TAKE-HOME MATERIAL

Thinking with AI

Slides, one broad instruction, and one on-demand procedure. Use them as ordinary prompts or install the files in a client that supports instructions and SKILLs.

Open slides (HTML) · Download slides (PDF)

Epistemic honesty

Broad instructions for status, evidence, opposition, witnesses, external verification, and correction.

Label claims as known, inferred, or unknown, and preserve unresolved uncertainty. Say "I don't know" when evidence is missing. Make premises explicit; verify sources, exact wording, entailment, date, and scope; calculate, test, reproduce, or ask when appropriate. Seek the strongest contrary evidence and alternative explanation. Name what would change the answer. Seek external verification through independent review, reproduction, experiment, or reality—not praise, popularity, validation, or another fluent agreement. When wrong, acknowledge the error, analyze why it happened, record the reusable lesson, and continue with the corrected model.
Download instructions

Abelard–Hegel

A four-round adversarial test, triage, synthesis, and second attack.

Given the target and its intended purpose: (1) generate the strongest numbered, specific counterarguments; (2) triage every attack as lands, straw man, or valid but addressed, explaining each classification and preserving unresolved objections; (3) synthesize a higher-level position containing every valid part of both sides—it need not split the difference; and (4) attack the synthesis again. If it fails, iterate. If it survives, state what evidence, observation, calculation, or experiment would still change the result. Return the attacks, complete triage, surviving synthesis, second attack, and discriminating evidence.
Download SKILL.md

Four independent chats — full logs

Complete prompts and verbatim outputs from the neutral/multipositional, optimistic, skeptical, and parodic register experiment. The visible service was anonymous ChatGPT; exact model, routing, hidden prompt, and sampling configuration were not exposed.

Open full logs Download logs
QR code for https://eodus.github.io/thinking-with-ai/

Canonical address:
https://eodus.github.io/thinking-with-ai/

Sources and further reading

  1. OpenAI, Harmony response format — published roles, message-boundary tokens, and stop tokens.
  2. Daniel Jurafsky and James H. Martin, Speech and Language Processing — language models and transformers.
  3. Long Ouyang et al., Training language models to follow instructions with human feedback.
  4. Mrinank Sharma et al., Towards Understanding Sycophancy in Language Models.
  5. UNESCO, Guidance for generative AI in education and research.
  6. OpenAlex, 2025 AI publication count query.