zlacker

[parent] [thread] 2 comments
1. ben_w+(OP)[view] [source] 2023-03-01 12:59:05
That's like saying:

> Ford straight up proved that Béla Barényi (of Mercedes Benz) ideas around crumple zones is all a big waste of time. The world didn't end with the 1938 Ford Prefect[0].

The world won't end overnight with an open fork of ChatGPT.

But it will mean the signal-to-noise ratio rapidly shifts, that spammers and scammers will be much more effective, and that even minor special interest groups (or individuals) get the ability to cheaply fake a diverse crowd of people to support any cause at a slightly higher standard of discourse than the current waterline for random internet comments.

[0] I don't know for certain it didn't have a crumple zone, but given when the patent was granted to Mercedes Benz…

replies(1): >>whywhy+V3
2. whywhy+V3[view] [source] 2023-03-01 13:25:56
>>ben_w+(OP)
That car analogy doesn't work because Mercedes Benz didn't actively work against other cars from existing.

https://www.vice.com/en/article/dy7nby/researchers-think-ai-...

replies(1): >>ben_w+2a
◧◩
3. ben_w+2a[view] [source] [discussion] 2023-03-01 14:03:06
>>whywhy+V3
First: Crumple zones have since become mandatory, which pattern matches what would happen if OpenAI gets what they ask for in that link.

Second: I'm just as concerned about automated generation of propaganda as they seem to be. Given what LLMs are currently capable of doing, a free cyber-Goebbels for every hate group is the default: the AI itself only cares about predicting the next token, not the impact of having done so.

Edit:

Also, the headline of the Vice story you linked to is misleading given the source document that the body linked to.

1. Of the 6 researchers listed as authors of that report, only 2 are from OpenAI

2. Reduced exports of chips from the USA are discussed only briefly within that report, as part of a broader comparison with all the other possible ways to mitigate the various risks

3. Limited chip exports does nothing to prevent domestic propaganda and research

https://cdn.openai.com/papers/forecasting-misuse.pdf

[go to top]