Open the Next-Token Prediction Game and choose one built-in sentence stem. Before looking at the model distribution, enter your own prediction as a one-line guess and tally it. Compare your choice with the model's top predictions, then name one place you agreed or disagreed and one reason why.
~15–20 min · one screenshot and two sentences
Choose one built-in prompt in the Tokenizer + Temperature Visualizer. Run it at low and high temperature, then compare greedy decoding with sampling. Document what changed in the output and what stayed constant about the mechanism.
~30–45 min · one written or screenshot comparison
Open the ELIZA Simulator and have a brief exchange. Then switch to its ELIZA vs. LLM tab and choose one paired prompt. Write down what ELIZA makes visible about the mechanism that the LLM hides, plus one moment where the outputs feel similar despite the mechanisms being different.
~25–35 min · one page or a few bullet points
Open the Whose Preference? Lab. Rank one prompt’s three responses before revealing the authored panel. Then change one thing: include or exclude a rater perspective, or switch the aggregation rule. Document which response became preferred, what the change rewarded, and whose priorities were left out.
~20–30 min · one before/after comparison and one bounded claim
Design an unplugged version of the next-word prediction game — one that works without a computer, internet connection, or AI account. What is the sentence stem? How do you tally the room? What is the debrief question? Create a one-page protocol, a set of slides, or a rough worksheet.
~45–60 min · one-page protocol, slides, or rough worksheet
Choose one pre-generated output from the Text Prompt Pack. Identify one default, assumption, or limitation you can support with an exact phrase from the output. Separate what the evidence shows from your hypothesis about why it happened, then name one question you would need answered to test that hypothesis.
~20–30 min · one annotated example · no live AI use required