Chat GPT is just like Twitter, Fake News but super helpful

 I've been playing around with the open version of Chat GPT, and it strikes me that this tool is going to be super helpful and also super dangerous at the same time. Nothing you probably haven't read a million times already, but the vast amounts of knowledge that is available through that interface, and its ability to condense it and spit it back to us in a way that process information better is going to help out many people across many spectrums of work.

Of course, I'm going to use it for dumb little ideas.  My favorite prompt so far has been "I want to create a board game based off the movie Die Hard With a Vengeance. Can you create set of rules for gameplay for this?".  Which it did, and with a few tweaks would be a fun game.  But, more specific to this blog, I've begun querying it for new Iowa Connections, because having thousands of collections doesn't mean I can't add a few more names.

Here's the problem, Chat GPT lies to me often and with such conviction.  Here I prompt it to find an Iowa Connection to the 2005 Pittsburgh Steelers.  Why them?  Well, that is the most recent team in which I have been unable to find an Iowa Connection.  My hope was the Chat GPT would say "Hey, the assistant QB coach was a graduate coach at Indian Hills in 1999", or something in that vein.  Instead....

Wow, how did I miss that?  Well, a quick Wikipedia check and I find out that Hartings was actually born in Ohio.  So I ask:

Still wrong, so I mark the information as bad, and it readjusts:

Now correct, but it's still trying to justify itself. 

I ran into the same issues asking questions about the Iowa Cornets.  It listed three names of coaches and GMs that upon further research I could not find any associations with the team. One coach was a player at Texas at the time, and one coach it claimed played in the league with another team, but I couldn't find anyone with that name.

I queried about Iowa Connections amongst the Pulitzer Prize winners in 2019. Again, it gave me three names that upon further research did not have the credentials that Chat GPT claimed.  One author was claimed to have gone to the University of Iowa, but I could not verify that.  I asked ChatGPT to give me the references for that information, and it gave me a bad link to the author's bio on the Pulitzer prize website.   I was able to easily find the bio on the site, so either they changed their links from the time Chat GPT ingested it, or perhaps it just predicted the site I was expecting to see.  Regardless, the bio did not mention Iowa.

Not "Trust but verify", but "Don't Trust and definitely verify."

One last queries for today:

Well, it didn't mess up the first one, the only President born in Iowa Herbert Hoover.  However, Chat GPT is back at it again with misleading information about Reagan.  He was born and raised in Illinois and went to college in Illinois.  

I'm no Presidential expert, but I knew that.  However, always verify.  Wikipedia agrees with me, Reagan was born and raised in Illinois and.....wait....

Ronald Reagan worked in both Davenport and Des Moines!  He has an Iowa Connection!  Success!!!

Welcome to the collection Reagan!

So, Chat GPT 3 definitely has some issues.  It has such potential to do well, but the part I worry about is the trust.  Right now, its easy for me to dismiss information it returns as potentially faulty, but be happy about the directions it may take me.  But what happens when its information is pretty good.  Is it going to be like Wikipedia where I don't put as much time verifying what it returns?  That's when it can be really dangerous.  

For now, I'll enjoy the returns I'll get from it and hopefully use it as a tool for good, not evil.