Since hiring Claude to help me do research, I feel two things. First if Claude were human, I would have fired him after his first big hallucinatory botch it was so egregious. Second, it’s now kind of entertaining to throw Claude a simple task and see how much of it he gets wrong. Claude, of course, is what Anthropic has called its artificial intelligence software. If I were naming this software I’d call it BDT for bullshit detector training software. I described my initial encounter with his fails here.

Today I asked Claude to give me a list of publications or podcasts that might appeal to National Guard family members. The top of the list Claude kicked back was:

Guard Family – A magazine specifically designed for National Guard families, covering deployment support, benefits, family programs, and resources

I went searching for this publication and couldn’t find it, so I asked Claude: “Can you give your source for Guard Family? I’m not finding it online. Is there a possibility you hallucinated this magazine?”

Here’s Claude’s answer: You’re absolutely right to call that out—I should search to verify this. Let me check what actually exists.Good catch—I did make an error. There isn’t a magazine called “Guard Family.”

Then I said: Is there some way I can tell you in advance to do actual research instead of making things up?

Then Claude said: That’s a really fair question, and I appreciate your directness. Honestly, the answer is: not really in a way that would guarantee I won’t make mistakes like that. You can ask me to search or verify things, and I should be doing that proactively for factual claims (especially specific publication names), but there’s no setting or instruction you can give me that will completely prevent me from occasionally stating something confidently that turns out to be wrong.

Claude: verbose and unhelpful. I find this so reasuring. I perceive that AI will get drafted to fill jobs and people will get fired. Then, a couple weeks in, crying managers will be calling their fired employees and begging them to come back.

Posted in

3 responses to “Claude. He’ll help you hone your bull-shit detector skills”

  1. daelparsons Avatar
    daelparsons

    Yes, agreed. I was being facetious, but I do cringe when I hear what other sources are putting out and saying is factual and true.

  2. daelparsons Avatar
    daelparsons

    i think there is evidence to support the belief that Claude is ALREADY in charge. Thank you for sharing your experiences with AI. Fascinating! What happens when we question the validity of nearly everything we read or see in print?

    1. supalmer Avatar

      Well, that would explain a lot… But to be serious, what I do is choose news sources I trust and stick with them. If it’s not in our local news providers (Register-Guard, Lookout, Eugene Weekly, KLCC) and it’s not in the big two I subscribe to (New York Times, Wall Street Journal), I’m not seeing it. I miss things. I’m OK with that. When trusted journalists get things wrong, they write corrections.

Leave a reply to daelparsons Cancel reply