Specifically, they believed it could be dangerous as a tool for generating really good fake news and related things, not necessarily AGI.

You don't need AI to do that, NI does it well enough.

So what? Lots of news is fake already, written by journalists who already have an agenda. You don't need any AI breakthrough to have that problem.

It's great for marketing though. Having something so 'powerful' you can't release it to the public. Also ties in with the 'fake news' and Russian bot hysteria very well.

Second order thinking here, it was a great narrative for supporting the next AI-winter

Sometimes it’s better to be cautious and err on the side of not releasing something safe, than to err on the side if releasing something dangerous.

I think the danger of neural network bots overrunning social media with garbage is entirely valid.

It seems pretty clear to me by now that GPT-2 is not as dangerous as OpenAI thought (or claimed to think) it might be. (It may yet be as deadly as OpenAI believed it was, but we won't know until it's actually released). That's not even mentioning that it's entirely likely that it will become the first machine learning platform to win a Turing Award.

What do you think? Are you ready to get involved in the GPT-2 debate? Or do you already have your own AI platforms and don't like GPT-1? Or do you have questions and would like the community to shed light on them? Please join us in IRC at #gpt: http://gstreamer.freenode.net/?channel=gpt

Please refer to the #GPT4 thread at Wikipedia.

GPG key fingerprint: A58F D3A2 CAF7 2E0D D4A3 8CD1 C721 5FB8 2F8E 15C4

Links

I like the drift in GPT numbers across this message.

What's the point of having a debate? GPT is proprietary IP belonging to a private company. They can release it or not as they please, regardless of whether their stated reasons are valid.

I believe that sharemywin's post is satire. It is at least, intended to appear to be gpt-generated.

Did you really mean Turing Award?

Or do you mean "pass the turing test"?

The GP comment was generated from GPT-2.

Social media is already pretty full of garbage. What is the actual threat here?

The actual threat is the danger of neural network bots eventually overrunning social media with far more entertaining content.

humans already do that and even provide advertising data along with it

It was a good publicity stunt that gave them wide media coverage and a lot of attention.

No, they're just responsible scientists. That's all.