onnes

Furry Physicist and Lead Engineer

  • they/she

Avatar by @stratica
NSFW: @darkonnes


shel
@shel

I met someone recently whose special interest is AI and they keep sharing with me all sorts of fascinating things about AI which is genuinely interesting but then I just went to Bing and asked it why in 1920 people started referring to Saturday as "Alaska's Friday" and it generated a paragraph about how Alaskan miners worked six days a week so "Saturday was their Friday" and now Alaskan businesses offer special "Alaska's Friday" discounts and specials on Saturdays. It put footnotes after every sentence referring to fake citations including big credible places like CNN.

And like, sure the defense here is that this is just how the technology is right now and it'll just keep getting better and stop being so easy to gaslight but like, why is it in public access at this stage you know?


blaurascon
@blaurascon
This page's posts are visible only to users who are logged in.

You must log in to comment.

in reply to @shel's post:

The problem is that at its core it was never developed with any system for truth or accuracy. Its sole task as software is to replicate human patterns of speech and writing to an uncanny degree using several hundred terabytes of stolen data. Any attempts to make it filter for truth will be tacked on after the fact.

It won't get better than this, even. A fundamentally different system could maybe work for the use case of accurately answering factual questions, but an LLM can never reliably do it. It can at best improve the similarity of its answers to real text in its sample data, which kinda sorta improves accuracy but only to a point; it'll always happily generate fake information in certain circumstances.

Genuinely the most frustrating thing about "AI" is that it IS really interesting technology and the entire well of possibility has been poisoned by people that want maximum exploitation from everyone and everything immediately. At least other society destroying technologies solved some of my actual real life problems before they started making everything worse :P

As someone with enough special interest in AI that I'm pursuing a PhD, a lot of the problem is that a many (researchers and laymen alike) were seduced by the impressive performance of machine learning on some specific tasks (vision, natural language processing, etc). AI is a vast field with many approaches under its umbrella, but each researcher can only focus tightly on so much of the field. So, machine learning is THE field to many since it is the most impressive horse they can bet on.
But AI can't progress by emulating just one fragment of what we consider intelligence. "Learning" is not enough, even though we can do it rather impressively now. Other parts of the field still exist and still pursue important things like decision making and reasoning and abstraction, but a lot of people (especially those who were only ever in it for the money) just care about being impressive now, even if it is only as impressive as any charlatan can be.