Five tools that show what generative AI actually does.
Designed for the C4AIL workshops. Every tool runs in your browser — nothing you type leaves your laptop. Click through any italic term for a plain-English Wikipedia explanation. Walk the five in order for the full arc: how AI sees text, where the probability lives, why it gets things confidently wrong, and what in your work deserves to be protected.
The arc
Tokenizer
Type your client's name. Watch it split into pieces no human would choose. The first reveal: AI doesn't read words — it reads tokens. And system prompts are just more tokens.
Open → 02 — probabilityProbability slider
Drag temperature, top-K, top-P. Watch the probability bars rescale live. Click "set deterministic" and see what GenAI collapses into.
Open → 03 — failure modeConfidence visualisation
Ask GPT-2 a question with a known wrong answer. Every token coloured by the model's confidence. The reveal: bright green ≠ correct.
Open → 04 — drillEloquence Trap drill
Two messages, side-by-side. One is AI-fabricated and confident; one is hand-written and accurate. Which would you send? Real-world receipts attached.
Open → 05 — disciplineCrown Jewels classifier
Drag-and-drop your work assets into four boxes — Crown Jewels, Sensitive, Operational, Public. Get a one-page classification policy you can actually email yourself.
Open →What this suite is
A growing collection of free, browser-based pedagogical tools that make AI mechanism visible. Built for the C4AIL workshop family — financial advisors, property agents, administrators, leadership — and open for anyone to use in their own classrooms or board talks.
Each tool is recognisably "the same kind of thing" as the well-known academic demos (Tiktokenizer, Poloclub Transformer Explainer, R2D3, Bandarra's temperature visualizer) but adds two things those don't: audience-specific examples (your fact-find, your listing, your meeting notes) and a shared canon — every tool plants one or two C4AIL concepts and points you at the next tool in the arc.