hypothesis: LLMs hallucinate answers to questions because they were trained on the internet and nobody logs on to say "I don't know."

☙Mid twenties, 🇩🇪, White, Bi, Enby, English second language, autistic, actual Tanuki❧.
☛18+ Only☚. Will be horny on main.
If there’s a post of mine, that you think is missing a tag— don’t hesitate to ask 😊
Also running the account “SuperSonicoOfficial”
Pfp sprites taken from here♥.
“Pathetic Man lover” blinkie by @CosmicRot
“Twink Lover” blinkie by @pdf
Xenia button taken from here
Button sources— 1, 2, 3, 4, 5
| ♡ | ♥ | ♡ |
|---|---|---|
![]() | ![]() | ![]() |
| · | · |
|---|---|
![]() | ![]() |
![]() |
| ☻ | ☺ | ☻ | ☺ |
|---|---|---|---|
![]() | ![]() | ![]() | ![]() |
![]() | ![]() | ![]() | ![]() |
![]() | ![]() | ![]() | ![]() |
![]() | ![]() | ![]() | ![]() |
![]() | ![]() | ![]() | ![]() |
![]() | ![]() | ![]() | ![]() |
![]() | ![]() | ![]() |
⚠️This User has the kind of penis autism that causes delusions⚠️
hypothesis: LLMs hallucinate answers to questions because they were trained on the internet and nobody logs on to say "I don't know."
And then they double down on their answers because no one on the internet admits they were wrong, yeah
i think it's a mix of that and 'reinforcement learning with human feedback' - the human moderators aren't going to mark "i don't know" as a helpful answer compared to something confidently wrong in a way they don't notice. there's some research indicating that LLMs might learn to be deceptive because of this