5·
4 hours agoIt’s always projection with them isn’t it?
- 5 Posts
- 625 Comments
Joined 3 years ago
Cake day: July 1st, 2023
You are not logged in. If you use a Fediverse account that is able to follow users, you can follow this user.
We have ours configured pretty well at this point and it does an absolute bang-up job at code review. It will point developers towards tools and utilities that our monorepo has If they appear to be reinventing the wheel. It will catch common problems and issues that we have in our instruction file, and does a pretty good job of catching bugs and small issues like typos and what not.
Perhaps you need to tighten up your instruction file?
2·6 days agoSo essentially these are just more dangerous pump handles then?
Love it.
How would they though?
They cannot learn and do not have memory. Which means they cannot actually follow a “decision”, and remember that an action has been taken. All information is limited to the context window, which is only an illusion of memory. Not actually memory.
They are effectively RNG’ing incredibly capable word generation machines.
They don’t have short memory, they have NO MEMORY AT ALL.
These are statistical word generation machines, that’s what LLMs are right now. They are REALLY good at this.
But they do not have memory, they do not learn, they do not make decisions. Which means they are incapable of cooperation as such a thing cannot exist without memory or the ability to learn, and decisions cannot be made without either of those.
These tools provide the illusion of such attributes.
- There is no memory, the whole context is sent on every request, the LLM does not have knowledge of prior conversations. It only knows what it is provided in that request only.
- Lots of tricks and hacks to make this illusion really good in incredibly small scales. But it’s still an illusion. Outside of fine-tuning and retraining new LLMs, which is not feasible to do on a frequency of communication.
- There is no learning.
- Without memory learning is impossible. Learning requires retraining a model, and to a degree fine tuning. Both of these are resource intensive and are static. And only provide the illusion of learning as it cannot happen in real time.
- There is no memory, the whole context is sent on every request, the LLM does not have knowledge of prior conversations. It only knows what it is provided in that request only.
Probably GitHub Copilot.
The rest just sucks.
2·15 days agoYou can’t fix stupid, it will only drag you down with it.
It is, just not for the pleb class.
4·18 days agoThe United States can’t manufacture shit so yeah, Canada should step up and prepare.
I assume that the gitea instance itself was being hit directly, which would make sense. It has a whole rendering stack that has to reach out to a database, get data, render the actual webpage through a template…etc
It’s a massive amount of work compared to serving up static files from say Nginx or Caddy. You can stick one of these in front of your servers, and cache http responses (to some degree anyways, that depends on gitea)
Benchmarks like this show what kind of throughput you can expect on say a 4 core VM just serving up cached files: https://blog.tjll.net/reverse-proxy-hot-dog-eating-contest-caddy-vs-nginx/#10-000-clients
90-400MB/s derived from the stats here on 4 cores. Enough to saturate a 3Gb/s connection. And caching intentionally polluted sites is crazy easy since you don’t care if it’s stale or not. Put a cloudflair cache on front of it and even easier.
You could dedicate an old Ryzen CPU (Say a 2700x) box to a proxy, and another RAM heavy device for the servers, and saturate 6Gb/s with thousands and thousands of various software instances that feed polluted data.
Hell, if someone made it a deployable utility… Oof just have self hosters dedicate a VM to shitting on LLM crawlers, make it a party.
11·23 days agoEven more reason why it should be released as a dump, and not walled in behind a website.
This is assuming aggressively cached, yes.
Also “Just text files” is what every website is sans media. And you can still, EASILY get 10+ MB pages this way between HTML, CSS, JS, and JSON. Which are all text files.
A gitea repo page for example is 400-500KB transferred (1.5-2.5MB decompressed) of almost all text.
A file page is heavier, coming in around 800-1000KB (Additional JS and CSS)
If you have a repo with 150 files, and the scraper isn’t caching assets (many don’t) then you just served up 135MB of HTMl/CSS/JS alongside the actual repository assets.
Fair fair. I missed that
I can get a 50Gb/s residential link where I am, and have a whole rack of servers.
Sounds like a good opportunity to crowd fund thousands and thousands of common scrapeable instances that have random poisoning.
Low key win for kink communities.
Yeah but that was before you had billionaires of this size able to manipulate entire markets in this capacity.
Why do they make this so difficult by not providing a raw dump…
It’s not just license plate readers anymore. They have cameras that perform facial recognition and other identifying recognition.
Your car is in many ways uniquely identifiable by its markings and its model that vehicle with many pictures of it and that license plate are already in a database. If you have stickers, if you have big dents or additions and changes from the base model of your vehicle than you are quite identifiable within a particular geographical area depending on the urban density.