logoalt Hacker News

ameliustoday at 9:33 AM1 replyview on HN

How do you trust what the LLM was trained on?


Replies

bravetravelertoday at 10:23 AM

Do I? Well, verification helps. I said 'prefer', nothing more/less.

If you must know, I don't trust this stuff. Not even on my main system/network; it's isolated in every way I can manage because trust is low. Not even for malice, necessarily. Just another manifestation of moving fast/breaking things.

To your point, I expect a certain amount of bias and XY problems from these things. Either from my input, the model provider, or the material they're ultimately regurgitating. Trust? Hah!

show 1 reply