@kj
One of the most frustrating things about AI health tools right now:
Who is tracking whether they actually work? With an increasing number of users turning to LLMs for health questions there is no real follow up.
There is the thumbs up 👍/down 👎 a user can click after a ChatGPT response but… not sure what that gets us in terms of health related questions and outcomes.
Sycophant behavior where model tends to agree and mirrors a user’s perspective (even when wrong) only adds to the issue.
Medical research, especially in orthopedics, turns to outcome measures. There are validated instruments for measuring patient outcomes such as: PROMIS scales.
Standardized. ✅
Research-grade. ✅
Used in clinical trials. ✅
Almost no consumer health AI is collecting this data. Which means almost no consumer health AI is actually learning…?🧐🤔
0 reply
0 recast
0 reaction