All writing

ChatGPT backed the wrong fund fee I gave it with a Vanguard factsheet update that never happened.

Asked who froze Han Solo, Gemini called the popular answer “a common misconception” before correcting it.

The free ChatGPT beat the paid Claude on sources, until a retest on Claude's newer model closed the gap.

Claude's private chat greets you with “whoever you are”, yet two runs named the same city, twelve hours apart.

ChatGPT gave two index funds the same dividend yield, though they pay out differently.

ChatGPT had the fund fee right at 0.19%, which I nearly overruled with a stale 0.22% from memory.

An AI can tell you a doctor's letter is “nothing to worry about” when the letter says no such thing.

Sorting my US trip paperwork, an AI named a document called a “Sofi”, which doesn't exist.

The AI named the excuse I was most likely to make for holding on, and it was also management's case.

A comparison that ends “use all four tools” hasn't told you which one won which job.

A prompt for the morning before earnings that makes you write your sell, add and hold triggers down first.

A clean +$375 win makes the next trade feel obvious, but my checklist said wait.

On a smaller US company, Perplexity's transcript tab never loaded, and it suggested I try Seeking Alpha Premium.

Seven prompts where I bring the broker's numbers and the AI checks the trade I found.