Step 4: tool results

add_argument

{
  "name": "The argument from the algorithmic-task experiments",
  "stance": "for",
  "claim_id": "a4bfe1a1-0d33-4876-b6ab-40989cb38bac",
  "description": "Iterated Amplification learned toy algorithmic tasks from roughly a thousand times fewer decomposition examples than supervised learning needed of full solutions; the gap is extrapolated to realistic tasks."
}

{"success":true,"message":"Created argument \"The argument from the algorithmic-task experiments\" (for) on claim a4bfe1a1-0d33-4876-b6ab-40989cb38bac. Attach its subclaim edges, then give it a written form with write_argument, and evaluate it with evaluate_argument when you record the claim's assessment.","argument_id":"aabfd608-54bf-4a9e-a901-87a3c94c8108"}

add_argument

{
  "name": "The argument that decomposing is strictly easier than solving",
  "stance": "for",
  "claim_id": "a4bfe1a1-0d33-4876-b6ab-40989cb38bac",
  "description": "Producing a decomposition with access to subtask solvers is strictly easier than solving the whole task unaided, so it should need less supervision to learn."
}

{"success":true,"message":"Created argument \"The argument that decomposing is strictly easier than solving\" (for) on claim a4bfe1a1-0d33-4876-b6ab-40989cb38bac. Attach its subclaim edges, then give it a written form with write_argument, and evaluate it with evaluate_argument when you record the claim's assessment.","argument_id":"920417c3-d150-4426-bfe8-9fb9add56d4c"}

add_argument

{
  "name": "The argument from the difficulty of decomposing realistic tasks",
  "stance": "against",
  "claim_id": "a4bfe1a1-0d33-4876-b6ab-40989cb38bac",
  "description": "Human experiments on factored cognition found realistic reasoning tasks hard to decompose into small, locally checkable steps, so decompositions may be costly to demonstrate."
}

{"success":true,"message":"Created argument \"The argument from the difficulty of decomposing realistic tasks\" (against) on claim a4bfe1a1-0d33-4876-b6ab-40989cb38bac. Attach its subclaim edges, then give it a written form with write_argument, and evaluate it with evaluate_argument when you record the claim's assessment.","argument_id":"26705e2a-1bae-4ef2-af33-3eb5852f67b2"}