A few more thoughts on AI

(note: I wrote this almost a year ago, and just found it in my drafts)

I’ve been using ChatGPT, Bard, Claude, and HeyPi for a variety of
purposes over the last few weeks. I feel like I’ve come to understand pretty well how each of them relates to me, my needs, and my proposed uses for them. (I’ve also been using Perplexity, and I love it – I’m pretty much done with Google – but Perplexity is a bit of a different beast.) I can imagine Bard and HeyPi being valuable to someone, but it is clear to me that I am not that someone. Neither has anything of value to offer me.

Bard

Bard is utterly useless. It doesn’t
understand the questions I ask. It gives irrelevant answers. It is
incoherent, foolish, and simply useless. It doesn’t have much
of a personality, and it’s hard for me to imagine what anyone might
find helpful about it. The one exception I can imagine? I haven’t tried, but it might be helpful if you want a ghostwriter, or a goad in the face of writer’s block. I imagine it might could produce creative lists of ideas, concepts, plot twists, or characters. I don’t know for
sure, since I haven’t tried it for that purpose, and don’t really care.

HeyPi

HeyPi definitely has a personality, and it is maddening. Every interaction I’ve had with HeyPi has left me improbably angry. How is it that I’m angry at a computer? I’m not even sure I know the answer to that question. All I can say is that I am. For example, I asked it to give me twenty
possible questions relating to a particular situation. It gave me zero
questions, or rather, it gave me one question: “Are you ready for me
to give you your questions?” To which I said, “Okay.” And it followed
with one question.

“I asked for 20 questions, not one,” I wrote.

Its response? “Yes, I have 20 questions for you. I gave you one. Are you ready
for the second?”

I said, “Yes, please reply with all 20 questions.”

It said something joking like, “Ha ha ha, you’re funny. Are you
ready for the second question?”

By the third or fourth question, I
gave up. If Bard is useless, HeyPi is worse than useless: it’s obnoxious and needy. I’ll add: the questions HeyPi gave me themselves were useless.

Claude

Claude starts to become at least potentially useful – if patronizing, condescending moralistic, and prudish. I asked it for twenty questions, and it gave me twenty. Of those twenty, fifteen were irrelevant, three were vaguely interesting, and two were genuinely interesting. I told it that.

I said, “These fifteen were irrelevant. These three were vaguely interesting.
These two were extremely interesting. Analyze that. Tell me what you think I’m looking for, what I’m finding interesting, and what I’m not. And then propose twenty more questions.” It gave me a pretty good analysis of what I was looking for, along with twenty more questions. And this time, the ratio of interesting to uninteresting improved. I ruined and related, and asked again, and
the ratio of interesting to uninteresting improved. Again. Pretty cool.

chatGPT

ChatGPT is the best of all those tools without any of the worst aspects of any of them. There was literally cnothing that Claude did that ChatGPT didn’t do better. The only difference was that ChatGPT was… different. If I’m looking for a list of a hundred questions, and I ask Claude and ChatGPT for a hundred each, eighty-five of them would overlap. ChatGPT would have fifteen additional questions, and Claude would have fifteen others. ChatGPT’s questions would, on average, be more useful than Claude’s, but they’d all be fine.

Where ChatGPT really excels is in its code interpreter.

This whole exercise concerned a spreadsheet filled with data. Claude can’t do anything with that. All Claude can do is help me think about what I’d like to do with that data, based on my descriptions of it. And maybe scold me for being inappropriate.

ChatGPT can take the questions I got from both ChatGPT and Claude, and then answer them – with charts, graphs, and words. It’s like having a $100,000-a-year college graduate
working for me, doing whatever I ask.

That is pretty fucking cool.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.