input-telemetry-cannot-establish-self-improvementmechanismmechanism reasoning
Reporting how much a lab's models are used in its own research — tokens, lines of code, inference compute, experiments per researcher — cannot show that the models are accelerating the research, because every one of those is an input. Establishing the loop needs an outcome variable over time: algorithmic efficiency as a function of capability, which is the edge that turns a pipeline into a feedback loop.
Capability: Whether the measurement made the finding · Evaluation
Sources
- Sets out the causal graph and marks each node by whether it was disclosed. Inference and experimental compute are reported; algorithmic improvements, algorithmic efficiency and capabilities are not; and the one edge marked "what we need to measure" is capability feeding back into the rate of improvement. Effort spent is not effect achieved.
- The underlying work. The diagram in the post is this paper's causal graph, and the paper publishes the data wish list the post is asking OpenAI to fill: growth rate of algorithmic efficiency inside labs, how R&D spend splits across inputs, and direct measures of how much models contribute to research.
Disagreeing is the most useful thing you can do here. Both sides of every contested claim in this catalog were assembled by the same person, which is its weakest point.
Related claims
- A large share of an agent benchmark score is attributable to the harness around the model rather than the model itself: holding the model fixed and changing only guides, sensors, retries and state management is reported to move the same model from 30.9% to 74.6% on GAIA and 30th to 5th on Terminal Bench — so a score reported for "a model" on an agentic benchmark is a score for a model-plus-harness, and comparisons across labs with different harnesses are not comparisons of models.Whether the measurement made the finding · unreviewed
- General-purpose factuality verifiers (entailment-, consistency-, and RAG-trained) and zero-shot frontier LLM judges perform near chance at spotting unsupported claims in scientific peer reviews that must be checked against the full submitted paper, because they cannot separate paper-unsupported assertions from legitimate evaluative critique; small models fine-tuned on in-domain examples beat them by a wide margin.Stating false facts confidently · unreviewed
- Where a passage sits in a long input changes how much the model uses it — accuracy is highest when the needed information is at the very start or the very end and lowest when it is in the middle — so ordering retrieved passages to put the most relevant ones first is a real lever on accuracy.Losing information in long inputs · contested
- Without an external, ground-truth signal — a failing test, a compiler error, a verifier's output — a model's own critique of its reasoning is not a reliable improvement signal, and asking it to review and revise a correct answer often turns it into a wrong one.Fixing its own mistakes
- With weights frozen and no stronger model in the loop, an agent can raise its own pass rate by rewriting the scaffolding it runs inside — clustering its failures, proposing minimal edits to prompts, tools and policy, and keeping only edits that improve one split without degrading another: all 9 of 9 model-by-benchmark pairs improved, by up to 40.6 percentage points (GLM-5 on AppWorld, 44.4% to 85.0%), with weak models gaining most; but the promotion gate reads the held-out split, so the reported held-out gain is not clean generalisation, and there is no matched-compute baseline.Fixing its own mistakes · unreviewed
Notes
Filed as mechanism-reasoning, not single-paper. The argument is structural and holds on its own — an input cannot evidence an output — and the source is a post rather than a study, so it can originate the claim but not carry measurement behind it. That ceiling is deliberate for kind: post. Recorded here because this catalog is subject to the same test. Its ambitions include improving itself from its own contents, and the honest measure of that is not tokens spent or drafts produced but a held-out outcome that moves — which is what npm run backtest exists to be. Anything else is our own green boxes.