3 ms·
Something that has been bothering me for a while is whether poisoning of OpenAI's dataset is possible, willingly or otherwise. An example here is getting chatG
by PartiallyTyped 4y ago
Something that has been bothering me for a while is whether poisoning of OpenAI's dataset is possible, willingly or otherwise.
An example here is getting chatGPT to accept that that 2+2=5, it's a lot of effort, but can be done. Then the users can give thumbs up when such responses are given.
Could this cause issues?