Good thread. However, training / prompting AI to give you the knowledge “territory” around a query seems like something the dreamy nature of latent space should lend itself very naturally to

One thing I cannot understand about Rust on Windows is the difficulty of debugging; really doesn’t make sense for a language of its caliber

The world needs an atproto version of TikTok

I loved Cave Story, it was my first foray into indie gaming; reminiscent of lots of late summer nights in 7th grade

Next he will score a perfect 1600 on his PokeSATs, gain admission to the prestigious Viridian University and find himself an elite 250k/yr finance job managing Rare Candy futures

2 parents with average IVs —> member of the affluent intelligentsia class; the American dream in a nutshell

if arguing online changed anything we'd all be running Linux

#MedSky

Ideal system might be some cross between free market for routine healthcare and coverage for catastrophe with cash subsidies for all for public health interventions like vaccines which are necessary to prevent outbreaks, etc.

Devil is in the details ofc - who defines “catastrophic?”

Screenshot bc he is unfortunately not on Bsky :/

There’s an incredibly thoughtful subset of health insurance discourse after the UnitedHealth incident if you can filter out the noise. This is one very sensible take

One issue with this is that insurance funds public health interventions like vaccines that are a social good - much likelier that more ppl skip measles vaccines and others if it now costs them additional $ out of pocket which means worse outcomes at large re: outbreaks, etc

Proof of zero knowledge? Just look at my Github 😂

This #AI model gives a fascinating glimpse into how AI is revolutionising #healthcare, and especially #radiology.

These sort of models look like they'll also have huge utility in terms of teaching.

Well worth taking a look at!

microsoft.github.io/BiomedParse/

Theming is also very difficult, tried using it with Bulma CSS and had to do weird hacky things like making a parent bulma style component that everyone inherits from, etc

AI research feels like it’s stuck at a local maxima, which is great because we’ve needed a breather for about the past two years or so

2) do healthcare orgs even have the requisite technical staff to maintain, audit, and keep these fine tuned models robust against model drift?

I think it’s a long play — the foundation model gets offered for fine tuning to develop custom automations - user gets a license to continue API calls to fine tuned version of model, if the model is good enough you’re locked in! The big questions are 1) whether the fine tuned models work and…

Very cool esp re: more companies offering this as proprietary solution (see: Bunker Hill, Hoppr); excited to see more offering foundation models with different modalities as well

Tfw you turn your nose up at $8/mo for Twitter premium but will go ahead and drop $150 for a Raspberry Pi and spend 20 hrs debugging your PDS just to get onto Bsky -

anyway, happy to be here, what's cookin?