Comments by "Brent Jacobs" (@br3nto) on "Kyle Hill" channel.

Are you serious? This video is basically just fear mongering. He says “we” don’t understand how these things work, which is the opposite of the truth; the AI engineers know exactly how they work. The so called “hallucinations” are an obvious artefact of how these models work. These models are currently static and deterministic, yet many the YouTubers are making out they think and contemplate. It’s so frustrating because the public discourse is around fiction rather the reality being led by people who should know better. Kyle needs to step it up.
5
@StoutProper your link to 8:35 is exactly proof of several of my points. A built model is static and therefore cannot adapt (at least that’s how AI currently works). The original Go AI model does not adapt to new adversaries because it’s model is static. Also, it should be obvious and not a surprise that the model doesn’t understand groups because it’s model is presumably just a relationship between different board configurations and transitions. Just like how language models are trained on relationships between configurations of words, rather than meanings of words, so too it should be no surprise about hallucinations and garbage responses.
4
Sounds like hitler (or any fascist leader)... the argument is we help them come to power to avoid risks of his wrath... I think the logic is wrong
1
When you say “we” don’t understand how these AI models work, you should clarify that the “we” refers to the general population. The AI engineers know exactly how they work. When you understand how these work, the concept dubbed “hallucination” is an obvious state an AI may get in. Language models are basically just hyper dimensional graphs relating disparate words. If you ask it if it’s telling you the truth, it’ll spit out related words based on that graph to make a response. So, stupid in stupid out generally, but your question might just be at an odd intersection that you aren’t expecting.
1
@thealmightyaku-4153 They aren’t black boxes though. There are videos and uni courses showing exactly how they work and what the models are. The linkages can even be visualised. To say they are too complex for us to understand is an underestimate of how smart humans are. Anyone can understand them if you put in the effort to try. It’s not some voodoo magic like a lot of these non-AI “experts” and articles are telling you. All your comments seem to be based from a fear reaction from all the fearmongering rather than logic and the reality.
1