Sounds like a very cool project. I often use the ElevenLabs app/extension to listen to articles, but I would love to have a local option to do this. I'm on Windows though, so I look forward to your future updates.
EDIT So, I did a very preliminary search and it appears some people are concerned that HuggingFace is going to be banned for some reason or another are are building some type(s) of repository for open source models.
Here are a few mentions of it on reddit and twitter but really no in depth conversation about why it exists, what purpose it serves or how it's meant to be used. [1] [2]
There was another project mentioned called Tensorbay available at tensorbay[.]net and seems to have torrents for different models, weights, old scrapped websites that might be useful for training data & etc. But, I can't confirm the security or anything about it, so of course, visit all this stuff at your own risk.
I'd love to hear from whoever runs this project what it is and what it's for.
This seems to be the only way forward from what I can figure. Helium's main selling point is that it's essentially degoogled chromium + a few miscellaneous patches & full uBlock. But once Google completely strips all that out of Chromium project, that won't be a tenable option.
I'm not sure what Opera/Vivaldi/et al. use for their native adblocking, but Brave's rust adblocker makes the most sense to me. Really it's uBlock's filtering lists that keep the whole thing working anyway.
I think you're going to have to standardize what you consider to be knock offs and what's not, or you're going to end up in continuous quagmires like this trying to make a determination based on vibes.
You might even end up getting together a team of volunteers to look through them or it'll probably get overwhelming.
What determination did you use to make the original list?
So, how did you come up with the brand names and decide on what is a knock off and what is not? I looked over the Github and really didn't see any specific criteria mentioned.
How often would you want to run it and do this though? I propose some kind of tool that constantly generates random GDID data and sends it to Windows telemetry.
I think we need to determine where the GDID is being sent and under what circumstances and just spoof that data.
It's wild how restricting China's access to our GPUs led them to create some of the top & most widely used open source ai models, while the US's models seem to be few and far between.
It looks like Meta has given up on Llama, which was one of the most popular, in favor of going proprietary to make more money.
This also leads me to believe the discussion about China slipping propaganda into their models (the whole "ask it about Tiananmen Square" thing) was never really a thing or at least not a priority for them.
Only way to see what's going on is testing to see what's going on. Hopefully, someone who knows more about it than me can take a look at the packets and see what they contain.
The article links to this page, which was shared on HN yesterday. [1]
I feel like using wireshark to look at what's being sent back and forth from Windows telemetry, when using Edge, Chrome & etc should reveal what's being sent and recieved. Using MITM SSL spoofing should be able to intercept the packets.
It's been preached for years in the piracy community that it's a terrible idea to use Bittorrent over Tor as it strains their network resources, is incredibly slow and ends up giving legal liability to the exit nodes if they get DMCA requests or whatever from copyright holders (amongst many other problems). [1] I know some clients provide the option to use I2P and that seems to be a better option for a variety of different reasons. [2]
How do we define "mythos level" exactly outside of marketing buzz? I don't even think the majority of us can access Mythos yet even to make a comparison.