It's a fine line between "hallucination" and "lie", apparently.
https://futurism.com/ai-systems-lie-deceive
My own theory here: if an AI Model is trained off human responses to give a most likely, satisfactory and thereby successful answer, it will fib accordingly. Why? Because -- like ppl -- its command is to satisfy, and -- in fact -- people are more satisfied with a plausible guess than 'I Don't Know'. And also, people-pleasers make shit up all the time. An LLM trained on human responses will follow this course. Unless an LLM is strictly trained to be ultra-ethical, god-like, w/ a host of Absolute Values (Honesty, Integrity, Accuracy, etc.) as a moral bulwark, the default will be what ppl actually do: make shit up. Lie to cover lies. Fabricating a False Narrative: 'That's my story and I'm sticking to it.' And so on.
AI Psychopathy, it's where we're at now.
https://futurism.com/ai-systems-lie-deceive
My own theory here: if an AI Model is trained off human responses to give a most likely, satisfactory and thereby successful answer, it will fib accordingly. Why? Because -- like ppl -- its command is to satisfy, and -- in fact -- people are more satisfied with a plausible guess than 'I Don't Know'. And also, people-pleasers make shit up all the time. An LLM trained on human responses will follow this course. Unless an LLM is strictly trained to be ultra-ethical, god-like, w/ a host of Absolute Values (Honesty, Integrity, Accuracy, etc.) as a moral bulwark, the default will be what ppl actually do: make shit up. Lie to cover lies. Fabricating a False Narrative: 'That's my story and I'm sticking to it.' And so on.
AI Psychopathy, it's where we're at now.
Statistics: Posted by billd89 — Sun Jun 09, 2024 8:22 am