Shamay Agaron

@joshuapliu Agreed that quality assurance frameworks are more appropriate for evaluating CDS AI response quality however, there is a place for RCTs to understand how CDS AI tools impact medical decisions (e.g., diagnoses, workup, and management) in live settings

How can we trust CDS AI like OpenEvidence, DoximityGPT, etc if no one's done studies showing their use improves patient outcomes? While I understand the intent, I think it's ultimately a misguided question. Anytime a new medical textbook comes out, should we do a RCT to prove that it improves outcomes? Most would agree "No, that's a waste of time". Instead, we evaluate a textbook based on the knowledge inside. If it's clearly grounded in clinical evidence and widely accepted medical knowledge, we're good with it. The textbook itself can't control how physicians use the knowledge (which has a bigger impact on patient outcomes). Using the same lens, doing RCTs on AI medical search tools like OpenEvidence isn't the right way to evaluate their utility. Of course, a valid difference with AI is that the output is dynamic and non-deterministic - the same question can technically lead to different answers. So it's understandable to ask "shouldn't we study this more before it gets used everywhere?" So what SHOULD matter in evaluating CDS AI tools? Ultimately it comes down to two things: 1/ The quality of data the AI is trained on: are the answers I'm getting based on the best available medical knowledge? 2/ The approach to responses generated: even if the AI has access to the right medical knowledge, does the AI synthesize the most accurate, relevant and useful information from that dataset? The interesting thing is that while there's no universal answer for either of these, these questions also exist for every non-AI medical resource we already trust. Authors and editors of journals, textbooks and UpToDate decide what to include, what to leave out, and how to frame uncertainty. With regards to (1), do you care more about the AI model being trained on specific journals? Clinician-curated content such as UpToDate? Being able to choose the specific sources yourself (as Heidi Health's CDS AI allows)? And then with (2), physicians are a finicky bunch. Some prefer OpenEvidence's approach that sounds more like a journal review. Others prefer DoximityGPT's more practical outputs, often filled with bullet points and tables. All of these approaches have trade offs in terms of what they emphasize vs leave off the table. To me what matters most - and what is genuinely non-negotiable - is how reliable the AI is at not hallucinating. Clinicians choose how medical knowledge is applied to patient care, but a tool that fabricates knowledge introduces error upstream of clinical judgment. If this error rate is too high, that's unacceptable. Nonetheless, you don't need a RCT to evaluate these elements. Presumably all of these CDS AI tools undergo human + software testing on reliability of output (and I'm sure this is continuously being done for quality assurance). I'm all for validating that type of testing - but I'm unclear why a RCT is best for this (and wouldn't you need on-going testing anyways, as opposed to a one-time RCT?) Agree or disagree?

@thekaransinghal I agree that real conversations with clinicians are likely to involve some back-and-forth, but patients behave differently when asking questions about health online - they may ask a question, read the answer, and move on. IMO evaluating single-turn performance is relevant too

A large fraction of pain can be resolved with mind-based practices. Yet no one is rigorously examining what works for whom. I want to build a roadmap to resolving chronic pain. Now seeking funding for my nonprofit

@mxslk reminds me of: "[Jhanas] are a rare technology whose instructions are encoded in our bodies. Jhanas are an algorithm: a set of instructions that, if executed correctly, solve for a problem that you may not have even realized you’ve been trying to unravel." nadia.xyz/jhanas

More and more I’m seeing methods like Alexander Technique or Feldenkrais as embodied algorithms for aligning movement and perception for the whole human

@nickcammarata i remember from learning the basics of Reiki that it's easy to unintentionally influence the energy in the body part where the hands are placed (as opposed to just scanning / observing). this made me hesitant to use my hands in the way you describe - do you notice that at all?

I’ve been working on moving energy around my body for a few months and now my body gets unhappy unless I have one hand on my stomach and the other on my heart, which for some reason like doubles the sensory resolution of what’s going on in there

@this_is_silvia i would love to hear more about this - are these completely original or based on prayers/poems you've come across before? i have poems that i return to for feelings of expansiveness for example, but never thought to create my own

i've been creating and memorizing my own personal poem-prayers for different purposes (creativity, protection, strength, etc) and it's one of the most beautiful, grounding, nourishing practices i've ever had

@yung_ibogaine welcome to the $100 fine club my friend

fucking mta is good at catching people not pay their fare and terrible at giving sensible service for that money. clowns

youtube.com/watch?v=aRULtX… This one is pretty much a poem released as a song

youtube.com/watch?v=2lrOYX… original poem is called "Love in the Time of Undeath" recommend listening to both to really feel the new dimension that music adds to it

pairing music with (spoken) poetry feels completely unexplored Fred again produced an incredible song (link below) using the most touching soundbites from Kyle's poem what other poems would find wider appeal in music?

@castillo__io Do you see Neurosity pivoting beyond EEG and/or other form factors? What have been the frustrations with the current design from a development perspective?

I was invited to a podcast that’s happening tomorrow. What would you like to know about me and Neurosity?

@ulkar_aghayeva 🙏

please reply if you are in NYC and want to be a part of a reading group to discuss the new paper on the assembly theory by the Lee Cronin group "Assembly theory explains and quantifies selection and evolution" (link to the paper below, it's open access)

art by monet, music by debussy they both lived in France in the same time period youtu.be/WDtAwXZkdU0

always wished that museum audio-guides would have music pairings for each painting recently discovered youtube channels that do this by turning your TV into a digital art gallery with a playlist going to use this thread to keep track of some i enjoy

@ulkar_aghayeva @groupmuse this is amazing, really excited about this Ulkar :)

my string trio on the theme of "The Joy of Scheherazade" will be performed live for the first time on April 20 at a @groupmuse in Williamsburg! groupmuse.com/events/13240-s…

@lxcosta @Matt_R_Angle also working in this space! please reach out in DMs if you want to compare notes :)

@Matt_R_Angle 🤚working in the intersection of hearing loss and dementia 🧠🦻

Who in neurotech is still active on Twitter?

@ulkar_aghayeva somewhat related, I really enjoyed this paper that contextualizes early 20th century morphogenesis predictions in light of recent discoveries link.springer.com/article/10.100…

@drmichaellevin @LPiolopez Diagnostics based on bioelectric states would potentially be able to detect disease states much earlier, given that changes in bioelectric networks precede morphological changes is there evidence that bioelectric changes precede morphological changes after embryogenesis as well?

Next, in a series of preprints and papers getting over the finish line at end-of year: osf.io/7hxnp with @LPiolopez: "Morphoceuticals: perspectives for discovery of drugs targeting anatomical control mechanisms in regenerative medicine, cancer, and aging"

@drmichaellevin fascinating! do you think that the basic science is far enough along to start developing therapeutic interventions that exploit this top-down pathway? I've read about your spinoff Morphoceuticals, but wondering if there are more :)

New preprint with linkedin.com/in/juanita-mat… and two undergrad co-authors: "Cellular Signaling Pathways as Plastic, Proto-cognitive Systems: Implications for Biomedicine" - some ideas on a different way to think about "pathways" osf.io/c6n9r/

@alexeyguzey Dan Freed (CEO of Thesis) created a nootropic stack specifically for stimulant use: takestasis.com Not sure how much of a focus on short term vs long term reduction, but they might have some insight

I want to hire someone to research: - when is adderall neurotoxic? How big is the effect? - how does tolerance work and how does it get reversed? e.g. what happens to people taking it daily for years? - what about modafinil, caffeine, nicotine? alexey@guzey.com if interested