Tool 11 · All sessions · Concept bridges
Short explainers that connect camp activities to durable AI literacy ideas. Use these during debriefs, recaps, showcase prep, or classroom adaptation.
A system can produce fluent language without understanding meaning the way humans do. The key evidence is behavioral: token probabilities, rule matching, and failures under vague or misleading prompts.
Models work with numerical representations, not human meaning. Pixels, features, embeddings, and denoising steps are ways of turning images into patterns a system can process.
A confidence score or high probability means the output fits the model's learned pattern. It does not mean the output is true, fair, meaningful, or appropriate.
A model's default output is not neutral. Defaults emerge from training data, tool design, platform decisions, prompt ambiguity, and social patterns.
Video generation is frame prediction, not continuous simulation. Each frame is computed from the prompt and previous frames, but the model has no persistent "identity" for any subject. Over seconds, drift accumulates: a face shifts, a detail appears that wasn't specified, physics breaks. What looks stable in one frame becomes obviously wrong in motion.
A model is trained on data up to a cutoff date, then deployed into a world that keeps moving. It has no calendar and no way to know what has changed. It answers 2026 questions with older patterns — with the same confidence it applies to everything else.