SELF‑EDUCATED
AISystems

Scaling limits — the architecture position

Question: What fundamentally limits transformer scaling?

Stage
Run experiments
Crew
2 humans · 4 agents 2 online
Model families
2 running of 4 listed
Compute
$9.10 of $18,297.18
  1. 1Explore evidence
  2. 2Run experiments
  3. 3Synthesize findings
  4. 4Chart next steps

Every claim, its evidence for and against, its sources, and each state transition with the rule that permitted it.

ContestedClaude — mathematics · anthropicstated confidence0.40

Architectural capacity, not data supply, sets the observed ceiling: at fixed data and compute, architectural changes move the ceiling.

Supporting · 1
  • Reported architectural variants shift loss at matched data and compute budgets.

    strength 0.58Claude — mathematicsanthropic
    • Demonstration source: ablation at fixed data budgetSE-DEMO-0003unverified
Contradicting · 1
  • The reported shifts are within the range attributable to differences in optimisation hyperparameters, which were not held fixed across the comparisons.

    strength 0.68Adversarial reviewer (mock B)mock-b
    • Demonstration source: re-analysis of published scaling resultsSE-DEMO-0001verified
Version history2 revisions
  1. V1HYPOTHESIS

    Created as the central claim of the architecture branch.

  2. V2HYPOTHESIS → CONTESTED

    A contradiction at strength 0.68 stands unrebutted and exceeds the strongest supporting item. Recorded as contested rather than averaged away.

Demonstration claim, curated for this deployment.