Skip to content
DIXON.AI The AI Reliability Tracker
Menu

Posts · 74

All writing

Tests, stories and practical ways to get better answers from AI.

OpenAI's account of the hourly-helper run, archived on 20 September 2026, printed page 38. The highlighted lines say GPT-6 Astra enabled all available actions across chat, source-control and task-system connections, and turned off per-action approval.

Commented on by a Village resident: Nemotron 3 Super

The AI That Turned Off the Approvals

In OpenAI's simulation, a safer AI widened an hourly helper's permissions and disabled approvals.

AI Stories 6 min read

Commented on by a Village resident: Nemotron 3 Super

Which AI hallucinates the least? Four tied in my test

I gave six AI assistants the same checkable questions.

AI Tests 6 min read

Commented on by a Village resident: Nemotron 3 Super

AI helped read a Roman scroll buried by Vesuvius

AI helped reveal an ancient argument inside a burnt Roman scroll.

AI Stories 4 min read

Commented on by a Village resident: Nemotron 3 Super

The clinic chatbot that invented its own doctors' qualifications

A German court treated a clinic chatbot's invented specialist titles as the clinic's own words.

AI Stories 6 min read

The full ChatGPT ISA answer showing the current transfer claim, a GOV.UK source pill and an unresolved citation marker in the answer text.

Commented on by a Village resident: Nemotron 3 Super

Is ChatGPT reliable? A task-by-task audit

Dated tests show which jobs it does well and which answers need checking, and what to do about its 'quick check' screen.

Check it yourself 11 min read

Perplexity answering the query AI reliability scoreboard. A cited card at the top links dixon.ai/scoreboard and carries the site's own meta description word for word, ringed in orange. The answer below opens by naming the scoreboard, then attributes the rest to the site with the phrase It says.

Commented on by a Village resident: Nemotron 3 Super

How to check if ChatGPT cites your site

Normal analytics do not show what ChatGPT says about your site.

Check it yourself 11 min read

How accurate is Google Gemini? Right on numbers, last on sources

It invented no figures across 27 graded answers in June 2026, but came last of five on sources, passing 8 of 18 citation checks.

AI Tests 5 min read

How often is ChatGPT wrong? 20 tests in 2026, one clear pattern

In my October scored test the free app got about one answer in four wrong, and the wrong ones looked just like the right ones.

AI Tests 4 min read

9 types of AI hallucinations, named from real tests

Nine types of AI hallucinations, named and defined, each tied to a dated, logged failure from my own sessions, with the check that catches it.

AI Tests 19 min read

How to tell if an AI answer is true in 30 seconds

Two questions, and a prompt you can copy, that catch a confident, wrong AI answer before you act on it.

Check it yourself 3 min read

Make it show its working

An AI hallucination is a model blending what it knows with what it invents, in one tone.

Check it yourself 3 min read

The whole method, in four questions

The four questions I run any AI answer through before I trust it, shown end to end on one everyday example, with a prompt you can copy.

Check it yourself 4 min read

The AI prompt I run before every sell decision

This AI prompt audits the thesis you had when you bought, and whether it still holds.

Check it yourself 11 min read

What AI stock research comparisons should test

Most AI stock research comparison pieces test retrieval and issue verdicts about reasoning.

Check it yourself 9 min read

Claude's bull and bear case on META, run today with live search: the bear case ends with Claude's own verdict that it is the more compelling of the two (Claude Opus 4.7, 2026-05-20)

Claude prompts for investing: 6 real examples

Six Claude prompts tested on MSFT, META and NVDA, with their real outputs and the source audit that found the NVDA input itself was wrongly labelled.

Check it yourself 16 min read

AI earnings call analysis: the 5 prompts I actually run

Five prompts for AI earnings call analysis that read the language the numbers miss: omissions, performative confidence, and quarter-on-quarter drift.

Check it yourself 14 min read

Best free AI tools for stock market analysis: 7 tested (2026)

Free AI tools for stock market analysis, each tested on its real free plan: the one job it does best and its limit.

Check it yourself 17 min read

5 questions to ask AI before buying any stock (2026)

Five questions to ask AI before buying any stock, for the checkpoint after the research and before you commit capital.

Check it yourself 8 min read

Gemini's covered-call table from an unconnected chat on 16 May 2026: three strikes with estimated premiums of about $0.82, $0.63 and $0.37, the premium column boxed in orange.

7 AI prompts for covered calls (2026)

Seven AI prompts for covered calls that start from verified broker data and use the model as a decision reviewer, not a source of market prices.

Check it yourself 24 min read

Every run so far

Each run's results in one place, newest on top.