January AI Reality Check: AI Chatbots aren’t Google

This sort of misconception can lead to a few problems, the first one being that AI is just finding the right answer from a list. The second is that a short prompt is good enough, because it’s just matching keywords and “finding” the response.

The truth is that AI models that power chatbots are trained on enormous amounts of information, but they don’t store that information as full sentences, any more than the human mind does. AI learns from that training, and when prompted, it uses what it learns to create a new response from scratch, again like people do. It’s worth noting that unlike a human, AI doesn’t learn from your questions and gets better at answering them, but the basic answering is the same.

Prompts that have more detail are better because they give the system more to match against in their memory when creating that response, so you are much more likely to get a more accurate, or more useful answer.

← Back to AI Reality Checks