Shahar Shelly M.D

Neurologist and Neuromuscular disease specialist

[Dvir Aran @dvir_a]

Gregory? This is our own multi-agent LLM that was taught to emulate the thinking of an expert diagnostician - based on the initial presentation it collects all the possible diagnoses (with RAG), and attaches confidence levels to those. Then it finds all the relevant tests, and measures the entropy of how the test may affect our confidence levels, and chooses the best test. It then reassesses its list of diagnoses and their confidence and moves to the next test or submits a diagnosis. It worked amazingly well! (8/n)

[Dvir Aran @dvir_a]

Since then, Google's AMIE outperformed docs in telemedicine (more empathetic!). Last month Microsoft posted on @arxiv a study on simulating hospital scenarios where docs order tests and diagnose.
We've been hard working for a while now on something similar which simulates specialist-level workflows, but with one student and no funding. Still, I think we built something really nice. (3/n) https://t.co/s2D8WjFGbx
https://t.co/x0cI0udGRc

[Dvir Aran @dvir_a]

We've seen the works from @GoogleDeepMind and @Microsoft showing LLMs acing USMLE-style exams, but those static multiple-choice setups don't test real diagnostic skills like iterative info gathering and hypothesis refinement. Last year, a @NatureMedicine paper showed LLMs struggle with such diagnosis paths. (2/n)
https://t.co/vSjvl4D8oT

[Shahar Shelly M.D @ShellyShahar]

Excited to share our latest NEJM editorial on B-cell depletion in myasthenia gravis!
We explore the emerging promise of anti-CD19 therapies and the potential for long-term remission in gMG rethinking chronic immunosuppression. @NEJM @MyastheniaOrg

https://t.co/9OVFAcVbXj

[Elie Naddaf @ElieNaddaf3]

Why don’t journals make the name of their downloadable pdf files as Year_Author_Title instead of ljdh6363638f_fgen45.pdf? 🤔

#AcademicTwitter @AcademicChatter

[Dvir Aran @dvir_a]

דירוג מדליות לפי מספר תושבים. אנחנו מקום 11.

[DailyAI @DailyAIOfficial]

AI research takes a turn with "Quiet Star," teaching language models to "think" before responding.

Researchers from @Stanford University and Notbad AI trained a language model to think before generating outputs.

https://t.co/knfrDUKbop

This innovation not only improves the quality of interactions but also paves the way for more reliable and nuanced AI communication.

Explore the impact of Quiet Star on the future of conversational AI in our article ⬇️

#AI #LanguageModels #TechInnovation #QuietStar #StanfordUniversity #NotbadAI #FutureOfAI #AICommunication #TechNews

[Shahar Shelly M.D @ShellyShahar]

@EliaSechi Great work!

[Nat Rev Neurology @NatRevNeurol]

Our February issue features articles on inflammasomes in neurological disorders, treatment of myasthenia gravis in the era of molecular therapies, post-COVID dysautonomias and multidimensional genetic diagnosis of rare movement disorders. Online now! https://t.co/GNuJvROwqQ

[Dvir Aran @dvir_a]

🚨Long due, but finally published! This paper will not make me many friends, but I think it's highly important. In metastatic melanoma, which treatment is better: monotherapy (Keytruda or Opdivo) or combination therapy (Yervoy+Opdivo)? 🧵https://t.co/nhKp6gTvZ3

[DailyAI @DailyAIOfficial]

Researchers @Dvir_a, @ShellyShahar & the team from @RambamHCC used GPT-4 to match & even outperform #Neurologists in #Stroke care.

The study shows GPT-4's potential in life-saving decisions, offering hope for faster, more accurate treatments.

Details ⬇️

https://t.co/Oh2TIGeU1I

#AI #AIinHeatlhcare #healthcare #medicine #treatment #stroketreatment #GPT4

[Shahar Shelly M.D @ShellyShahar]

@lpachter @dvir_a @RambamHCC Several times for kappa score .. 💪🏻

[Dvir Aran @dvir_a]

🚨New preprint alert. GPT-4 has been shown to work well for diagnostics, but rarely as a clinical decision-support tool. In joint work with @ShellyShahar from @RambamHCC we used it to decide how to treat patients with stroke. How did it do? Amazingly well! https://t.co/8q31Mu2PRB

[Dr. Eli David @DrEliDavid]

🚨 Presidents of @Harvard @MIT and @Penn: “It's OK to call for genocide of Jews” 🤯

[Marina Medvin 🇺🇸 @MarinaMedvin]

“I was forced to leave my study group because my group members told me that the people at the Nova music festival deserved to die because they were partying on stolen land.”

MIT.

[Anthony Goldbloom @antgoldbloom]

A new survey suggests TikTok is a meaningful driver of a surge in antisemitism. #TikToxic

Spending at least 30 minutes a day on TikTok increases the chances a respondent holds antisemitic or anti-Israel views by 17% (compared with 6% for Instagram and 2% for X).

[Lyell Jones MD @LyellJ]

@EduardoPegna Also caused by the ICI in this case.

CK elevation is mild in these patients (mean 600-700) and most have fibs. If found should workup for cardiomyopathy given high rate of overlap syndromes

Great series from @ShellyShahar @MarcusVPinto @TLiewluck ⬇️
https://t.co/C3GIzvIUgZ

[Ben Caspit בן כספית @BenCaspit]

שימו לב למיזם הזה:
נעים מאוד, IronTruthBot, מיזם הסברתי שנלחם בתוכן שקרי ומסית נגד ישראל. זה פשוט: מצטרפים לבוט דרך הלינק בטלגרם ושולחים לו קישורים לפוסטים מסיתים, שקריים ומכפישים נגד ישראל מכל הפלטפורמות. הבוט החמוד הזה, שפותח ע"י קבוצת מתנדבים מקסימה, מקבל כ-700 דיווחים ביום והצליח להסיר מאות פרסומים לא ראויים.
עוד משהו שהחברה האזרחית המדהימה שיש לנו כאן מפתחת לבד, בלי לחכות למדפסת של גלית דסיסטל אטבריאן.

https://t.co/kpcsbTSwgs

[רמב''ם הקריה הרפואית לבריאות האדם @rambam_hospital]

ממשיכים לסייע ולתמוך בכל מטופל ומטופלת שצריכים אותנו

https://t.co/aM0cByH6QZ