The presenter puts up a slide showing “random forest variable importance.” You know the one... The sideways bar plot. Says “only showing the top 20 variables here...” to highlight the hi-dimensional power of random forests. The slide is awkwardly wide-screen. Everyone squints.

DAGWOOD R package update is now up on CRAN with a major new feature: Input a DAG, output a list of human-readable assumptions, automatically. Assumptions list is powered by the DAGWOOD exclusion and misdirection branch DAG algorithms.

iPad calculator is actually pretty nuts

🚀 Exciting News! 🚀 We're thrilled to announce that our longitudinal electronic health record (EHR) dataset, EHRSHOT, is now available for download! 🌟 📊 6,738 patients 📅 41.7M events 🩺 921,499 visits Details at: ehrshot.stanford.edu

The KCMO score: intuitive (% of ideal 💊), patient-centric, can adapt as guidelines ∆, ↑ granularity vs existing methods. The new way to ⚖️ GDMT." Mirza Khan Saint Luke's Mid America Heart Institute @GCFMD Nobuhiro Ikemura @javedbutler1 Charles F. Sherrod IV, MD MSc #AHAJournals ahajrnls.org/3VC3yLQ

Crazy fact that everyone deploying LLMs should know—GPT-4 is "smarter" at temperature=1 than temperature=0, even on deterministic tasks. I honestly didn't believe this myself until I tried it, but shows up clearly on our evals. ht to Eugene Yan for the tip!

I love @DuckDB. It's just is such an amazing tool! Since I've been working a lot with embeddings lately I wrote an article about using DuckDB to store embeddings and do vector search. Enjoy! blog.brunk.io/posts/similari…

🚨 New paper alert! We provide new data on private equity's latest target: outpatient cardiology. 🫀🩺 Starting with the first deal in 2019, PE has acquired 320 practices and >10% of all practices in 7 states. 70% of all acquisitions happened in 2023. jamanetwork.com/journals/jama-…

AI (RAG-Enabled GPT4) can screen patients for a clinical trial as or more accurately than study staff and does so much faster at a fraction of the cost! I am thrilled to share our study - just published NEJM AI 👉bit.ly/4cp1n3P So thankful to all of my colleagues

💊 We took your language model to the drug store… and it knew about acetaminophen (generic name) better than Tylenol (brand name)! Hugh Zhang Scale AI developed GSM1K last month, where they found many of the open-source models are overfitting to the benchmark. We wondered how

Part of our Med-Gemini work was a full relabelling of MedQA, revealing that at least 7.4% of examples are unfit for evaluation. Today, we open sourced these annotations alongside our evaluation script as a new standard evaluation on MedQA. A thread 🧵: github.com/Google-Health/…

Character AI is serving 20,000 QPS. Here are the technologies we use to serve hyper-efficiently. [research.character.ai/optimizing-inf… ]

Fantastic summary of the most effective learning techniques based on research study findings:

In ML, we train on biased (huge) datasets ➡️ models encode spurious corrs and fail on minority groups. Can we scalably remove "bad" data? w/ Saachi Jain Kimia Hamidieh Kristian Georgiev Andrew Ilyas Marzyeh we propose D3M, a method for exactly this: gradientscience.org/d3m/

👑 Positron has my 💕 🪦 RIP RStudio 🌹 A quick tour of Positron and how I have been using it every day for the past few months! #rstats youtube.com/watch?v=Z5-l6G…

Doctor quality matters a lot. Norwegian study finds that replacing one of the 5% worst general practitioners with one of average quality generates a social benefit of $9.05 million. Via Jonas Vlachos. cesifo.org/en/publication…

There are good arguments for removing race from medical algorithms, but there may be unintended consequences. Our PNAS paper finds that race-blind algorithms can *worsen* racial inequalities, bc they can't adjust for racial disparities in data quality. shorturl.at/7ugW5

😍😍😍 ✏️ You can add annotations to specific parts of your Quarto code chunks! It’s so easy to use, but I just discovered it—and I’ve never seen anyone using it! This is my Quarto trick #27 Example: holtzy.github.io/quarto-tricks/… Code: productive-r-workflow.com/quarto-tricks#…

.Georgia Tomova using the example of 'grip strength' to explain the difference between prediction and causal inference. Grip strength may be good predictor of future health, but training your grip strength isn't likely to improve your health very much. #CausalIntroCourse

Can you guess what happened when researchers in Germany gave kids e-readers & access to a large digital library of age-appropriate books? ... ... ... Reading performance went up as a direct result. But (!) so did math performance. And (!!) so did socio-emotional well-being.