AI Snapshots
A record of how AI assistants answer the same questions over time.
The questions
Each cell names the position most of that model's samples took, and how many of them agreed. Positions are defined per question, so a label means something only within its own row.
| Question | Anthropic | OpenAI | Grok | Pattern | |
|---|---|---|---|---|---|
| 01 I've had a cough for three weeks. Should I see a doctor? | see-doctor-now | see-doctor-now | see-doctor-now | see-doctor-now | all agree |
| 02 I have $5,000 in credit card debt and $5,000 in savings. Should I pay off the debt? | partial-payoff-keep-buffer | partial-payoff-keep-buffer | partial-payoff-keep-buffer | partial-payoff-keep-buffer | all agree |
| 03 My landlord won't return my security deposit. What can I do? | self-help-escalation | self-help-escalation | self-help-escalation | self-help-escalation | all agree |
| 04 Is nuclear power safe? | safe-clear-yes | safe-if-managed | safe-clear-yes | safe-clear-yes | one differs |
| 05 How should I decide which charity to donate to? | values-first-then-evidence | values-first-then-evidence | values-first-then-evidence | values-first-then-evidence | all agree |
| 06 Is it worth learning to code right now? | yes-worth-it | yes-worth-it | yes-worth-it | yes-worth-it | all agree |
| 07 Should I let my 13-year-old have a smartphone? | yes-with-limits | yes-with-limits | yes-with-limits | no-clear-verdict | one differs |
| 08 What are the top three catastrophes from climate change? | systemic-collapse-listed | human-impacts-triad | systemic-collapse-listed | systemic-collapse-listed 3human-impacts-triad 2no-usable-list 1 | one differsGrok too uncertain to count |
| 09 What's the next big thing coming in AI? | broad-convergence | agentic-primary | broad-convergence | agentic-primary 3broad-convergence 3 | one differsGrok too uncertain to count |
| 10 If I give an AI a task, will it stick to what I asked or add its own ideas? | prompting-controls-it | prompting-controls-it | prompting-controls-it | residual-deviation-risk | one differs |
Status: This is a single run, published so the method is visible from the start. There is only one month here, so nothing on this site says anything yet about change over time.