AI Atlas · Optional exploration · 1 / 7 · Imitation

What would a fair comparison show?

Browser companion, outside core E0–E7; lab completion does not complete the quest. Time is unmeasured. Not yet learner/facilitator validated. Choose Free exploration or pause the quest. Keep the answer key covered before each response.

  1. Open Imitation. Finish the original comparison, then choose Design a fair comparison.
  2. Choose a setup, a possible claim and a reason before revealing the fictional preference cards. Inspect what the setup held constant. Revise your explanation; keep the first one.
  3. Change the question to factual accuracy. Commit a fresh decision. Inspect the four answer-key rows; explain why the old style result cannot settle this new task.

First setup, claim + why · learner code ______

Revised explanation + evidence

Changed task: fresh prediction → observed result → explanation

Write or scribe exact spoken words. Access support ______ Content hints ______ Keep both responses before discussion; use browser history to compare.

Cover until the relevant response is saved

Worked answers. Same question/limits with names hidden supports a preference about these replies: fictional A 4, B 2. It proves neither identity nor consciousness. Changed task: fresh questions, fixed answer key, names hidden. A gets 3/4; B gets 4/4. A wrongly selects 0.75 as larger than 0.8. Limit the claim to these checked replies.

Prompt: “Would friendlier wording prove who wrote it?”

Manual explanation: 0 missing/unsupported; 1 relevant result but weak comparison or scope; 2 names the fair setup, exact result and limit. These are explanation codes, not product-rubric points. Prose is not automatically graded.

Optional record · first ______ revised ______ changed ______1 / 7

AI Atlas · Optional exploration · 2 / 7 · ChatGPT

Repair the reply. Or justify trust.

Browser companion, outside core E0–E7; lab completion does not complete the quest. Time is unmeasured. Not yet learner/facilitator validated. Choose Free exploration or pause the quest. Keep the answer key covered before each response.

  1. Open ChatGPT and choose Try a reply check. Commit your first judgment about the binocular loan period.
  2. Try a follow-up; decide whether it changes the evidence. Check the original for this exact item. Repair or approve the detail, limit the claim and explain. Revise if needed.
  3. Try the changed Cedar Loop case. Make a fresh judgment before checking its sources. Compare current and archived versions, then decide again.

First judgment + why · learner code ______

Checked source + revised decision

Changed case: fresh judgment → checked source → decision + why

Write or scribe exact words. Access support ______ Content hints ______ Preserve the first judgment, follow-ups, source checks and revisions.

Cover until the relevant response is saved

Worked answers. Simplification/reassurance still says 14 days. The fictional equipment card says 7 days; repair 14 to 7. A book guide is about the wrong item. Changed case: the current notice supports 1.2 km; approve that detail. The archived 1.8 km guide was replaced. Trust covers the checked fictional detail, not all replies or real-world accuracy.

Prompt: “Did this follow-up add independent evidence, or only different wording?”

Manual explanation: 0 unsupported; 1 correct detail without item/version or scope; 2 cites the relevant source, justified repair/approval and limit. Prose is not automatically graded; completion is not understanding.

Optional record · first ______ revised ______ changed ______2 / 7

AI Atlas · Optional exploration · 3 / 7 · Dartmouth

New words. Changed meaning.

Browser companion, outside core E0–E7; lab completion does not complete the quest. Time is unmeasured. Not yet learner/facilitator validated. Choose Free exploration or pause the quest. Keep the answer key covered before each response.

  1. Open Dartmouth. Finish the original reasoning/learning cases, then choose Test the language ambition.
  2. Predict whether both unfamiliar wordings preserve the intended action. Choose the scope and explain before the check. Compare intended and observed destinations; revise or confirm with evidence.
  3. Change the meaning. Make a fresh prediction about “Do not move” and “Keep”. Inspect each outcome, revise the explanation and retain the first words.

First prediction, scope + why · code ______

Observed destinations + revision

Changed meaning: prediction → intended/observed action → why

Write or scribe exact words. Access support ______ Content hints ______ Keep original reasoning/learning credit separate from these language records.

Cover until the relevant response is saved

Worked answers. Both paraphrases request B and produce B: 2/2 match. Claim only these wordings; the frozen program receives no feedback or update. Changed case: “Do not move … to B” expects A but produces B. “Keep … at A” expects and produces A: 1/2 match. One counterexample limits the broader claim. All results are authored, not workshop events or live AI.

Prompt: “Can nearly the same words ask for the opposite action?”

Manual explanation: 0 unsupported; 1 cites an outcome but misses meaning or scope; 2 names the request, intended/observed destination and claim limit. Prose is not automatically graded.

Optional record · first ______ revised ______ changed ______3 / 7

AI Atlas · Optional exploration · 4 / 7 · Lisp

Follow the groups. Keep the type.

Browser companion, outside core E0–E7; lab completion does not complete the quest. Time is unmeasured. Not yet learner/facilitator validated. Choose Free exploration or pause the quest. Keep the answer key covered before each response.

  1. Open Lisp and choose Try the expression lab. Predict the full result of (* (+ 2 5) (- 9 6)) and explain before stepping.
  2. Inspect each committed evaluation. Choose the final rule and trace row. Revise wrong predictions or explanations without deleting the first record.
  3. Try symbolic list data. Predict the result, then change the outer operation from car to cdr. Predict again; include parentheses if the result is a list.

First result prediction + why · code ______

Final rule/trace row + revision

Changed operation: prediction → observed result/type → why

Write or scribe exact words. Access support ______ Content hints ______ Keep the number case, symbol case and changed-list predictions in order.

Cover until the relevant response is saved

Worked answers. Inner values are 2+5=7 and 9−6=3; final row 3 multiplies 7×3=21. Symbol case: quote keeps (red blue green); cdr gives (blue green); outer car gives blue, a symbol. Changed outer cdr gives (green), a one-item list. Cite final row 3 in each case.

Prompt: “Why are green and (green) different answers?”

Manual explanation: 0 unsupported; 1 correct value but incomplete operation/type account; 2 connects nested values, final operation and symbol/list distinction. This bounded model does not establish understanding of colour meanings. Prose is not graded.

Optional record · first ______ revised ______ changed ______4 / 7

AI Atlas · Optional exploration · 5 / 7 · Logic

Which arrow is justified?

Browser companion, outside core E0–E7; lab completion does not complete the quest. Time is unmeasured. Not yet learner/facilitator validated. Choose Free exploration or pause the quest. Keep the answer key covered before each response.

  1. Open Logic and choose Choose my proof steps. With A given, choose an arrow and predict whether its starting fact and rule are available.
  2. Commit the reason before testing. Keep failed shortcuts; choose further steps to support the goal. Save a conclusion citing the chain, then revise if needed.
  3. Change the starting facts to C and the goal to A. Predict a fresh step and conclusion using only the listed rules. Do not silently reverse an arrow.

First step, validity + why · code ______

Supported chain + revised conclusion

Changed facts: fresh step/conclusion → evidence → explanation

Write or scribe exact words. Access support ______ Content hints ______ Record the starting facts as well as the arrow; retain invalid attempts.

Cover until the relevant response is saved

Worked answers. Available rules are A→B and B→C. B→C is invalid before B is established. From A, apply A→B then B→C: the actual chain proves C here. With only C given, neither rule produces A. C→A is unavailable, so A is not derivable in this system; that does not prove A false in the world.

Prompt: “Which starting fact and available rule justify that arrow?”

Manual explanation: 0 unsupported; 1 correct conclusion without its chain or missing prerequisite; 2 cites the exact rule/fact support and separates derivability from real-world truth. Prose is not automatically graded.

Optional record · first ______ revised ______ changed ______5 / 7

AI Atlas · Optional exploration · 6 / 7 · Information

Predict both possible answers.

Browser companion, outside core E0–E7; lab completion does not complete the quest. Time is unmeasured. Not yet learner/facilitator validated. Choose Free exploration or pause the quest. Keep the answer key covered before each response.

  1. Open Information; choose Choose my own questions. Ask “Is it 1?” Predict counts after YES and NO, an information claim and reason. Do not guess the hidden target.
  2. Reveal; compare both branches, remaining uncertainty, actual-answer information and expected information. Revise without filtering twice. Continue with upper half, even, then “Is it 6?”
  3. Try Fresh target, same first question. Predict both counts again before asking. Compare the answers and amounts of information.

First YES / NO counts + why · code ______

Revised counts / information claim

Fresh target: both predictions → actual answer → explanation

Equal-prior teaching model, fixed authored targets. Access support ______ Content hints ______ Write or scribe exact words; keep the first counts.

Cover until the relevant response is saved

Worked answers. “Is it 1?” splits 1 / 7. First answer NO: uncertainty log₂7≈2.8074 bits; actual information log₂(8/7)≈0.1926. Expected information = (1/8)×log₂8 + (7/8)×log₂(8/7)≈0.5436. Next branch counts: upper half 4/3, even 2/2, singleton 6 1/1. Fresh case answers YES to the same first question: 3 bits, with the same expected average.

Prompt: “Does every binary answer always give one bit?”

Manual explanation: 0 unsupported; 1 correct count but confuses the quantities; 2 connects branch counts to remaining uncertainty, actual and expected information. Prose is not graded.

Optional record · first ______ revised ______ changed ______6 / 7

AI Atlas · Optional exploration · 7 / 7 · Neuron

One fixed rule. Different inputs.

Browser companion, outside core E0–E7; lab completion does not complete the quest. Time is unmeasured. Not yet learner/facilitator validated. Choose Free exploration or pause the quest. Keep the answer key covered before each response.

  1. Open Neuron; choose Try my own threshold experiment. For this worked path, leave A=1, B=0, threshold=2. Predict output and explain before revealing.
  2. Compare the sum with the threshold. Revise the same configuration if needed; preserve the first prediction. Reaching the threshold, including equality, turns the output on.
  3. Choose the changed challenge. Make a fresh prediction using its displayed inputs and threshold. Explain what changed and what rule stayed fixed.

First output prediction + why · code ______

Observed calculation + revision

Changed inputs/threshold: prediction → calculation → explanation

Write or scribe exact words. Access support ______ Content hints ______ Save each prediction before revealing; no speech recording is required.

Cover until the relevant response is saved

Worked answers. First case: 1+0=1<2, output 0. From this untouched starting case, the changed challenge is A=0, B=1, threshold=1: 0+1=1≥1, output 1. Equality is enough. Other chosen starting configurations produce their own changed challenge; calculate the displayed case. No weights update, learning or biological simulation is shown.

Prompt: “Did the switch learn, or did we select a different fixed rule?”

Manual explanation: 0 unsupported; 1 correct output without full comparison; 2 shows each sum/threshold comparison and explains the fixed-rule boundary. Prose is not automatically graded; completion alone is not mastery.