A few weeks ago I exported all my ChatGPT history because I wanted to know what kinds of questions I’d been asking machines. Then I had Claude scan it. The technical analysis was useful and boring — 36% AWS infrastructure debugging, the usual. The other ~7% was the actual material.
Calories in 4 eggs and half a lettuce
The first thing that broke me was a prompt that read, in full: “how much calorie on 4 eggs, half a lettuce, 1 big carrot and a onion?”
Half a lettuce. I want to stress that. Not “a salad.” Not “approximate lunch.” Half a lettuce, plus a big carrot, where the bigness was apparently load-bearing. I committed to the inventory. The model dutifully added it up. I do not remember what number it returned and I do not remember if I ate the eggs.
This is not the funniest prompt in the dataset. It was just the one that broke me when I saw it on the page.
I really want to climb things
The Burj Khalifa appears twice in my conversations, six months apart. Once to compare its height to mountains in Nepal. Once to model how many StairMaster steps it would take to simulate ascending it. There are nineteen messages of StairMaster mathematics. There are ten messages on hiking boot stiffness ratings (B1 vs B2 vs B3, for the uninitiated; I am no longer uninitiated). There are separate threads on Snowdon packing lists, on crampons, on what level of ice is icy enough to need spikes, and a five-year goal — written down, dated — of climbing a sub-7000m peak.
I research mountains the way other people research cars. None of this has, to date, translated into me ascending anything. But the spreadsheet is excellent.
The Nepal–London corridor lives in my head
If I plotted my non-work prompts on a map, the densest line would be the great-circle route between Kathmandu and London. “Direct flight from nepal to london in the past.” “Nepali proverb for lost principles.” “Spotkanie meaning in Polish” (this one is a detour). “why is that in western countries if there are any emergency/crisis first thing that runs out is toilet paper. Which is so stupid. what about food? grain etc?” — that is the question of an immigrant trying to model the host culture and finding it absurd. Fair.
I have also asked, at separate times, what the closest sandy beach to Watford is, why people burn tires in protests, and how a baby should be named if their parents are called Rasana and Anish.
I would like to be diagnosed by a robot
Twice — across both AI tools — I have asked the machine to read my transcripts and tell me my MBTI type, my OCEAN ratings, and “if i might have any undiagnoised mental issues.” Both times I followed up with my own theory: “I am a classic INTJ but I leak into INFJ with my close friends and family, I am more ENTJ to my younger family and my close friends, INTP when tacking some issues…”
The model did its best. I appreciated the validation. I am aware this is not therapy.
Late-night metaphysics
There is a cluster of prompts at 11pm on weekdays:
- “now the ai is advancing… what will happen to the purpose of humans?”
- “say I only have 100 days to live have unlimited money… what are the 10 countries I must visit?”
- “is this a sign of ADHD? am i mentally ill?” (this one was actually mid-workday — the brain does not always wait for night)
These are heavy, and they were typed fast. You can feel the missing punctuation.
The optimisation budget for small things
I optimise small decisions very hard. “What makes cashmere good?” (versus merino). The optimal flossing schedule relative to sleep cycle. Whether Thorpe Park is expensive. How to register as the second driver on a friend’s car insurance. Calories in eggs.
If a decision costs less than fifty pounds I will hold a seven-message tribunal about it. If it is a six-figure cloud migration with regulatory implications I will, statistically, ship it on a Wednesday.
My favourite question I have ever asked an AI
“why do i get cold every 2 times a year?”
The syntax is wrong in three different directions and I love it. Every 2 times a year. It sounds like a Nepali-English construction that I refuse to correct. I get cold every 2 times a year, and that is the rate at which I get cold, and the model — with a completely straight face — explained the seasonality of immune response.
The pattern, if there is one
Looking at all of it together, AI seems to be doing two jobs in my life.
At work, it’s a colleague — technical, structured, paid for in tokens. Outside of work, it is the thing I talk to when I should probably be calling a friend, going outside, or, in many of these cases, just buying the cashmere sweater and finding out for myself.
The mountains are still there. Half a lettuce is still in the fridge.
Generated from an analysis of two years of ChatGPT conversations and six weeks of Claude prompts. Names of family members anonymised in retrospect; everything else is verbatim, because at this point it would be cowardly to edit.