Twitter | Pretraživanje | |
OpenAI
We've fine-tuned GPT-2 using human feedback for tasks such as summarizing articles, matching the preferences of human labelers (if not always our own). We're hoping this brings safety methods closer to machines learning values by talking with humans.
We’ve fine-tuned the 774M parameter GPT-2 language model using human feedback for various tasks, successfully matching the preferences of the external human labelers, though those preferences did not...
OpenAI OpenAI @OpenAI
Reply Retweet Označi sa "sviđa mi se" More
𝔊𝔴𝔢𝔯𝔫 19. ruj
Odgovor korisniku/ci @OpenAI
𝘠𝘦𝘴𝘴𝘴𝘴𝘴𝘴. Excited to see if the code is usable for improving poetry generation.
Reply Retweet Označi sa "sviđa mi se"
Andre Infante 19. ruj
Odgovor korisniku/ci @gwern @OpenAI
Surely if you can do this, you can use another transformer as an adversarial module to clean up some of the artifacts (repetition) right? No reason these supervision signals have to come from a human.
Reply Retweet Označi sa "sviđa mi se"
Dean P 19. ruj
Odgovor korisniku/ci @OpenAI
Such a cool update to GPT-2. What I like about this is the qualitative examples showing improvement. So much better than “we’ve beat the SOTA by 0.01%”
Reply Retweet Označi sa "sviđa mi se"
rdwrt 20. ruj
Odgovor korisniku/ci @OpenAI @permutans
Just adding "safety" to this tweet and hope that's enough to fool everyone into assuming we're up to something good.
Reply Retweet Označi sa "sviđa mi se"
Louis Maddox 20. ruj
Odgovor korisniku/ci @rdwrt @OpenAI
Well it's in their paper too, they have been outspoken on the dangers of large language models able to imitate humans. 'Safety' is also RE: biases in the text gen. models following incompletely specified objective functions, here they phrase it in terms of rules, like "don't lie"
Reply Retweet Označi sa "sviđa mi se"
Cem Say 19. ruj
Odgovor korisniku/ci @OpenAI
Will you be sharing any examples of this phenomenon?
Reply Retweet Označi sa "sviđa mi se"
Tony Abram 19. ruj
Odgovor korisniku/ci @OpenAI @sama
Is there a fast way to try these advances online? Summarization seems to be pretty easy to test on a website
Reply Retweet Označi sa "sviđa mi se"
Gus the Winged Wild Dog 19. ruj
Odgovor korisniku/ci @OpenAI @AdamDanielKing
I can't wait for the day it can write me stories about whatever I ask it to.
Reply Retweet Označi sa "sviđa mi se"
sahil 20. ruj
Odgovor korisniku/ci @OpenAI @prakharaga19
😎😉
Reply Retweet Označi sa "sviđa mi se"
Josef Vojtek 19. ruj
Odgovor korisniku/ci @OpenAI
Who wrote this tweet? :)
Reply Retweet Označi sa "sviđa mi se"