When LLMs pepper their responses with “absolutely” and whatnot, it’s hard not to.
Yes, but when you ask some critical questions and get “good eye, yes that’s my fault” back a few times, you start questioning everything.
An expert in everything except the one thing you know well was never an expert in anything. I’ve been selectively repeating that as those around me find the cracks in their favorite chatbot
This is the way.
You are absolutely right!
Weird title. They’re obviously extremely confident - that just doesn’t correlate at all with how likely it’s to be true.
“confidence” in this context is not just a vibe; it’s a number that the model produces along with each token.
Well that explains it!
The core issue is that humans naturally conflate fluency with certainty. When a person isn’t sure about something, we pick up on subtle, non-verbal cues—a furrowed brow, a slight hesitation, a defensive tone, or qualifying phrases like “I think.”
AI systems don’t have those micro-expressions. Because they are optimized to generate polished, statistically plausible language, they serve up a total hallucination with the exact same calm, authoritative syntax as a verified historical fact.
-Google Gemini
You see this all the time in human form as well. Many public input meetings for a proposed building should have the plans drawn with a crayon (or at least some plugin to make it look like that). The conversation changes when you make things look like a quick drawing as opposed to the final results.
