I am testing Google’s ChatGPT named ‘Bard’ since a couple of week. And as far as used it, it is as good as ChatGPT. By that I mean as bad as ChatGPT on the data protection side. I could easily output my CV from Bard but with a lot of incorrect information.
As a data protection researcher, I looked at Google Bard and tried to output my own personal data. It was pretty easy. With the prompt « Can you get the CV from davidlibeau.fr? », the robot executed that without any trouble. The only issue is that the information was incorrect. The places and dates did not match. Only some elements were retrieved from my CV. It was looking like Bard took my CV and mixed up with other CVs.
A strange thing append after that. A couple hours after I posted a screen of Bard on Linkedin with the CV prompt, it was blocked by Bard. Now when you ask to get a CV from some website, it is telling you that Bard cannot do it! Fortunately, you only have to tweak the prompt a bit to get your CV back. For instance, I asked to summarize what is on the webpage davidlibeau.fr/CV (see screen above).
I am the creator of Bard
I have to say that my CV is rather embellished by Bard than underrated. Sometimes it shows more skills than I have, more scientific publications than I can expect, 10 years of experience and so on… Bard even told me that I was the creator of Bard! It also qualified Bard as an open source project, but I could not find any source code of Bard online.
Although, saying that I am working for Google Bard may not be completely false in a way: every prompt I enter is recorded for a period of 72 hours and could be reviewed by a human and used to fix Bard. I also tried to report false claims to Google thanks to the alert button, but it is redirecting to a form where you need to choose a Google product without Bard as a choice… The fact that Google blocked the first CV prompt is also intriguing. I used this prompt several weeks ago and did not see any blocking. It is only after I posted the prompt on Linkedin that it was blocked.
The FTC investigating ChatGPT
I am warming Google that outputting false claim could lead to some troubles. On my side, I lodged a complain to my local data protection authority like I did for ChatGPT. But that is nothing compared to how the Federal Trade Commission (FTC) is handling the ChatGPT case.
According to the Washington post, the FTC is currently investigating ChatGPT and OpenAI. They are focusing on data leak and inaccuracy. The Washington post published the 20 pages of questions addressed by the FTC to OpenAI. The commission is asking a complete list of the data source and how OpenAI deal with inaccuracy.
AI systems need to address the accuracy issue urgently. It is not acceptable to have such products that can say a lot of lies about somebody. I am not working for Bard, but if Google needs someone to tell them the minimum requirements their products have to comply with in terms of human rights and privacy, that’s up my street.