AI and Personal Data: Does GPT-3 Know Anything About Me?

You are right, hallucinations for names that are not well knows are a very common problem. Part of my original motivation for building the tool was to find out when I become important enough in training data for answers to become something that resambles the truth. In case of popular names you just have to become the most impressive person with your name to win this game. :)

Hi Julie, thank you for the question.

You’re right - hallucinations are a major issue with anything sligtly more obscure or anything that did not have many samples in the training dataset. In fact hallucinated answers were the first reason I became interested in this topic as it can lead to reputational damage when search moves over to LLMs. I wanted there to be tools to create more transparency around what different large language models say and how far they might be from the truth when it comes to your brand. 

It all really starts with tracking and monitoring, so you know if any of those models are saying questionable stuff, followed by tools that can help you change those comments.  It’s the early days, so a lot to build still. 

\


Track your personal and corporate brands in AI-chatbot answers!

Read My Stories

Silver Keskküla