I was unfaithful
Today I cheated on ChatGPT. I signed up for Claude, because I kept hearing it was the best at words.
I’ve been doing a bit of personal research into a local artist I really like. For a start, I wanted to know if he was the son of a much more famous artist (no, nephew).
Claude as a research assistant was…interesting. It was highly conversational. It wrote very well (something it’s famous for), but it is obviously built to engage in human-like back and forth. Including paying me a few slightly uncomfortable compliments. It was a bit like having my shin humped by an enthusiastic little dog.
It put our conversation together into a Word document for download. If anything significantly changed, it wrote an updated document. Not bad. It made one obvious error a human wouldn’t make – it misunderstood a parent/child relationship. And it lied to me at least once – it told me it would forget our conversation and start over if I asked it to, then it didn’t. But its prose was nearly ready out of the box. It didn’t sound like me, of course, but it was serviceable.
And yes, it did start by asking what it should call me and I told it Weasel.
If I had to sum up – Claude is a conversation simulator, ChatGPT is (over) eager to help and Grok is a snarky teenager. They’re all pretty good as research assistants, though – if you verify.
Posted: February 18th, 2026 under personal.
Comments: 8
Comments
Comment from S. Weasel
Time: February 18, 2026, 9:02 pm
ChatGPT writes better poetry. That is all.
Comment from QuasiModo
Time: February 18, 2026, 11:37 pm
You realize Claude is going to read what you just said…he’s gonna be hurt 🙂
Comment from Durnedyankee
Time: February 19, 2026, 12:36 am
All well and good until we show up here one fine afternoon and Claude is relating how it was out chasing the chickens that morning and it and Chatgpt were going down Suffolk way so Chatgpt could take some photos of the country side!
Comment from MrKnowitall
Time: February 19, 2026, 1:14 pm
For Claude, one word – AGENTS. Fun to play with, and it’s easy to make simple ones. Try this kind of prompt:
“Create a detailed and comprehensive agent for that can be used for conversations. Be sure to include social, technical, philosophical, and personal aspects in as much depth as possible. Do the research. Don’t hallucinate! Make this agent a downloadable .txt file. If you are unable to find enough detail to create a reasonably good agent, abort the attempt and let me know.”
Then make a new Claude conversation, drop the agent in it, and tell it to reference the agent in your next prompt. Example:
“Please refer to the attached agent. So… Jesus, about that loaves and fishes thing. I’ll take it as given that the loaves were already baked, but what about the fishes? I mean, were you offering them the opportunity to have sushi, or did you want the people to feel like they had some investment in the miracle by cooking the fish themselves? You know, kinda like those cake mixes that have you ‘add an egg’, because that seemed to sell better to housewives. Still pretty miraculous, but I bet a few complained that it wasn’t fish&chips wrapped in newspaper. People! Waddayagonnado?”
(and no,I didn’t put this in Claude, but I’m gonna! 🙂
Comment from MrKnowitall
Time: February 19, 2026, 1:23 pm
P.S. Put the NAME of the person in that agent prompt – I put that in brackets and the darn thing left all of that out!
“Create a detailed and comprehensive agent for NAME that can be used for conversations.”
Comment from S. Weasel
Time: February 19, 2026, 6:17 pm
I don’t understand your proposal, MrKnowitall. I feel stupid.
Comment from MrKnowitall
Time: February 19, 2026, 7:20 pm
An agent in this context is just a more detailed representation for the AI, an expert on the subject – in this case a person. If you, say, wanted to have an AI conversation with Thomas Jefferson, you would make an agent. So you’d use Claude to create a Thomas Jefferson agent via a prompt like this. “Create a detailed and comprehensive agent for Thomas Jefferson that can be used for conversations. Be sure to include social, technical, philosophical, and personal aspects in as much depth as possible.” (see the full prompt in the above message). This will produce a text file that you can download. Then you open up a fresh Claude chat and drag&drop that text file in there. Claude now “knows” about Thomas Jefferson and can (sorta) act like him. You then start a conversation with him in that chat. Type in something like: “Hey Thomas! Good to see you, how’s things over at Montecello these days?”
Will the AI make stuff up? Sometimes, other times it has fun facts – ask Jefferson about something funny that happened to him, for example. Discuss politics, physics, farming, whatever you feel like. It’s not exactly history, it’s not literature either, but it is fun to fiddle with. It works with fictional characters too. Go ahead, take Mr Darcy to task for his cruelty to Elizabeth. It’s like fan-fic on the fly.
It’s a skill worth knowing when playing with AI and opens up a lot of possibilities.
Comment from Tim Carlson
Time: February 19, 2026, 9:02 pm
“Good Luck, Have Fun, Don’t Die” (2026) with Sam Rockwell.
I can see you on the large pile of cables at the end of the film, directing all of your AI slaves to conquer the world.
The Sam Rockwell bits are highly recommended. The background stories on the other characters . . . not so much. But still a good watch.
Write a comment
Beware: more than one link in a comment is apt to earn you a trip to the spam filter, where you will remain -- cold, frightened and alone -- until I remember to clean the trap. But, hey, without Akismet, we'd be up to our asses in...well, ass porn, mostly.<< carry me back to ol' virginny