,
5 tweets,
6 min read
Read on Twitter
@jackclarkSF @OpenAI @Miles_Brundage Preface: I love the team at @OpenAI, I hope that's clear and obvious.
My main issue: way more than once @OpenAI have been the source of AI hyped narratives, intentional or not. The responsibility falls on the AI community to try to hold off the avalanche of misinformation.
My main issue: way more than once @OpenAI have been the source of AI hyped narratives, intentional or not. The responsibility falls on the AI community to try to hold off the avalanche of misinformation.
@jackclarkSF @OpenAI @Miles_Brundage My feelings are almost exactly the same as two years ago from the @OpenAI DotA 1v1 narrative and I'm disappointed as I thought we'd moved beyond that.
@jackclarkSF @OpenAI @Miles_Brundage I was asked to comment on the research by journalists - but that was soon before release and all I got was the paper. I explained the technical details but those weren't important - it was the rest of the narrative. Specifically the narrative I wasn't aware of.
@jackclarkSF @OpenAI @Miles_Brundage I have no clue what @OpenAI's internal thinking is or what exactly they said or provided journalists. If I knew you weren't releasing the large model I would have pointed journalists to @NvidiaAI's work on training a language model on 40GB in 4 hours.
arxiv.org/abs/1808.01371
arxiv.org/abs/1808.01371
@jackclarkSF @OpenAI @Miles_Brundage @NvidiaAI How does the malicious/state actor discussion factor in there? I have no clue as I have zero additional information from @OpenAI. I didn't even know the number of machines used (only disclosed in an article by a journalist) or how long it's trained for (totally unknown still).
