AI Lies, so WHY should we continue using it?

Yes, AI lies! It lies A LOT! It halucinates. It halucinates A LOT! And if you're asking it questions directly, it's a crap shoot if you get an honest or factual answer out of it. But WHY would we bother using it? To be clear, I'm talking about LLMs (Large Language Models). ### Here's why they lie: They don't know they're lying. They're text generation machines and they generate text based on probabilities of prior text (like your question) combined with billions of things they "know". If they don't know the answer, they have no choice to do anything other than produce probable text. It's how they're designed. ### Here's how to overcome that: 1. Do NOT ask LLMs questions from their own knowledge. It's out of date. It's incomplete. And lots of what it "knows" is imply not so. 2. Ask it questions based on documents you provide it or from web searches. It's not good at providing direct answers. It's good at extracting data from unstructured text and restructing it in a form that's bettter for your needs. When you do THAT, it will (usually) do a very good job. It's not perfect on that either, but it's orders of magnitude better than asking it directly without referencing documents. Unlike other tips, AI was not involved in writing this tip in any way.