I don't want AI labs to be the arbiters of model-generated speech, and this feels very close to lab censorship.
It's not all that far from the logic used to censor covid "misinformation", some of which was true.
OpenAI has banned multiple accounts that were using ChatGPT to post content here on X in attempts to influence the public against datacenter construction, and against tariffs on technology.
Users objected to OpenAI banning accounts using ChatGPT to oppose data center construction, calling the approach ineffective and a potential step toward broader censorship.
Questions this raises:
Do you then intend to modify your models to refuse to generate this kind of content?
In COVID, would you have shut down people who were documenting vaccine-related myocarditis risk because the health authorities (incorrectly) were saying it wasn't real?
I don't want AI labs to be the arbiters of model-generated speech, and this feels very close to lab censorship.
It's not all that far from the logic used to censor covid "misinformation", some of which was true.
Dealing with a supposed influence operation by banning them from using a single model provider is an ineffective method, and opens the door to pressure in the direction of other censorship in the future.
The speech-generation layer is not the right layer to deal with this.
I don't want AI labs to be the arbiters of model-generated speech, and this feels very close to lab censorship.
It's not all that far from the logic used to censor covid "misinformation", some of which was true.
OpenAI has banned multiple accounts that were using ChatGPT to post content here on X in attempts to influence the public against datacenter construction, and against tariffs on technology.