Gabe Erion Barner @[email protected]

EM resident physician @bidmcem, MD/PhD. Research in machine learning w/focus on interpretable ML & applications to trauma and ICU care. He/him.

[Gabe Erion Barner @[email protected] @gabeerion]

Our preprint is out! We present CoAI, a method to reduce the data-gathering cost of any machine learning model, and show it reduces time, effort, and monetary costs for health care providers and patients.
@joejanizek @suinleelab @SayreMR @utarnachitt

https://t.co/pGFKVj9ntC

1/

[Gabe Erion Barner @[email protected] @gabeerion]

@joejanizek I guess it's interesting to me because it feels like the implication is "don't worry about deskilling" which feels wrong to me! But I don't think relying on CT is bad at all

[Gabe Erion Barner @[email protected] @gabeerion]

@joejanizek Congrats, this paper is awesome! Do you think there are applications in text/tabular settings like the ED?

(though I guess it's common to turn a history-taking problem into an image classification problem by pan-scanning :)

[Red Thread Games @RedThreadGames]

🚨 SWEEPSTAKE ALERT 🚨

To celebrate the release of Dustborn we are giving away 5 Dustborn superpacks!

To enter, follow @redthreadgames, like and repost!

Sweepstake ends on August 27!
Good luck!

[EFF @EFF]

Without notice, X has opted all users into training its "Grok" AI Model. To turn off this setting and stop your "posts, interactions, inputs, and results" from being used for training and fine-tuning Grok, visit https://t.co/hhKe2I8u9d and uncheck the checkbox.

[Oladimeji Oki, MD @TheRealDr_O]

12 years ago today was my first day as an intern. Residency was such a time of growth and realization. Good luck to all the new interns who are going to have the highest of highs and lowest of lows this next year. You'll do great! Just try to learn something every day

[Gabe Erion Barner @[email protected] @gabeerion]

@jmschreiber91 @IMPvienna @UMassChan Congrats Jacob! Would love to meet up with you in MA!

[Gabe Erion Barner @[email protected] @gabeerion]

@joejanizek Congratulations, Joe!! So excited for you!

[Lindsey Erion Barner @LindseyBarner]

We're hiring at the @broadinstitute! @michlipi and I are looking for talented research associates to join the Spatial Technology Platform and advance spatial profiling and transcriptomics technologies. Come join us!
https://t.co/8ZiRZ6gZne

[Gabe Charbonneau, MD @gabrieldane]

The EHR should be a decentralized infrastructure platform for the public good. Patients should own and control our own data. Not this current mess.

[Tal August @tal_august]

I am recruiting PhD students to work with me at @IllinoisCS at the intersection of HCI and NLP! I am especially excited about understanding how we use language to communicate and building tools that help us communicate better.

More: https://t.co/jkrcQpLckn

[Robert Graham @robertgraham]

For example, to prove my point, I opened the podcast (https://t.co/a4TeapBKM2) and skipped forward to a random location, around 37 minutes into the thing (I can't bear to watch all 3 hours and debunk point by point).

At this point, he's talking about a "Lazarus Report" that said 1 in 37 people had an adverse reaction to vaccine. I'm pretty sure he means this:
https://t.co/gYVrUzXWWl

This is the sort of things that live, I can't rebut, but written, I can.

I googled, I found the thing, I read the thing. It doesn't say what RFK claims it says.

The way VAERS (vaccine adverse event reporting system) works is that you should report any event after a vaccine that happens 30 days after. This includes things that couldn't possibly by related to the vaccine, such as a pedestrian getting hit by a car.

This floods VAERS with garbage, where 99.9% of the reports have nothing to do with vaccines, because on average, within 30 days, things happen to people.

In other words, 1-in-37 is the same chance that in the next 37 months, you'll have one of the 890 conditions that VAERS wants reported: falling down the stairs, getting a cold, severe headache, ingrown toenail, and so forth. This is normal.

If you get a placebo vaccine instead of a real one, there's a 1-in-37 chance in the month after you'll have some event that VAERS wants reported.

The point of VAERS isn't the absolute numbers but relative numbers. There's a spike in the number of people getting ingrown toenails after a new measles vaccines, something so totally unexpected, this system will help find it.

During the pandemic, lazy doctors who did a poor job reporting to VAERS suddenly got diligent, and the number of VAERS reports shot through the roof -- including all 890 categories, including getting hit by a car. It didn't mean covid caused anything, because almost all 890 categories went up mostly equally. It's how we know myocarditis was actually a problem, because those reports went up more than the rest.

The point is that you really can't debate this sort of thing live. I'd never heard of the "Lazarus Study" before, though I have researched VAERS thoroughly, so I wouldn't be able to debunk it. Conversely, RFK just brings it up out of nowhere and misrepresents it. Moreover, the RFK/Rogan audience have such low levels of education, they simply can't follow the complex explanation debunking it.

You can't live debate crazies. It just won't work.

[Chanwoo Kim @ChanwooKim_]

We've developed a framework to promote transparency in medical image AI, leveraging an image-text foundation model. We introduce MONET (Medical cONcept rETriever), a tool that empowers large-scale data auditing, model auditing, and inherently interpretable model building!

[Roxana Daneshjou MD/PhD @RoxanaDaneshjou]

Dermatologists need to be aware that Google is giving lists of diagnoses using AI for patients who upload images of skin disease. They claim it’s just “informational” so doesn’t need regulation. https://t.co/jdWUoPVD1b

[Jeremy Howard @jeremyphoward]

I teamed up with philosopher @sethlazar and AI impacts researcher @random_walker to investigate the "Statement on AI Risk" that proposes:

"Mitigating the risk of extinction from AI should be a global priority".

tl;dr: We're not convinced.🧵
https://t.co/fdvKXkmotR

[Gabe Erion Barner @[email protected] @gabeerion]

@joejanizek @David_Ouyang The one research idea we've ever discussed that brought visible sorrow to an attending's face and it's the one Joe shouts me out for 😅 really appreciate the discussion above though!

[Joseph D. Janizek @joejanizek]

This project has everything! In-depth audits of 5 dermatology AI models, architectural innovations for generative XAI methods, and a framework that promises to help medical AI developers improve their algorithms using the data they already have (and some domain expertise).

[Joseph D. Janizek @joejanizek]

Want more reliable feature attributions in biomedical machine learning tasks like predicting optimal combinations of anti-cancer drugs from tumor expression? Ensemble multiple models (even if it doesn’t increase your pred performance)! https://t.co/rgZJGQfGEc

[David Ouyang, MD @David_Ouyang]

@zakkohane Patients w rare or missed diseases often mention if they put in their one-liner into Google, their diagnosis would come up and chalk that up to the incompetence of their docs, but I think it’s the synthesis into that one-liner that creates the incantation leading to the dx.

[Gabe Erion Barner @[email protected] @gabeerion]

@EricTopol While I see a couple plausible mechanisms for LLMs to show more empathy (especially more "time") I worry using Reddit posts as the source of physician advice is a big study limitation and gives doctors a less empathetic "tone" by design.