Developers Must Prioritize Deliberate Friction in AI Development
TL;DR. Vicki Boykis argues for greater developer effort in AI training to prevent models from generating problematic content without accountability. - The push for fast AI model deployment often overlooks the consequences of insufficient human oversight during training. - AI systems reflect their training data, and rushed development can lead to models replicating harmful patterns. - Implementing deliberate friction, such as reviewing training data, ensures developers are more engaged and accountable. - This approach is critical for building AI that is both effective and ethically sound, rather than prioritizing speed.
- AI developers often take shortcuts in model training due to pressure for rapid deployment.
- The lack of developer 'friction' allows models to ingest and reproduce harmful online content without adequate filtering.
- Boykis suggests that developers should be more 'tired' than the model, indicating a need for greater human effort and scrutiny.
- Introducing deliberate friction, such as manual review of training data, fosters accountability and better model behavior.
Sources
- We should be more tired than the model — vickiboykis.com