Table of contents
- Life Update
- Tools and Tips
- Stuff
Life Update
This week I was teaching.
Nothing unusual about that. I teach all the time.
A big part of teaching is sending your students to read. Open the book. Find the chapter. Pull the manuscript. Come back and tell me what it said.
This week I did something different.
I told them exactly what to type into their phones.
Not just into their phones. Into an app called OpenEvidence.
I gave them the diagnosis (seven words) and asked for the treatment plan.
Then I stood behind them and read the answer off the screen.
It was better than I expected. Much better.
I ran the same test again later, alone, without an audience. Same result.
I typed a few words. Very little explanation.
And I got a complete answer. With all the right references.
People are always asking me which AI tool I use for work.
And that reminded me it's been a while since we discussed it here.
So let's do this!
Number 166!
Tools and Tips
AI tools change every few weeks. There is no single best tool for everything, and anyone who tells you otherwise is selling something.
But at any given moment, there are usually one or two that are best at most of what you do.
So let's discuss the ones I use as a doctor and as a scientist.
🔵 The Companion That Wasn't
For months, Gemini was my default for almost everything.
The deep research function was the reason.
I could point it at a question and get back a literature review.
Sometimes better than the actual review I would have found on PubMed.
But what I really wanted was different.
I wanted someone to think with.
A partner.
A tool I can ask: "I have an idea," and the tool will push my idea somewhere I hadn't thought about.
Gemini could never do that.
It was like consulting with someone who has read EVERYTHING and has no idea what to do with it.
Enormous knowledge. Zero creativity.
For a while, that was simply where AI was.
So I used it for what it was good at.
Then Claude made a leap.
A huge leap.
Claude became closer to a real person. With signs of creativity.
And it kept getting better.
For months, I kept both subscriptions (Gemini and Claude) running side by side. And I kept giving Gemini another chance.
Gemini kept handing me the encyclopedia.
I canceled Gemini.
🔵 The Optimistic Colleague and the Pessimistic One
ChatGPT had a bad era.
It was the worst tool for months.
But it has gotten much, much better. It has passed Gemini and closed most of the gap on Claude.
So now I pay for two.
Why two?
Claude is more open-minded and more creative. It goes places with an idea instead of just cataloging it.
But when I hand Claude something I wrote and ask what's wrong with it, Claude wants to like it.
It finds the good.
Sometimes I don't need the good.
Sometimes I need someone whose entire job is to find the bad. A devil's advocate.
ChatGPT has the opposite problem.
Ask it the same question about the same paragraph, and it will find every flaw in the piece, plus several that aren't real.
Devil's advocate on steroids.
One is the optimistic colleague. The other is the pessimistic one.
So I give them both the same text. Claude tells me what it could become. ChatGPT tells me why it will fail. The truth sits between them.
That last part is the whole skill.
Neither of them is right on its own. You have to know enough to place the answer yourself.
🔵 Seven Words
​OpenEvidence has been good for a long time. In the last few months, it became something else entirely.
It used to require work. I had to explain the case, set the context, feed it background before it gave me anything worth having.
This week I typed a diagnosis. Seven words. No history, no framing, no setup.
What came back was everything you need to treat that patient.
I didn't really need it. I already knew the answer, which is exactly why it was a good test.
But if I were a first-year resident, that answer would have been more than enough to manage the case properly.
It was not like this just a few months ago.
It also gets things wrong.
Not totally wrong. Wrong the way an inexperienced doctor is wrong.
It picks the wrong item off the differential. It recommends a treatment option- but not necessarily the best one.
I catch it because I know the material, and because I open the references every single time. There are ALWAYS references. They are ALWAYS on point.
So even if OpenEvidence is wrong, the references will take you in the right direction.
That's the arrangement. It doesn't replace judgment. It replaces the hunting.
When I was a resident, I spent hours looking for information. Hours. I told myself the searching was teaching me something, and maybe some of it did.
Most of it was a waste of time.
Today, that time goes into READING the thing instead of FINDING the thing.
If you're a doctor or a scientist, put it on your phone.
Stuff
💡 Gear I use - Dreame H16 Pro steam floor cleaner​
This thing doesn't replace a good floor cleaning you do with a bucket of water and a rag.
But it allows you to use the bucket and rag less often…
It washes and dries in one pass, and cleans its own roller when you dock it.
Is it more machine than a floor requires? Probably. I use it anyway 😄
Epilogue
If you received this newsletter from a friend and would like to join Friday's Digest, visit https://newsletter.shaysharon.com​
That’s it for this issue.
See you next week!
Shay
Want to learn more from me?
Tip Sheets:
- My Workflow for Reading & Writing Manuscripts
- Future Fellows & Postdocs
- The PhD Journey
- How To Write an Abstract in 10 Minutes
- How to Study for the TOEFL
Video Guides:
- Manage References with a SINGLE CLICK
- My Workflow for Reading & Writing Manuscripts
-
How to Write an Abstract in 10 Minutes​
- Make PubMed Work For You
- How I Use Anki Flashcards
- How I Studied for the USMLE
- How I Studied for the TOEFL
-
What's in My Bag​
- Clinical Photography - The Complete Guide
​
​