Yes they were, so I’m offering you an actual theory as to why this may actually be true, yet difficult to “prove”.
Smoking was bad for your health long before anyone sat down and took the time to prove it. Autoregressive LLM tokenizer are a very new field of computer science and it’s going to take a while for the community to collectively understand everything we’re currently doing by trial and error.
Anecdotally, I use it a lot and I feel like my responses are better when I’m polite. I have a couple of theories as to why.
More tokens in the context window of your question, and a clear separator between ideas in a conversation make it easier for the inference tokenizer to recognize disparate ideas.
Higher quality datasets contain american boomer/millennial notions of “politeness” and when responses are structured in kind, they’re more likely to contain tokens from those higher quality datasets.
I haven’t mathematically proven any of this within the llama.cpp tokenizer, but I strongly suspect that I could at least prove a correlation between polite token input and dataset representation output tokens
I think that holds true in this case… but I’ll be damned if the ltt screwdriver isn’t the best hand tool I’ve ever owned.
When ML training farms run out of new text to train on, “they” may very well want your original writing too…
Me too brother, but I disagree with your assessment on value
An non-blacklisted residential IP address with reasonable throughput is valuable in and of itself. DDOS botnets, proxies to bypass geo blocks or to obfuscate illicit traffic, etc. Also your gaming PC could be used for distributed compute workloads of compromised, usually crypto mining.
Any hardware/connection has value if it’s “free”. It’s just a numbers game beyond that.
For now, ctt winutil does a pretty good job at removing the cruft. I’ve long since switched to debian for my daily driver, but as a remote-access sunshine host for games that require kernel level anticheat, it’s surprisingly usable.
For anyone looking to keep windows around in some capacity, I strongly recommend it. https://github.com/ChrisTitusTech/winutil
Every packet you send/receive relies on passive security. Your nic drivers, the driver kernel model, all of the userland applications that sit on top of it. I get that in practical terms, your firewall will do a lot of the heavy lifting but there are passive rce vulnerabilities in previous unsupported versions of Windows that are trivially exploitable today.
Your dad probably got lucky, and your router’s firewall probably did a lot of the heavy lifting. If you were to connect a win 2000/XP computer to the internet today without a firewall between, it would be compromised in minutes (there are loads of videos of people demoing this).
While I don’t have proof that 7 would be the same, I strongly suspect it would be the same. 10 will get there soon too. Firewalls will stop most of the low hanging fruit, but an application that bridges connections through the firewall are that much more vulnerable to exploitations that won’t be integrated by your running kernel.
They won’t brick it, but you can bet that a lot of people are sitting on unreleased 0-days for win10. It will likely be dangerous to connect to the internet on day 1.
It’s almost certainly related to cloud-init, (the canonical tool for handling deployment automation) or Ubuntu pro (extra long support for backporting security packages to older distros, plus some conveniences). They’re pre installed as a convenience to paid users of those services, that’s the (IMHO, quite reasonable) model they use to fund the distro. I would expect that some or all of that traffic would disappear if you disable/remove those two services.
https://cloud-init.io/
https://ubuntu.com/pro