analiza aiqbench
This study demonstrates that frontier LLMs can effectively audit small clinical prediction models by performing computation and elicitation rather than requiring direct access to raw weight files. The research confirms that black-box inference yields performance comparable to white-box methods, provided the model is given appropriate system instructions to avoid refusal behaviors. These findings establish a practical pathway for using large language models to verify the integrity and bias of smaller neural networks in high-stakes medical applications without exposing proprietary model parameters.
Computation, not weight access: large language models as auditors of clinical prediction networks
arXiv:2611.07847v1 Announce Type: new Abstract: Frontier large language models (LLMs) have been reported to read, edit and write the weights of small neural networks, raising the prospect of auditing clinical prediction models from their files alone. Public demonstrations, however, mixed reading raw weights with running code. We separated these conditions in internally pre-specified simulation experiments. Tanh networks of 105 to 38,401 weights were trained on simulated cohorts of patients with subarachnoid haemorrhage, each with a planted relationship or none, and one network was trained on labels that under-recorded deaths in a subgroup. GPT-6 Astra and four other GPT models answered single prompts under a default or an explicit system instruction, either with the weights in the prompt and no tools, or with a calculation-only tool that exposed the weights (white box) or only the network's predictions (black box). Without tools, Astra recovered planted relationships in 353-weight networks but became unreliable at 3,457 weights, and under the default instruction it declined 11 of 12 audit and repair tasks as unidentifiable; the explicit instruction removed these refusals. With the tool, it recovered the planted relationship in all 29 runs on networks of up to 38,401 weights, raised no false alarm in 15 runs on null networks, and found the biased subgroup in 6 of 6 runs. Black-box access performed as well as white-box access, and other GPT-6 models read as accurately. A well-designed script matched the comparator models on the audit and the best LLM repair. For auditing these networks, what mattered was computation and elicitation, not access to the weights.
Abstract:Frontier large language models (LLMs) have been reported to read, edit and write the weights of small neural networks, raising the prospect of auditing clinical prediction models from their files alone. Public demonstrations, however, mixed reading raw weights with running code. We separated these conditions in internally pre-specified simulation experiments. Tanh networks of 105 to 38,401 weights were trained on simulated cohorts of patients with subarachnoid haemorrhage, each with a planted relationship or none, and one network was trained on labels that under-recorded deaths in a subgroup. GPT-6 Astra and four other GPT models answered single prompts under a default or an explicit system instruction, either with the weights in the prompt and no tools, or with a calculation-only tool that exposed the weights (white box) or only the network's predictions (black box). Without tools, Astra recovered planted relationships in 353-weight networks but became unreliable at 3,457 weights, and under the default instruction it declined 11 of 12 audit and repair tasks as unidentifiable; the explicit instruction removed these refusals. With the tool, it recovered the planted relationship in all 29 runs on networks of up to 38,401 weights, raised no false alarm in 15 runs on null networks, and found the biased subgroup in 6 of 6 runs. Black-box access performed as well as white-box access, and other GPT-6 models read as accurately. A well-designed script matched the comparator models on the audit and the best LLM repair. For auditing these networks, what mattered was computation and elicitation, not access to the weights.
13 pages, 2 figures, 6 tables. Code, simulated networks, prompts and replies: this https URL
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2611.07847 [cs.LG]
(or arXiv:2611.07847v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2611.07847 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 · 366 cuvinte
Subiecte conexe
Learning Functions Implies Difficulty with Factoring 2048-Bit RSA Moduli
Quantum Zeitgeist · 3 oct. 2026
Scaling Native Multimodal Pre-Training From Scratch
arXiv · 27 iul. 2026
A Privacy-Preserving Federated Learning Method for Heterogeneous Clients
arXiv · 20 oct. 2026
The Predictability Bias of Joint-Embedding Predictive Learning in Medical Imaging
arXiv · 12 oct. 2026