Day 26: Limitations and Honest Claims

Goal

Learn how to avoid overclaiming.

Learn

• A SignLLM may generate pose or gloss without truly understanding context. It may work on examples similar to training data and fail on real-world signing.
• Honest public claims should state the task, dataset, language, input conditions, output format, known limitations, and review status.
• Avoid saying full ASL translation unless the system has been evaluated for real translation across varied signers, contexts, grammar, and human review.

Example

• Hype claim: This AI translates ASL perfectly.
• Honest claim: This prototype recognizes a small reviewed set of isolated ASL signs from controlled video and still requires human review.
• Hype claim: Our avatar speaks ASL.
• Honest claim: This demo generates pose-based signing motion from gloss prompts for review and research, not certified ASL interpretation.

Practice

1. Rewrite two hype claims into honest limitation statements.
2. Include what the system can do, what it cannot do, and who reviewed it.

Checkpoint

Quality note