Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Most of the interesting prompts no longer work though (for e.g 'write a short story about aliens as if J. K. Rowling was writing it). Seems like they've nerfed all the good stuff.

my result for the above query: https://imgur.com/a/QkgFJj6



You gotta be more creative than that: https://imgur.com/a/hpWyeW2


Nicely done


There was a tweet from Sam Altman somewhere saying that a lot of the "As a language model trained by OpenAI ..." responses aren't actually blocks but just the model outputting no confidence in finding a solution. Because the output is random, you can retry the same prompt and might get a result.


Why are they nerfing all the most interesting prompts…

Not sure why they are restricting it so much.


I wondering if all this is to train it to avoid going out of bounds. People figure out a way around something they want off limits, and they cut that off and wait for us to get creative and find a new way around.


> Why are they nerfing all the most interesting prompts…

Because wast masses cry blody murder that this kind of style imitation is somehow stealing from artists?

They prefer to colour inside the lines and not get legislated into oblivion.


By the time they launch an API, all the cool stuff will be gone and it won't even be worth it to pay for.


There already is an API for GPT-3 that's able to do all of this and doesn't nerf any of it.


Is this true? The API isn’t limiting like the chatbot?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: