Model over-confidence perception is a Design problem. The linguistic style of LLMs is parked in overconfident persuasion.
Both the pro and the anti come wrapped in overconfidence. The model is confident that the pro answer is grounded in reality and is also confident that the caveat hits reality.
This style is like that friend we all have that "knows-it-all" and commits to nothing. We know we can't really trust a word they say.
Trust.
Design units in frontier labs are absolutely confused right now, it seems.
When a design team is working under your watch on a chat interface, and they think their job is smoothing out buttons for the user to RAG files... Well, they are absolutely confused (and plz stop spreading more of these buttons).
The job should be: the Design of language.
Get out of Figma and Claude Design, stop reviewing components, and go study Tolkien, Chomsky, Pinker, Vignelli, Eames. Eames. And, have I said Eames?
Go understand the brilliance of what the Eames did with IBM in 1964 to explain mathematics to the masses.
Then look at the problem again. You're welcome.
That's the job.