Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
beatthatflight
49 days ago
|
parent
|
context
|
favorite
| on:
"ChatGPT is I presume broken"
say hi first, then the prompt. seems to work better.
ksaj
41 days ago
[–]
Confirmation bias. Simply repeating does the trick. The interesting part isn't that this error comes up. It's that it very often decides "weird" means "pornographic" and starts to generate such an image before catching itself.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search: