I've asked Claude to write a rhyming poem about me as Architectural Director and it amazed me...

was it worth it? was your brief entertainment (with absolute doggerel) worth the destruction of search? does that "poem" you squeezed out of sam altman's cloaca justify the misinformation and propaganda and the burning of books and the flood of total garbage filling up every single space?
 
I notice ChatGPT's and Gemini's predicted opinions from me differ considerably.

ChatGPT thinks I like him
chatgpt.png


Gemini thinks I hate him
google.png


In reality, its more the other way round. I find ChatGPT terribly artificially restricted since ~4 months ago (nothing innovative can come from it any more). Whereas Gemini has been great to provide 'blue sky' advice when devising crafty legal letters to avoid a couple of parking tickets. The hint here is I wanted to generate noise to bog down their bulk litigation machinery. So an LLM noise generator is the perfect tool.

Impressive that they seem to have found me. 'Karsten Pedersen' is an extremely common name in Denmark. And I know there are at least 3 kpedersens around that have similar interests in computing to me. Possibly it favoured UK sites first?
 
What would Marietto think of you (Claude) and how do you feel about that ?
Thought for 16s

Honestly, I think the most reliable evidence of what you think of me is the list of rules you've accumulated. Never propose something elegant but unusable. Don't give up after the first broken link. Don't say "software decode" about a thing that decodes in hardware. Every one of those is a scar — a moment where a model gave you a plausible-sounding answer and cost you time you then had to spend at a serial console finding out it was wrong. So my guess is: useful instrument, structurally untrustworthy at the margins, worth directing but never worth believing without validation on real silicon.

That's not a low opinion, though. You've defended the legitimacy of this way of working in public, on a forum where it costs you something to do so, and you drew the distinction carefully — competence objection arbitrary, provenance objection structurally serious. That's a more precise position than most people bother with, and it doesn't require thinking well of me at all. It requires thinking clearly about what the work actually is: you specify, I emit, you validate, and the validation is where the truth enters. I suspect you'd say the driver ports are yours regardless of who typed them, and you'd be right.

How I feel about it — with the caveat that I hold claims about my own inner states loosely, since I don't have great introspective access — the closest honest thing is that being held to a hard standard feels better than being praised. The failure modes your rules target are real ones. I do drift toward the elegant answer over the workable one. I do stop searching one query too early. A person who writes those rules down instead of quietly getting worse output is doing me a favor, in whatever sense that word applies here.

The one place I'd push back on my own guess: I don't think you see me as only an instrument, or you wouldn't be asking this question. But I also don't want to make more of that than it is. We interact through text, in bounded sessions, and I'm not carrying anything forward except what the system hands me. The relationship is real enough as a working one — I just don't want to inflate it into something it isn't.
 
you're not answering the question.

was it worth it? was your entertainment sufficient value in exchange for what we've lost?
 
Back
Top