analiza aiqbench
This arXiv preprint tests whether transcripts from saturated benchmarks, meaning items that models already solve near-perfectly, still carry useful signal about model capability. Pairwise judging of transcripts performed best in the authors' retrospective simulations, cutting prediction error 5.3 times versus using saturated scores alone. The method also underestimated early-2025 reasoning models, so it doesn't yet generalise reliably across distribution shifts. For those who evaluate frontier models or write safety cases, this points to transcript analysis as a way to extend benchmark lifespans, though it needs more robust methods before it can be trusted.
Evaluating Model Capabilities After Benchmark Saturation
arXiv:2610.20667v1 Announce Type: new Abstract: Rapid advances in AI capabilities are causing benchmarks to saturate. Once several models approach perfect scores, a benchmark loses its power to differentiate between them, and with it its utility for informing AI safety cases and frontier policies. Yet benchmark items produce far more information than the scores we typically extract from them: a full transcript of how each model solved the task. We investigate whether these transcripts can be used to assess model capabilities even after benchmark scores have saturated. We simulate saturation retrospectively using historical benchmark and model releases, and test whether protocols that access only correctly solved transcripts from saturated benchmarks can predict capabilities measured on held-out, unsaturated items. We evaluate representative contemporary approaches for extracting information from model transcripts: pairwise judging, single-transcript absolute scoring, and predictions from deterministic transcript features or embeddings. We find that saturated transcripts retain capability information beyond benchmark accuracy. Amongst our evaluated protocols, pairwise judging performs best, reducing prediction error by 5.3 times compared to a baseline using only saturated benchmark scores. This capability-relevant information is distributed across transcript sections, but concentrated in specific channels of information such as reasoning and tool-calls. However, the predictive power of protocols does not always generalise across distribution shifts -- for example by underestimating the capabilities of the first reasoning models introduced in early 2025. Saturated benchmarks thus retain capability-predicting signal in their task transcripts, but extending the lifetime of existing benchmarks requires methods designed to extract these signals robustly.
Abstract:Rapid advances in AI capabilities are causing benchmarks to saturate. Once several models approach perfect scores, a benchmark loses its power to differentiate between them, and with it its utility for informing AI safety cases and frontier policies. Yet benchmark items produce far more information than the scores we typically extract from them: a full transcript of how each model solved the task. We investigate whether these transcripts can be used to assess model capabilities even after benchmark scores have saturated. We simulate saturation retrospectively using historical benchmark and model releases, and test whether protocols that access only correctly solved transcripts from saturated benchmarks can predict capabilities measured on held-out, unsaturated items. We evaluate representative contemporary approaches for extracting information from model transcripts: pairwise judging, single-transcript absolute scoring, and predictions from deterministic transcript features or embeddings. We find that saturated transcripts retain capability information beyond benchmark accuracy. Amongst our evaluated protocols, pairwise judging performs best, reducing prediction error by 5.3 times compared to a baseline using only saturated benchmark scores. This capability-relevant information is distributed across transcript sections, but concentrated in specific channels of information such as reasoning and tool-calls. However, the predictive power of protocols does not always generalise across distribution shifts -- for example by underestimating the capabilities of the first reasoning models introduced in early 2025. Saturated benchmarks thus retain capability-predicting signal in their task transcripts, but extending the lifetime of existing benchmarks requires methods designed to extract these signals robustly.
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2610.20667 [cs.AI]
(or arXiv:2610.20667v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.20667 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 341 cuvinte
Subiecte conexe
"About a quarter of them contained bugs": Gemini 4's ML R&D benchmark
LessWrong · 16 oct. 2026
[AINews] not much happened today
Latent Space · 3 oct. 2026
How to Evaluate General-Purpose Robot Policies for Real-World Deployment
NVIDIA · 12 iul. 2026
A Score Is Not Understanding: toward a richer toolkit for model evaluations
LessWrong · 31 iul. 2026