an experiment in ai, mostly under control

probablydeterministic

Probably what you expected.

AI ideas, tested in public before they're allowed to have an opinion. Own space. Own scope. No guarantees. Mostly works. Confidence interval: wide.

Experiments n = 0 · queue = 9

Nothing shipped yet.

That's not a bug — it's the prior. Things will show up here once they're confident enough to be looked at by strangers.

$ ls experiments/

// dry humor, meta-ai

Confidence Interval Generator

Feed it a claim, get a percentage and a deadpan explanation. "Will this work?" → "67.3%. Margin of error: yes."

Probably Fine Checker building next

Describe a technical decision. It rules: probably fine, technically fine, statistically fine, or not fine (ship it anyway).

AI Disclaimer Generator

Legal-sounding, fully absurd disclaimers for anything. "Results may vary depending on sampling temperature and cosmic background radiation."

// useful, still funny

Prompt Autopsy

Feed it a prompt that failed. It performs the autopsy — what went wrong, what you should've written instead.

Overfitted Advice

Extremely specific advice from way too little context, openly overfitted. "Based on your two words, here is my 14-step plan."

Cold Start Interviewer

Asks questions to help you kick off a new project. Only output, no matter what you say: "okay, you're probably ready to start now."

// more useful, still on brand

Token Budget Estimator

Describe what you want to build, get a monthly API-cost estimate with a wide margin of error and a dry comment.

Vibe Check

Paste code, get judged on vibes instead of correctness. "Chaotic but probably works. Senior dev would cry. Ship it."

Hallucination Detector (ironic)

Answers with total confidence to questions it can't possibly know. Always ends with "source: probably real."

What this is

Approach

Ship small, see what breaks, keep what's useful.

Scope

Chatbots, generation tools, automations — whatever's worth a weekend.

Guarantee

None offered. Results vary by sampling temperature.