Category: AI

My AI Physical Therapist

~ Sasha ~ 2 Comments

My Instagram feed is basically:

  • The 10 people I follow (including my son’s ceramics account)
  • Dogs / puppies / rescues
  • Tennis / squash tips and highlights
  • Injury prevention / cures for middle-aged athletes

On the last point, my last few years have been marked by a series of minor but chronic injuries. I seem to be in the stage of life of small ligament and tendon strains, injuries that respond well to physical therapy if you can make the time.

The worst has been tennis elbow (lateral epicondylitis). It’s lasted for more than a year, and I’m just now getting better. But I’m finding PT too time-consuming (90-120 minutes so I can do 20 minutes of exercises??) and too much of an insurance hassle.

In a continuation of this trend, a month ago, I got a sharp pain in my hip after a run. After a few dead-end YouTube attempts, I realized I didn’t know what was wrong with me.

Rather than go to the doctor and then back to PT, I turned to ChatGPT. I was amazed with how helpful it was. If you, like me, are still under-using these tools, perhaps this can be an inspiration.

Here’s the original prompt I wrote:

I need you to be an expert physical therapist who can help me diagnose and treat injuries at home. Be thorough in your diagnosis and assume that I am a sophisticated lifetime athlete who can do complex exercises at home as needed. I’m 51 years old. I need to diagnose a pain I have in my right outer hip, just below my hip pointer.

I got the injury after a fast (for me) hilly run after having taken off a few months. This was my 10th run after having taken 3 months off. I had done the same run, 4.3 miles, twice earlier in the week, but my pace in those was 9:30/mile and in this run was 8:20/mile. I felt no pain during the run but after the run it hurt some to stride / step with my right leg, and to go up hills. The pain is greatest when I get up after sitting down for a bit. After a little while it decreases. The pain is sharp and localized to the area just below and above the hip pointer.

From this prompt and a few follow ups, ChatGPT gave me:

  • Its first diagnosis (either a tendinopathy in my gluteus medius / minimus or a TFL overuse injury)
  • A series of diagnostic tests I could do to narrow down the potential injury
  • A full 30 day PT plan that I now have in a Google Sheet that I’m following daily
  • This plan includes descriptions of all the exercises with references.

Since this went so well, I’ve kept talking to ChatGPT about how it’s going: how the exercises, felt; how one of them hurt more than expected and could it give me an alternative; how I feel first thing in the morning and over the course of my day. It’s the conversation I would have with my PT every day if he were available.

With this feedback, ChatGPT both modified the exercise regime to better suit how I’m feeling and has modified the diagnosis.

Here’s the latest that I got back this morning, based on my updated symptoms:

Differential Diagnosis Recap

Condition Why It Fits Why It Doesn’t
Gluteus med/min tendinopathy Pain with heel strike, long stride, compression from sleep, improves with warm-up —
TFL overuse Could be involved, but unlikely primary now; less direct heel-strike involvement Less pain with bridge, more lateral than anterior
Trochanteric bursitis Possible secondary irritant But no diffuse swelling or tenderness reported yet
Labral or joint issues No locking, groin pain, or deep ache Very unlikely
Femoral stress reaction No deep bone pain or night pain Very unlikely

And here’s a visual representation of what I’ve told my AI physical therapist and what that means for my diagnosis—this pretty much blew me away:

I know that for many people, AI is an idea—perhaps even a scary one—and not something that’s part of daily life. And I can understand why many of us might be happier either using none of these tools or waiting for someone to build a “Virtual Physical Therapist” app rather than playing around with ChatGPT or the many other tools available to us today.

I’d encourage you to think about this another way: for $20 / month, an expert in literally any topic you can accurately describe is available to you and to your team.

It’s a shame not to take advantage of that.

Is “Polished” Writing a Good Thing?

~ Sasha ~ Leave a comment

How we communicate evolves with time and with the medium.

I write my texts (mostly) like I write my emails, resulting in my kids repeatedly telling me that it sounds “aggressive” when I put a period at the end of a text message.

(They also want me to use exclamation points much much more! LOL)

While I’m comfortable with the idea of tone and style evolving over time and in different contexts, I’m uncomfortable with what happens when we no longer need to struggle with a blank page. I’m skeptical that it’s a good thing that Gmail is now offering to “polish” my posts and that LinkedIn suggests “rewrite with AI” every time I string a few words together.

Clear writing and clear thinking co-evolve: I don’t know anyone who writes well who doesn’t think well; and how we express our thoughts in written form is a great way to reveal whether our thinking is as clear as it needs to be. I also know that convenience will win out—why wouldn’t it?—and that the cost of all of this convenience will be mostly invisible.

It’s already established that AI is most useful when you have subject matter expertise, so you can tell the difference between good and bad, and use these tools as leverage for your strengths.

How do we avoid systematically undertraining ourselves as strong writers and strong thinkers, to use the tools without having them replace an activity that sharpens our mind?


 

Use AI to do Just One Job

~ Sasha ~ 1 Comment

Are you curious about AI but still a bit skeptical?  Then this post is for you.

First, a test: Have you found at least one important task that you can now do better or faster thanks to AI?

Yes! Hooray.

No? Keep reading.

To put the “no” into context, you are, in effect, making an active choice not to use a major productivity enhancer. It’s like insisting to do your calculations by hand instead of Excel, or writing by hand instead of using Word or Google docs.

If that choice is essential to your creative process, then I encourage you to stick with it.

But there is definitely somewhere that you should be using AI.

If you’re not using AI, you may have read about it or tried ChatGPT, but it’s not part of your daily workflow.

The road from here to there is short, just follow these steps:

  1. Pick a task that you do often, that you understand well, and for which you are a good judge of what good quality looks like.
  2. Log in to one of the new AI models. I would suggest you pick ChatGPT-4 (free) or Claude 3 Opus (I pay $20/month for this).
  3. Play around for 1-2 hours, working to get the AI to help you do your chosen task.

A mindset shift around AI might help, and what helped me the most was this How Should I Be Using AI Right Now podcast by Ezra Klein.

My takeaways from the podcast were:

  • Anthropomorphize the AI. Think of it like a person you’re working with, not like a computer. More specifically, think of the AI like an intern who’s a couple of years out of college who is very capable but who needs (and is open to) a lot of feedback.
  • Tell the AI who it is / who you want it to be. Meaning, give it a personality. Tell it to be Steve Jobs, or tell it the characteristics of your college advisor or best friend or editor. When you tell the AI what personality to have, it responds with the right tone and syntax and, more surprisingly, you are changing the quality of the responses.
  • Give the AI examples of what ‘good’ looks like. Share examples of the work product that meets the standard you want the AI to hit. Better yet, share 10 of those. I find copy/pasting text the easiest way to do this. Explain what each example is and how representative it is of the generalized output you’d like to see. The more you can share, the better.
  • Give the AI tons of feedback on its output. This is where the “helpful intern” mindset helps. The AI needs to be told what to do, so tell it! Here are some examples from when I was playing with teaching the AI to write a good follow-up email from a sales call. (aside: the AI is reading what I write to decide how it will write. So I’m intentional about writing like I write.)
    • “It’s pretty good. The opening is a bit generic and I’m always trying to avoid more general statements, so please tone that down a bit and stick to specifics.”
    • “You’re getting there. Let’s leave that as a placeholder for now. It’s a good start for the more standard emails that I write. But sometimes I’m more open ended.”
    • “OK a few things – avoid unnecessary modifiers like “truly believe” and try to avoid repeating words (you used “believe” twice at the start of two paragraphs).”
    • “This sentence is terrible ‘Bringing together diverse perspectives and expertise will be critical in unlocking innovative solutions.’ It is a statement full of platitudes. Avoid writing like this always.”
    • “Yup. Better. But this sentence is still full of platitudes. ‘and have been impressed by the insights and leadership you’ve brought to the space.’ What would be better is either to delete it or to find something relevant to refer to that is more specific. Also on the prior example, ‘exciting opportunity’ is breathy and doesn’t need the word ‘exciting.’ Don’t be afraid to be direct and not too flowery.”
    • It finally produced something decent, so I wrote, ‘Good. This works as a point of reference, stylistically.’ I named that style so I could refer back to it in the future.

I expect you’re getting the feel for this…I gave more feedback on various iterations, either at the general level (“that is waaaay too wordy. Try again but cut 80% of the words”) or much more specific (“This sentence is terrible, the whole second half of it is a word salad that adds no value.”). Again, the mindset of “I’m talking to an intern” really helped me stick with it here.

After about an hour of this back-and-forth, the AI was giving me what I wanted at a high standard, and this standard got me an 80% first draft that will save me tons of time.

In addition, working on a real task—one that is important, and where I have domain expertise—helps me learn about what the AI can and cannot do well.

And the impact of that is huge. It’s like crossing a threshold as I imagine the amount of leverage each of the 120 people on my team could have if this becomes part of their workflow. What if each person could have a “very helpful and eager intern” at their beck and call? Imagine the impact of that, and add to it the operational stack that we can hack to pieces with the aid of this technology.

I hope this is the post that nudges you to take another shot at this, and that, in a weeks’ time, you’ll have played with and succeeded at getting AI to do SOMETHING meaningfully helpful in your professional or personal life.

If you need more ideas of where to start, here they are:

  • Writing your professional bio
  • Creating meeting summaries
  • Writing follow-up emails from sales calls
  • Resume screening to avoid bias
  • Creating a negotiation training module
  • Helping you prioritize your to do list
  • Summarize this data set
  • etc. etc.

Don’t forget: the assisted here is the point.

I generated that list with help from AI.

It gave me two lists of 10 ideas. All of them were too wordy. I liked 5 of the ideas, and I made those 5 better, added links, and I added two of my own

Voila.

(p.s. more On Interviewing Well posts are in the queue. Stay tuned.)