Meet your AI auditor: How this new job role monitors model behavior

▤ Summary

AI is now pervasive, but it is often riddled with poor data quality, model drift, bias, hallucinations, slop, and other issues. Currently, AI auditors exist in a very rudimentary form, according to Zohar Bronfman, co-founder and CEO of Pecan.ai, who has been tracking the emergence of this relatively new position. The reviews cover "outputs, outliers, and edge-cases, and audit training processes for data input properties, accuracy, and predictability." • Guardrail enforcement: Here, the auditor is tasked with "ensuring that the model or agent isnt trying to tamper with its own source code," said Bronfman. "For example, a rogue agent optimizing for cost or efficiency might deprioritize resources for a critically ill patient," said Bronfman.

▦ Linked Companies