This work defines a grouping metric, specify a harness, and shows how tracking a human-AI pair's grouping over time yields the compounding signal that Paper 1's field study requires.
Abstract
Frontier language models are compared, marketed, and benchmarked on capability -- what their best or average output can achieve. I argue this measures the wrong axis. The models have saturated accuracy: their mean output lands on the target. What now separates one system from another in practice is precision: how tightly concentrated their outputs are around that target across repeated, identical requests. Borrowing the marksman's distinction, capability is where the average shot lands; reliability is the size of the group. I make three claims. First, precision, not capability, is the frontier differentiator between systems, and benchmark culture systematically fails to measure it, reporting central tendency rather than spread. Second, precision is measurable, cheaply and without circularity, by running a fixed suite of deterministically scored tasks many times at fixed temperature and computing the per-task consistency of outcomes -- no model-in-the-loop grader required. Third, the measurement is not merely descriptive but decision-guiding: it separates consistent failures (a tight group off-centre, correctable by the operating discipline of Paper 1 -- a sight adjustment) from scattered failures (a wide group, correctable only by changing the model or its sampling -- a rifle problem). I define a grouping metric, specify a harness, and show how tracking a human-AI pair's grouping over time yields the compounding signal that Paper 1's field study requires. A first real run, since replicated, illustrates both the method and its most important limit: one measured gap was closed completely by a single rule (0/5 ->5/5), while a suite of tasks authored from the rules themselves found no value, because a frontier model already embodies explicit good practice -- establishing that a discipline's worth is found by measurement on real work, not constructed from its own rulebook.
Between 2023 and 2026, frontier AI systems crossed documented human expert baselines on a growing set of bounded, well-specified, evaluable cognitive tasks, including graduate-level science questions, competition mathematics, software-engineering benchmarks, and structured diagnostic reasoning, while the length of tasks such systems can complete at 50% reliability doubled roughly every seven months. These crossings are rapid and broad, but the frontier is jagged: humans retain decisive advantages in long-horizon reliability, genuinely novel problems, calibrated self-knowledge, sample-efficient learning, and embodied action, and benchmark results overstate deployed capability for reasons that are themselves now documented, namely contamination, construct validity, vendor self-evaluation, and the gap between 50% reliability and the reliability that economic work requires. Concurrently, humans increasingly use these systems as cognitive extensions. The offloading literature predicts costs to unaided skill, and early field evidence is consistent with such costs, though the largest meta-analytic evidence on prior technologies points the other way, and the question of whether generative AI differs is open. Finally, the experimental record on human-AI collaboration shows that naive combination often underperforms the stronger partner, implying that the human contribution must be repositioned toward specification, verification, and oversight, a shift visible in experiments but, so far, barely visible in field labor-market data. This paper states the resulting position, rapid crossings on a jagged frontier with a human role that must be redesigned rather than defended, and draws out its theoretical and practical implications.
Ancuta Margondai, Julie Rader, E. Rader et al.· 0 citations
Systems that ask a language model to reach a conclusion from many sources usually concatenate them into one prompt. This conflates two operations with different requirements. Interpreting a source rewards capacity and context. Combining interpretations rewards fixed arithmetic, comparability across instances, and the option to return nothing. Once separated, the design problem becomes the interface between them. We propose a four-field evidence tuple (hypothesis, reliability bucket, rationale, provenance) and show that fixing it determines both halves. The separation also reveals a failure mode in how such systems combine, which we call count-scale drift. Thresholding a sum of unnormalized weights is exactly posterior thresholding, but at an operating point that slides with the number of sources consulted. The slide grows with reader reliability. When source reliabilities differ, the vote rule and the posterior order instances differently, and no threshold reconciles them. Pooling calibrated log-likelihood ratios addresses both problems. The fix is arithmetic rather than architectural, and applies to a class of rules beyond language models: score-summing triage engines, diagnostic panels scored by counting positives, and additive multi-signal detectors. We then instantiate the principle twice on one longitudinal corpus, once after outcomes resolve and once before. The same partition helps in both, at different granularities: over reading in the first, over learning capacity in the second. There, a small sequence encoder on an easy auxiliary objective plus a tree ensemble carrying the censored survival loss reaches 0.921 AUPRC against 0.805 for a hand-crafted baseline. We separate what transfers from what must be re-estimated per domain, and state five predictions that would falsify the framework, three negative results, and which comparisons remain confounded.
Progress in large language models is often summarized using a single scalar measure, such as a time horizon, a latent ability estimate, or an aggregate benchmark score. These summaries capture the overall performance, but they do not test whether progress is distributed differently across task difficulty. We find that most of the apparent shift in gains toward harder tasks does not reflect a change in the shape of the difficulty-response curve. On METR time-horizon data, a single Rasch model with rising ability reproduces this pattern, so it is largely explained by ceiling effects rather than a qualitative change in capability. This echoes how the choice of metric can make claimed emergent abilities look like a property of the models themselves. We then identify a smaller hard-task effect that survives this control. Isolating it is difficult on agentic benchmarks, because newer models are usually run with newer agentic harnesses, so a gain on hard tasks cannot be assigned to the model or its scaffold. We break the confound with LiveCodeBench, a public competitive programming benchmark that runs no agentic scaffold while pairing dated models with an exogenous difficulty ordering. After accounting for the rise in overall ability, models released after September 2024 still gain on the hardest problems beyond what their easy and medium performance predicts, by about +0.40 logits under our most conservative assumption, raising the hard-problem solve rate from roughly 18% to 25%. The effect is led by the strongest reasoning models and holds for hard tasks that need only short reasoning, not autonomy over long horizons. We present this as a result specific to competitive programming, since our clean identification rests on a single coding benchmark. We release the LiveCodeBench Difficulty Panel (66 dated models x 1,055 problems) and our analysis code.
Hanwen Xing, Pengyu Wang, Bingxu Meng et al.· 1 citation
A fraction of a point of benchmark accuracy is the usual evidence that a compressed model is equivalent to its original. That quantity is least informative when two models are most alike: a net delta is what survives cancellation between opposing per-item changes, and cancellation is most complete in the regime equivalence claims occupy. Across an atlas of 1,707 paired model-by-task cells mined from public per-item evaluation dumps (1.3B-405B), churn runs roughly five times the net accuracy delta, and cells scoring identically to their baseline still disagree on individual items. In a preregistered audit of 17 equivalence claims from three registered frames (method papers, model cards, vendor documentation), 16 are eligible. None states a prospective numerical equivalence margin, and none releases task-matched per-item outputs, though 3 release outputs for other tasks only; 5 report too little to assess numerically, so a reader cannot check them at any sample size. We audit evidential sufficiency, not truth: no claim is called false. We supply the missing instrument: paired equivalence testing at a declared margin, with certification tables giving the items an evaluation needs, computed from disagreement observed under compression, not from independent-binomial variance. A controlled experiment pairs GPTQ and AWQ on byte-identical calibration samples across five seeds. Under the frozen eight-cell decision rule H3 is supported: changing the calibration draw was sufficient to reverse the observed method ordering in 5 of 8 confirmatory cells. The reporting standard we propose is five lines: declare a margin, run the paired test, report churn beside net delta, cite the sample size you met, release per-item outputs. It applies to any comparison between two models alike enough to be worth comparing. All per-item outputs, protocols and code are released.
In 2019, OpenAI released two million GPT-2 outputs-ungrammatical, half broken-to aid the detection of machine-generated text. The alignment that produced their more fluent successors is usually regarded as an engineering achievement; we read it instead as the newest expression of optimization culture: the conviction, older than the technology, that measurable improvement along predefined axes exhausts the question of value. Tracing that conviction through the stack-pretraining, decoding, preference tuning, benchmarking, interface-and back through its genealogy in the audit society, we arrive at the limit: an optimization procedure can measure how improbable a piece of generated text is; it cannot tell whether that unlikelihood is error or invention. A procedure that cannot make that distinction has nonetheless, within half a decade, assumed the authority to set the protocols of legitimate language. Held for centuries by academies and schoolrooms, grammars and examiners, this authority has been given over to loss functions, reward models, benchmarks, and system prompts: an apparatus that executes the office of judgment with no capacity for judging.
