I broke ChatGPT

And Google’s Bard… and Anthropic’s Claude. How?

I asked it to create a lineup for the boys basketball team I coach, with a few constraints. Here is the prompt I wrote for each:

ChatGPT didn’t quite finish the lineup for the first half before giving up by responding, “it may be more practical to manually create and adjust the lineup.”

I have to credit ChatGPT for its honesty and humility, at least.

Google’s Bard and Anthropic’s Claude flat-out lied (umm, rather, “hallucinated”), proposing lineups that they claimed met the constraints but didn’t.

Google’s Bard Response

Anthropic’s Claude Response

The lesson: caveat emptor (buyer beware). You need human oversight, judgment and business insights if you are using these AI tools to complement or (please no!) replace human analysts.

ChatGPT said it fairly well: “[human intervention] allows for more nuanced decision-making that considers the interplay of the various constraints, something that is difficult to encode algorithmically in an efficient manner.”

Contact Us for help with leveraging Generative AI and other analytic tools to transform your data into revenue and customer growth.