This is almost certainly an independent discovery/implementation of the same idea, there's really only one sensible way to do it. There are, of course, several non-sensible ways to do it as well, but...
This was first done by Markku-Juhani Saarinen in June 2000 for Komplex Labs, https://www.komplex.org/. Apart from the multi-line text message that it contained as the key it had other suspicious markers like a negative validity period over 1 January 1951 and it not being signed (it used e=1), it was still accepted as valid by both Windows and Firefox.
It's a good thing we're not depending on this stuff to keep us safe or anything.
It's fine if you stay away from it's raid5/6. And realize it isn't going to let you know when you have a mirror fail until you try to boot it and it won't come up.
Oh, neat, this mentions my patch to rust-brotli! FWIW, I do not, an never have, worked for Dropbox. ;-)
A similar tool would be possible for gzip - I made one a long time ago that could append a literal stored block to the end of a gzip stream.
One note about brotli - appendable chunks after the first will generally have a worse compression ratio because they can't reference brotli's built in preshared dictionary, but my use case was appending a small amount of dynamic data to an otherwise static file, where this isn't a significant problem.
I've experienced the same, IIRC also using NVIDIA's NVFP4 quant. Also just decided to ignore because it didn't seem to cause any real issues. I figure it might be a training thing, since the hallucinated user messages seem to occur immediately after tool calls or when it's checking its work.
No they don’t. The GDPR is location scoped so the data of an EU citizen given to an hotel while on vacation in the US isn’t protected by GDPR but the data of an US citizen giving their data to a hotel in the EU while on vacation is.
That is my understanding as well. Not a lawyer but recently looked into it since we have US, EU, and non-EU European employees and we were looking at token usage tracking tools (which would appear to potentially fall under employee surveillance or at the very least require explicit consent)
Not sure the exact legalese but legal put together a consent form for anyone wanting to do the PoC
To be clear, I find the claim that the models are sentient and have feelings to be preposterous, but I feel like the key takeaway is:
> It doesn't actually matter to you if they have feelings, because you will still get demonstrably better results across the board if you treat them as if they do.
However, personally, my main reason for that mode of interaction is that I interact with a lot of humans via text, and I do not want to habituate myself to being an asshole when the medium is words on a screen (or otherwise).
Yes, I know that (see the venerable https://github.com/trapd00r/colorcoke, etc.) but those tricks are not used widely enough for them to be a concern. Using those tricks is also a deliberate choice so it is definitely on the user if my lovingly crafted 256c colorscheme is broken.
Having all terminal emulators run the equivalent of colorcoke without asking the user is not a very bright idea.
This is almost certainly an independent discovery/implementation of the same idea, there's really only one sensible way to do it. There are, of course, several non-sensible ways to do it as well, but...
reply