3 ms·
This is one of the more major risks of chatbot style generative AI. People are very open to well argued answers to problems even if the answer is wrong. I was r
by trescenzi 2y ago
This is one of the more major risks of chatbot style generative AI. People are very open to well argued answers to problems even if the answer is wrong. I was reading this article about how those with less educational attainment seem to use chatbots more[1].
On one hand this is amazing because it increases access to good enough writing and information. On the other hand it raises the chances that people, regardless of educational attainment, are convinced that falsehoods are true.
1: https://arstechnica.com/ai/2025/03/researchers-surprised-to-find-less-educated-areas-adopting-ai-writing-tools-faster/ https://arstechnica.com/ai/2025/03/researchers-surprised-to-...
- roenxi 2y agoIt raises an interesting theoretical problem. If the AI comes up with a good argument that translates a coded message into something legible can it truly be said to be wrong? It'd be more of a flaw in the art installation. The problem with asking people to interpret gobbledegook is that it might map to English in any number of ways.
- ethbr1 2y agoLLMs are literally truth-sounding machines. Which isn't to diminish their utility, but is to say it's smart to treat them like used car salesmen. Maybe they're telling a true-sounding truth... or maybe it's just true-sounding.
- trescenzi 2y agoRight their purpose is to produce bullshit in the most technical sense[1]. There is no concern for the actual truth, just that the output sounds good. Edit: That’s not to discount the value of said bullshit or to say they are never correct. Simply that correctness isn’t what they are optimized for. 1: https://en.wikipedia.org/wiki/On_Bullshit https://en.wikipedia.org/wiki/On_Bullshit
- timbit42 2y agoThe need to add an inference engine to them for fact checking.