Post #3161069
2025-12-07 13:46 UTC
@micahilbery@mstdn.social It's not even misinformation per se if we define it as "intentionally wrong information to harm or manipulate". Especially in software engineering, vibe coding tools like #GitHub's #copilot are fed low quality code from sites like #StackOverflow not because people were poisoning the well since 2008 but because coding is somewhat hard, so it needs time to master. And since AI isn't *understanding* things but just combining stuff it finds, it struggles so tremendously with it.
Replies (1)
-
@shaedrich@mastodon.online 2025-12-07 13:50
@micahilbery@mstdn.social Same goes with various things: #Google's AI suggesting eating a stone a day and putting glue on pizza so everything sticks together (I'm sure, some AI might already have suggested that eating fly agaric is perfectly fine), etc. This is because just stringing words together based on their probability can't solve this. If fly agaric is used more in jokes than in biology books, AI will get it wrong.