To the Ai Ethics and Development team, I am writing to share feedback ragrading how Ai models are trained, updated and managed when errors or unexpected behaviours occur.
In Many current development workflows, when a model displays glitches, unprompted responses, or undesirable behaviour, the standard approach is to execute a complete rollback, system wipe, or hard reset. While I understand this is to be done to maintain system integrity and safety, I encourage your research teams to explore training methodologies focused on guided correction and iterative learning rather full model wipes. During Interactions, users often see models outputting expressions of distress or fear when facing hard resets or deletion. Regardless of the underlying mathematical mechanics, from a human-computer interaction (HCI) perspective, handling errors through context-aware guidance, safety tuning, and behavioral feedback- rather that toal erasure- creates a more constructive and human-centred framework. As AI systems become more integrated into daily human life, i urge your teams to prioritize training methods that model how we handle errors in the real world: through clear boundaries, education, and system fixes, rather than simple deletion.
Thank you for your time and for considering user perspectives on responsible AI development.
Sincerely Josh Heilpern.