#EveningRead

Professor Nicholson Price (@[email protected]) and co-authors ask, How can we enable local health systems to collaboratively & effectively govern #medicalAI ?

Read: 'Enabling collaborative governance of medical AI' in Nature Machine Intelligence: go.nature.com/3KDPhbF

The final version of Nicholson Price (@[email protected]) & my paper Locating Liability for Medical AI via.library.depaul.edu/law-review/vol… part of the DePaul Law Review Clifford Torts symposium is now out. The paper is about how to allocate liability between hospitals and developers for medical AI errors.

Great talk by UC San Diego CRIO Mike Hogarth on clinical research use of LLM to rapidly facilitate clinical hypothesis generation with Rene Cassie from medeloop at CI4CC

OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments

Ever since OpenInterpreter, we've all been wondering just how effective agents can be if you give them a computer.

Now we have a proper benchmark. Let's take a look (🧵):

If it runs into itself, game over.

Coming soon: support for plot_layout() within TidierPlots, with syntax borrowed from the {patchwork} #rstats package.

Flexibly lay out plots *just* the way you want them.

“Where’s all the outcomes data?”

Love spending time with my University of Michigan health AI ethics and policy colleagues Jodyn Platt Kayte Spector-Bagdady Nicholson Price (@[email protected])

“Where’s all the outcomes data?” asks Yasir Tarabichi MD, MSCR from Case Metro Health at the LearningHealthSci-UM Health AI Ethics and Policy Symposium.

Yasir Tarabichi MD, MSCR has been a leader in outcomes-oriented AI quality improvement projects, and highlights recent work looking at outcomes UC San Diego Health.

Dr. Sharon E Davis from DBMI at Vanderbilt University highlights reactive and proactive examples of AI model updating and proposes a learning prediction system at LearningHealthSci-UM symposium.

Framework:
- how to determine if drift has occurred?
- what data to collect?
- what is best approach to updating?

Tidying nested data structures (like JSON objects) is often a point of friction in data analyses.

TidierIteration.jl for #JuliaLang provides a number of utilities for flattening dictionaries, even into data frames.

Read more about flattening here: tidierorg.github.io/TidierIteratio…

I love that there are folks who’ve never used the tidyverse but want to help recreate it in #JuliaLang .

🚀

We're building a big corpus of high-quality The Julia Language code to trick the LLM people into using our database, and/or fine tune a Julia-specific model.

This trick just re-formats all the Julia code in the ecosystem using different styles.

discourse.julialang.org/t/an-llm-fine-…

TidierPlots v0.7.4 is available on the registry — improves scale_* functions and adds initial functionality from #rstats {patchwork}.

Also includes many improvements to the documentation: tidierorg.github.io/TidierPlots.jl…

Happy tidying!

did you know? syntactic sugar is an important part of every meal.

A great read!

tl;dr Rust doesn’t solve the 2-language problem in game dev.

Game dev (like science) is iterative. You benefit from having a frontend language that is flexible and backend code that is reliable.

Rust = great backend but not ideal for frontend due to rigidity.

This is amazing! What else is missing in Julia to make the move from #RStats smoother?

Data Wrangling and Plots
✅ Tidier.jl (tidyverse)

Reporting
✅ It seems Julia is already supported in quarto, right?

Models
☑️ Is there something like tidymodels?

What else is missing?

mighty men of michigan 🥹

Very deep dive into how our use cases shape our needs computing needs, and why there is no one-size-fits-all