AI as a centralizing force

[dngray (Daniel Gray)]

Quoted from my other post.

@dngray

@Libre_Software_Enjoyer

AI is maybe the worst thing humanity has ever invented

I think a lot of these doomsayers saying AI is going to become the next Skynet are ridiculous. It’s like arguing a hammer wants to hit people. It’s just a tool. If AI is making certain industries redundant because it simply can do a better, more efficient and accurate job, then that is progression. Projects which claim “no AI contributions” for example, are not doing themselves any favors.

The real fact is, that the contributions should not be distinguishable from human contributions on the basis that they are accurate and correct. Some projects refer to this as thoughtful use of AI. AI is particularly good at developing test cases, eg telling you all the things you’re not checking. This is useful for any kind of regression testing.

As for “Ai being the replacement to search” - which was a quote from that video, that’s not necessarily a bad thing. Some topics are not controversial and frankly with the way search engines been lately because of all the SEO garbage, I find that saves me a lot of time. Then you have the bloggers which have all sorts of interests and aren’t even writing factual information.

The latest example I can think of was in regard to the arguments by a privacy blogger about Chrome including a weights file, and the particular blogger claiming that it was an environmental catastrophe (climate) based on preconceived usage of a data center’s power requirements, then pivoting and trying to claim that was a violation of GDPR. It literally amounted to an argument that local LLMs are worse for privacy, which makes no sense whatsoever in any context.

Being able to input a specific error produced by a program and have it tell me why that is and and what to do about it, seems like a improvement over randomly trying to find some internet post on a forum where someone has had the same issue, or issue in a bugtracker. It also means i don’t have to visit a bunch of websites with consent dark patterns and annoying things that might trip up an adblocker.

Then there are simply tasks a human cannot do efficiently, like analyze assembly or a binary like a fuzzy analysis on steroids. There are specific usecases like pgvector which use a small language model to improve database efficiency.

It comes back to the old “garbage in, garbage out” if you put good data in and a lot of it, you get excellent results.

Please also be mindful we do have rules against unsubstantiated content that is not based on fact.