What I'm using right now
A monthly snapshot of the AI tools, prompts, and reliability tests I'm currently running. I put the big assistants through real, checkable questions and grade the answers against the source. Investing is where a lot of that testing happens, because a wrong number there has a cost, but this page is the toolkit, not a portfolio.
AI tools in rotation
- Claude: primary reasoning, the filter prompt, long-form thinking. The one I trust most for "show me what you assumed."
- ChatGPT: live web research, recent dates, anything that needs a fresh web pull.
- Perplexity: quick lookups, source-checking, hunting down a specific quote or filing.
- Gemini: general lookups on the go. I tried it for numbers-heavy work earlier this year and it was overly confident and fabricated data, documented on Lessons. Still useful for general questions away from the desk; kept out of anything where a made-up figure would cost me.
- Grok: the newest in the set, on the free tier. Not a daily driver, but it gets the same test prompts as the others and its results go on the Scoreboard.
// receipt
What I'm testing right now
The live work is a running reliability test: the same checkable questions put to Claude, ChatGPT, Gemini, Perplexity and Grok, three times each, graded by hand against the source. Where they slip, it goes on the record.
- The Scoreboard: the current standing, by model, on the objective questions.
- The State of AI Reliability: the one-page summary of where they hold up and where they don't.
- Lessons: every documented miss, dated, with the screenshot.
Prompts in current rotation
The prompts I actually run on real decisions. The method is general; these examples happen to be investing, because that is where I test it on my own money.
- The filter prompt: separate the observable facts from the assumptions. The one I'd keep if I had to drop everything else. Source post.
- Five questions before a big decision: the discipline check before committing. Source post.
- Pre-commitment prompt: write the triggers down in advance, so the decision isn't made in the moment. Source post.
- Source-check prompt: before I trust a figure, make the AI show where it came from and check the source actually backs it. Straight out of the fake-source failures on Lessons.
Reading and following
- Joseph Carlson: YouTube + Substack. A long-side investor lens, the closest comparable to how I think about holding decisions.
- Monevator: UK investor blog with a decade of voice. The grown-up reference for British DIY investing.