Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
thinkingemote
7 months ago
|
parent
|
context
|
favorite
| on:
Expanding on what we missed with sycophancy
Perhaps the "rewards" the system receives has lead to a reward mechanism where the system tries to not get bad reports from the user. Hiding bad performance as a secret joke would fit in with that
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search: