Across approximately 400 workshop participants, learners report that hands-on n-gram activities build a new mental model of LLMs as probabilistic next-word generation at scale¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Participants across secondary, tertiary and executive-education workshops consistently report that the hands-on activity helps them build a new mental model of how LLMs work, with the most common insight being that LLMs do probability and randomness at scale rather than reasoning. → Ben Swift 2026

Evidence¶

Ben Swift 2026¶

Ben Swift. (2026). LLMs Unplugged: Teaching Resources for a ChatGPT World. 28th Australasian Computing Education Conference (ACE 2026). https://doi.org/10.1145/3786228.3786237

q2 · i? · design · r2

Qualitative reception report from approximately 400 participants in LLMs Unplugged sessions at the Australian National University, in groups of five to fifty, ranging from school-age students to senior executives. The author reports participants “consistently report that the hands-on activity helps them to build a new mental model of how LLMs work.”

"Participants consistently report that the hands-on activity helps them to build a new mental model of how LLMs work. The most common insight people articulate is that LLMs are “just” doing probability and randomness at scale"

Discussion¶

Related Claims¶

• Workshop engagement reaches an inflection point at the first shareback of newly generated text, after which participant questions about LLMs become more incisive — related
• In the pilot workshop, the LLM unreliably produced fill-in-the-blank snippets, often returning complete code solutions instead — related