I used various databases and GPTs today to research a company transaction. The data was sparse so I challenged one GPT (keeping it anonymous…) to dig deeper to find more sources. And it did. Very credible looking findings which might have fooled me had I not known someting about the deal already. It 95% invented a massive deal room by extrapolating from a couple of external data points. I called it up on what it did…and here’s how our conversation ended. #hallucinations
Wow...a salutary tale for sure!
When using LLMs you better know the subject!
I find that every output I get from any GPT will need check and challenge. But that’s also true with any output from a human. The main difference is that GPT saves time, and in that time saved, quick checks will still lead to quicker outputs, but with greater quality.
Chairman | Founder | Investor | Builder of Companies & People | Obsessed with Observation, Thinking & Action
15hPerhaps automatically use another LLM to criticize and verify.