Not my intent at all, the safety/patient comfort data is interesting. I just never thought of arterial lines as something that would affect mortality.

I view a lines as a tool, not a therapy, so it strikes me as counterintuitive to prioritize them. But yeah I guess some people prioritize the wrong things 🤷‍♂️

Ultimately, the power to distribute educational materials through vibe coding is HUGE. You can make an aesthetic, practical website for it in a few days. But keep the projects small, ideally internal, and be sure to always review legal and security measures it made
#AI #meded

Knowing 0 code can lead to $$ by a thousand papercuts. I spent $12 developing this site (plus any deployment maintenance costs I incur), but I edited a lot of small things because they were easy and would save me $0.05-$0.30 per edit, which adds up when you edit 100 small things.

Now, can you do this without knowing any code at all? YES! I was really surprised and impressed with how well it developed and added features, plus Replit does seem better prepared than others to handle secrets and maintain security best practices, but there's one caveat:

So ask it! Prompt: "Make the website comply with all international laws for websites". Attach it to a separate entity if you are able. And make sure no copyrighted materials, illegal links, or other illegal materials are on your site.

As SOON as you gather user data, you are REQUIRED to add a privacy policy, cookie banner, terms, etc. The default is that the site is connected to you (and not an entity separate from personal assets), meaning fines and/or lawsuits for mishandling data could fall to YOU.

After I asked for the main architecture and added the major features, I had a working prototype without writing a single line of code myself. It didn't look half-bad, and I was blown away by the speed... until I realized that the website was breaking multiple laws...

It helps to know some architecture. I specifically asked for a react and express setup, and I'm not sure what it would've done if I hadn't. If you don't know about them, you should consult with an LLM about the best architecture for your purpose BEFORE a single line of code

I built the POCUS website feature by feature. Main framework and curriculum pathway that connects to the database -> add auth -> add groups -> add an instructor dashboard for the groups -> and so on

To start, build feature by feature

❌ Don't say "build me duolingo but for POCUS"

✅ Say "Build me a POCUS website with **exactly the main feature you want and exactly how you picture it** using a *** framework"

Trying to one-shot it makes an absolute mess of bugs and errors

First, #meded vibe coding is good for SMALL, SIMPLE extensions for curricula or assessments like this.

It is NOT good for large, integrated platforms that connect to external APIs. These are HUGE security threats if you don't understand them.

It’s time that we in #meded ask ourselves if churning out pubs is educationally beneficial for medical students, or if we can reshape the system to reward a focus on quality rather than quantity

Ultimately we found a dramatic increase in the quantity of medical student publications with pretty flat metrics on quality. All while costs to publish are rising which can be a huge burden for those trying to match into competitive specialties.

💸 An increasing cost burden:
In 2012, we estimated only 3.7% of student papers were in journals with required fees.
By 2022? 14.8%.
That’s nearly a 4x increase in pay-to-publish trends for student work.

📝 Study types shifted:
• Case reports ↓ from 26% → 14%
• Perspective/opinion pieces ↑ from 6% → 23%

📊 Journal quality (by Scimago Rank) had a very small increase:
Median SJR rose from 0.470 → 0.630
(For reference NEJMs rank is 19, JGMEs is 1.1)
But…
🧵 23% of all articles were never cited even once.

Over the ‘12-‘22 span, 5,181 med student-authored articles were included.
🧑‍🎓 50% listed a student as first author
📉 That % actually dropped from 63% → 48% over time
So more papers, but proportionally fewer led by students.