AI TrackerFrontier desk · AEST
Back to news
normalaitech

Weight-space programming for auditing clinical decision networks

arXiv cs.AI · 4 Nov 2026, 04:41 AEST

A study explored the use of weight-space programming to audit and repair clinical decision networks in radiotherapy. The findings suggest that this method can identify and fix flaws in small networks but cannot certify their correctness.

Key facts

Read aloud

A study explored the use of weight-space programming to audit and repair clinical decision networks in radiotherapy. The findings suggest that this method can identify and fix flaws in small networks but cannot certify their correctness. The purpose of this study was to determine whether such a model, given only a clinical protocol and a network's weights, can audit and repair the network. Claude Opus five point five audited each network blind, without tools, data or code execution, and every claim and repair was tested on about four hundred twenty thousand requests. The language model named the broken rule in sixteen of eighteen flawed networks ninety five percent CI sixty seven percent to ninety seven percent, including all fifteen with a confound or a threshold shift. It cleared two of five healthy networks without being forced and found one of three triggers. The results indicate that, in small networks, weight-space programming can locate and repair flaws that silence an input or shift a threshold, but cannot yet certify that a network is correct.

Sources in this cluster