Post #3161072
2025-12-07 13:58 UTC
@micahilbery@mstdn.social Even if we said, we only feed the models high quality information, we still have two huge problems:
– we either have to code "conditional exception branches" into the AI, to branch out of it automatisms and let it follow a hard-coded path instead to arrive at a preferable solution: This would fundamentally go against the principles of AI working like a black box and making decisions without explicit code
1/
Replies (1)
-
@shaedrich@mastodon.online 2025-12-07 14:01
@micahilbery@mstdn.social – or we can curate the training data, which would be a massive undertaking – data enrichment workers would have to do even more with the data than they already do today (not to mention their often precarious work situation) – remember what I said about humor? We would have to apply this to truth. While we agree on 1+1=2, some things are more nuanced, biased or even philosophical. These would not easy to even define 2/2