evs91

so you re-invented kubernetes but in a unicorn type way is how I read what you designed...

and by this I mean: that time that the scrapers ingested too much trash / slop to be accounted for and thus became "mediocre" because all that LLMs do is make output that is the sum "average" of humanity's knowledge.

you mean that time that the scrapers started becoming the Ouroboros? When you put in trash don't expect gold output

lol....hardly groundbreaking stuff my guy. and also: "precompiled powershell" doesn't exist. just lame slop

consumer EDR "could" detect it if everyone knew what to look for and the pricing was good. Unfortunately (or not) EDR for consumers is limited to really just the MS365 addon for Microsoft Defender for Endpoint (P2) which is $3 a month on top of your MS365 license (so looking at a good value if you already have an enterprise tenant even if solo). Downside: it's a firehose of information and is a full-time job managing for SMB. But to the other comment here: sandboxing / runtime isolation helps. It's more an onion than a strict wall. One failure shouldn't cause the city to collapse.

yeah; no way with the data currently collected and lack of any audit certifications would I ever allow this on enterprise tenants. lmao.

not even sure what is going on here....

You are pretty much going to experience that no matter what. Even if you update UFW to only allow Cloudflare IP ranges, you will get scans against SSH and if you turn on deny logging you will see that your IP still is being scanned. Changing the IP just moves your target somewhere else for it to be indexed again. Fail2Ban rules, like you mentioned, will reduce your attack surface and get the "background" automated attacks somewhat at bay. You can do things like only allowing SNI HTTPS requests and not direct IP connections (which is what you doing with the Cloudflare proxy). From what you are saying: you are doing a solid middle of the road start. I would focus on making sure you keep the security posture up as you implement other services. Is this just for your projects or are you providing a service to customers? If it's just you - does the effort merit the work. If its customers, there are more things you can do but "do you need to" is going to be more up to you.

TL;DR - sounds like a solid starting place; don't worry about the IP address

yeah; and then there is the ongoing scandal still of the Legionaries of Christ. Been there - done that. I'm still always amazed at how Opus Dei has avoided any long term bad press unlike the LC's which seem to only come in waves.

So creating synthetic knowledge for AI and thus contributing more to the issue by using the tools causing the issue for the issue to just index again?

not available at link as of 21:24 UTC

Lol - so way to not actually say anything over the dozen docs I read. It’s a bunch of beating around the bush, vague metrics, and not a single: how does this actually change outputs. The outputs of the different models “internalizing” could just be induced by previous prompting.

So I'm going to the website and all it does is ask me to sign up; not sure this is much better as there is zero information on how the system works.

Cerebras alone for the speed of the response - it was quite jarring the first time to see a 70B model respond near instantly.