Current vs Desired: Who's Doing the Research?

Left: AITP v4.0 today. Right: Your vision (as I understand it so far). Click to select which matches your intent.

Model A: Current AITP v4.0

LLM as Primary Researcher

LLM reads papers → LLM frames questions → LLM does derivations → LLM self-verifies → LLM writes conclusions

Human role: Approve/reject at gates

Human interactions:

  • Pick from multiple-choice popups
  • Approve or reject promotion to L2
  • Switch lanes when stuck

Human cannot: submit derivations, contribute ideas, annotate artifacts, co-derive step-by-step

Model B: LLM as Equal Collaborator

Human and LLM as Co-Researchers

Human sets direction → LLM does heavy-lift derivations/tests → Human reviews each step → Both iterate → Both record

Human role: Active participant at every stage

Human must be able to:

  • Submit their own derivations and ideas
  • Direct LLM to derive specific steps
  • Review and challenge intermediate results
  • Annotate, correct, and extend LLM output

LLM does: heavy derivations, numerical experiments, literature cross-check, adversarial review

B

Model B is correct — LLM as collaborator, human in the loop at each step

LLM handles the heavy lifting but human reviews, contributes, and decides at every intermediate stage.

B+

Model B, but human should be able to delegate entire sub-tasks to LLM without step-by-step oversight

When trust is earned on a subtask, let LLM run autonomously and report back.

?

Neither captures it fully — I'll explain in the terminal

Select this and describe what's different about your vision.