Re:It sounds like Google will unveil its ChatGPT clone Febru...
by amanaplanacanal on Saturday February 04, 2023 @11:59AM
(#34655918)
Won’t there always be ideological weights though, depending on what you select for the training data?
[ Parent
| on HN ]
Re:It sounds like Google will unveil its ChatGPT clone Febru...
by thatguy0900 on Saturday February 04, 2023 @12:09PM
(#34656013)
Yeah, certainly. I should have specified the manual weighting they do. I wouldnt mind a bot that was trained on training data to lean left or right, but the bot just spitting out the same boilerplate paragraphs of text it was told to say anytime something political comes up gets old fast.
Re:It sounds like Google will unveil its ChatGPT clone Febru...
by hbfijcgu on Saturday February 04, 2023 @12:50PM
(#34656413)
I think that, like with newspapers there is unavoidably a bias. Hopefully there will be a big cast of AIs with different ones.
Re:It sounds like Google will unveil its ChatGPT clone Febru...
by great_psy on Saturday February 04, 2023 @01:05PM
(#34656537)
I am ok with bias in the data. That bias can often be overridden with explicit prompts (used to be able in early versions of chat GPT).
The problem is explicitly censoring certain topics and forcing a bias that cannot be overridden by explicit prompts.
I understand the concerns about generating harmful content. But the cat is out of the bag now. We can slow it down, but we cannot put it back in. There will be models that have no censorship, it’s just a matter of time