deprecated. Direct integration, forking, or transitioning to Alamofire is recommended moving forward. Thanks for your support over the last decade!
Does anyone else who uses multiple programming languages daily also unintentionally switch between camel case and snake case for variables within even the same function or method? My code looks like a psychopath wrote it.
Apparently I suck at coding FFT’s because I could not for the life of me get this fucking goddamned Log Mel Spectrogram to work with vDSP. I can’t remember being this annoyed with a coding challenge in a while.
So, after a small false start this weekend, Whisper CoreML is working 4x realtime with the base multilingual model. 0.7x realtime for Small ( sad face) - with no ANE Transformer optimizations.
https://github.com/vade/OpenAI-Whisper-CoreML/tree/feature/RosaKit…
Using RosaKit for the Log Mel generation with minor tweaks.
Whisper CPP is actually faster! (Amazing work btw holy cow - I didn’t anticipate decode loop to be so slow). This however uses way less power / CPU. In theory ANE performance can be greatly improved:
Good news about our environment!
The Montreal Protocol has succeeded in safeguarding the ozone layer.
The phase out of nearly 99% of banned ozone-depleting substances has the ozone layer on track to recovery within four decades.
More from
have some ideas? Sorry for the spam. I’m ‘ ‘ close, and I’m 99% sure the Log Mel Spectrogram being off is whats giving me bad output. This thing is fast otherwise!
😅
Did Photoshop remove the ability to open just like, raw data without any format? Am I crazy or wasn’t there a way to just open a file, say “pretend this is x y resolution, this is the bit depth just fucking go for it?