I'm one of the co-founders and practicing physician on the team.
AI is changing the nature of the patient-doctor relationship where patients are already using AI tools to research material and prepare before and after. Sidekick was purpose built to lean into new interaction dynamics where the goal is to support the patient fully in context of what their doctor says. We are definitely being very mindful of the doctor side and are working to set the right balance of content generated and build intersecting tooling for clinicians.
Don't forget pre-visit. Doctor visits go much better if the patient advocate (this thing) has questions in hand. The patient can spend hours talking to the tool and get a summary so they are better prepped for the 5-minute doctor conversation. A tool that summarizes observed symptoms for the doctor might be an idea.
Doctors are good with getting a shorthand note they can scan prior to appointment about updates, etc reverent to the appointment that day. My experience is that AI models even with expert curations cannot currently do well by itself and would waste a doctor’s time and could be misleading too. As such as noted by the parent comment, that note would need to be summary sheet that AI touches would have to list facts and context with patient review that it accurately would represents what the patient would say in their own words.
Pro tip: check out the NoHarm benchmark and tracking to see how the current models are generally trending on the medical frontier. Make sure to select Clinical not General:
https://www.arise-ai.org/mast
And if you look at some the data carefully as one of the models highlights on their blog, there’s concerning behavior for Fable on this front:
https://www.amboss.com/us/newsroom/noharm
Yes we've taken a look at this. In fact we had a pretty good test case from our teammate's real visit where the text to speech was mistranscribed (despite using a medical model) to the point where it looked like the doctor said to eat a topical steroid. However, because of that case and a few other complicated cases in our internal benchmarks, we've made the after visit prompt much more reliable, both identifying and cleaning up mistakes in the transcript, and marking points where you really ought to confirm with your doctor if it's unclear from context what it was actually supposed to be.
Please try it out and lmk what you think! We would love your feedback!
Note that our tool is not diagnosing or prescribing, simply noting what the doctor said and giving suggestions to the patient of useful questions to ask and contextual plain English definitions of medical jargon
Thanks for the thoughts! I’m one of the co-founders and a practicing physician. One of the big gaps with medical AI benchmarks like NoHARM, MAST is they are all primarily tuned for clinician reasoning and none of them look at the patient context and perspective. It's a huge gap right now.
Yes, actually that was our first experiment! We stripped it for this launch so we could more intelligently handle the cost and latency.
Every time I talk to a doctor they suggest we should do this. Honestly there's often enough time in the waiting room even if the patient doesn't do it at home lmao. Then just show the doctor a few bullets with quotes.
I’m sure doctors will love this. “My AI bot thinks I…”
I'm one of the co-founders and practicing physician on the team.
AI is changing the nature of the patient-doctor relationship where patients are already using AI tools to research material and prepare before and after. Sidekick was purpose built to lean into new interaction dynamics where the goal is to support the patient fully in context of what their doctor says. We are definitely being very mindful of the doctor side and are working to set the right balance of content generated and build intersecting tooling for clinicians.
>> balance between live help vs after-visit
Don't forget pre-visit. Doctor visits go much better if the patient advocate (this thing) has questions in hand. The patient can spend hours talking to the tool and get a summary so they are better prepped for the 5-minute doctor conversation. A tool that summarizes observed symptoms for the doctor might be an idea.
Doctors are good with getting a shorthand note they can scan prior to appointment about updates, etc reverent to the appointment that day. My experience is that AI models even with expert curations cannot currently do well by itself and would waste a doctor’s time and could be misleading too. As such as noted by the parent comment, that note would need to be summary sheet that AI touches would have to list facts and context with patient review that it accurately would represents what the patient would say in their own words.
Pro tip: check out the NoHarm benchmark and tracking to see how the current models are generally trending on the medical frontier. Make sure to select Clinical not General: https://www.arise-ai.org/mast
And if you look at some the data carefully as one of the models highlights on their blog, there’s concerning behavior for Fable on this front: https://www.amboss.com/us/newsroom/noharm
Yes we've taken a look at this. In fact we had a pretty good test case from our teammate's real visit where the text to speech was mistranscribed (despite using a medical model) to the point where it looked like the doctor said to eat a topical steroid. However, because of that case and a few other complicated cases in our internal benchmarks, we've made the after visit prompt much more reliable, both identifying and cleaning up mistakes in the transcript, and marking points where you really ought to confirm with your doctor if it's unclear from context what it was actually supposed to be.
Please try it out and lmk what you think! We would love your feedback!
Note that our tool is not diagnosing or prescribing, simply noting what the doctor said and giving suggestions to the patient of useful questions to ask and contextual plain English definitions of medical jargon
Thanks for the thoughts! I’m one of the co-founders and a practicing physician. One of the big gaps with medical AI benchmarks like NoHARM, MAST is they are all primarily tuned for clinician reasoning and none of them look at the patient context and perspective. It's a huge gap right now.
Yes, actually that was our first experiment! We stripped it for this launch so we could more intelligently handle the cost and latency.
Every time I talk to a doctor they suggest we should do this. Honestly there's often enough time in the waiting room even if the patient doesn't do it at home lmao. Then just show the doctor a few bullets with quotes.
Thoughts?