Not everything in black and white makes sense.
If I understand it right you need to enrich and filter data with human input so as not to collapse the model.
Wouldn’t that imply if the human enrichment is emulating AI data too closely it will still collapse the model, since it’s now just the human filtering that’s mimicing AI data?
I’m speaking about the firmware and other blobs that are there because devices wouldn’t work without it.
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/tree/WHENCE
Nice, something running in an eBPF context with a blob in the middle, what could go wrong …
Also there are already a lot of binary blobs in the kernel, that also makes me nervous a bit.
What are you using to ship the logs to VL?
If you want to exclude “normal” logs you should start excluding them before they reach VL, so the only logs you have are the interesting ones.
The more you can filter and label at the source, the less you have to work out in VL.
I use alloy (which is kinda heavy) to extract and prepare only the data I want and it works great so far.
I agree with you on some points here. The problem is that these crawlers are hostile to the point of DDOSing sites
So the problem is not that someone archives your public accessible data, the problem is that in doing so either breaks your site or makes you pay for the excess traffic.
I think the web is now broken beyond repair. The commercialisation killed it and the tech monopolies are all that’s left.
So I think small invite only fully encrypted enclaves are all that is left, until someone comes up with a “new Internet”, that can resist the " Techbros", but for now I don’t see that.
Also I don’t see the Fediverse as a solution, it’s just under the radar for now, but if it gets bigger it will be coöpted and sunk.
First they filled the web full of ads, then they let loose an army of bots and now they have to come up with solutions to identify and track real humans, so as not to waste their precious advertising.