Viren ✧˖°

Daanish & Co. are building something amazing here! I’d encourage my healthcare peeps to check it out!

We’re announcing Kos-1 Lite, a medical model that achieves SOTA on HealthBench Hard at 46.6%. As a medium sized language model (~100B), it achieves these results at a fraction of the serving cost of frontier trillion-parameter models.

Today Northwood has raised $100M in Series B funding led by @WashingHarbour and co-led by @a16z to support a simple objective – take space missions further faster. This financing follows millions in signed contracts including a $49.8 million contract with the Space Force to support the Satellite Control Network – critical infrastructure used for launches and early satellite operations, to track and control satellites, and to provide emergency support to tumbling and lost satellites.

@jjuulliiaallii claude

(1/5) New post: "Mismatch Praxis: Rollout Settings and IS Corrections". We pressure-tested solutions for inference/training mismatch. Inference/training mismatch in modern RL frameworks creates a hidden off-policy problem. To resolve the mismatch, various engineering (e.g., FP16 unification, deterministic kernels) and algorithmic (e.g., importance sampling) fixes have been proposed. In this work, we examine how rollout settings (temp, top-p, and top-k) affect mismatch, and how importance sampling corrections bear out in practice. We find that while Sequence-TIS is theoretically optimal, it can succumb to catastrophic variance in long-horizon contexts. Additionally, non-standard rollout settings create subtle mismatch patterns that require careful engineering fixes. Token-TIS with default rollout settings proved to be the most robust setting for long-horizon training.

"What nature does is rigorous until proven otherwise; what humans and science do is flawed until proven otherwise." - Nassim Taleb, Antifragile

I've just written a new post raising a series of concerns about Nucleus Genomics, a company that offers embryo selection services based on polygenic scores for couples undergoing IVF. I was shocked by the degree to which Nucleus's work is obviously plagiarized or simply wrong.

@connerchyung There was a patient I had who was on max dose ozempic for 4 months and gained 40lbs. It’s a matter of will and dedication.

when are they gonna invent reverse ozempic

@corinavcruz @growing_daniel putting this on every mandatory zoom session.

@growing_daniel When even the chad filter can’t save you.. you’re down bad

Putting your face through a chad filter for an internet show interview is insane

@EM_RESUS need a meta-analysis on this ASAP!

Hi, ER Doc here. I can’t believe l actually have to say this but pee, in fact, is NOT “stored in the balls”.

the chat format fits perfectly with image study recommendations! Especially with the way the ACR has organized it via conditional patient criteria.

This database from the @RadiologyACR to integrate in @EvidenceOpen or @PathwayMedical ! Could help reduce unnecessary scans and put a smile on your local radiologists face 😅 acr.org/Clinical-Resou…

@EM_RESUS Bro forgot to renew his left lung subscription

Here's one of the most impressive CT Scans I've ever seen of a 40-year-old man who was involved in a brawl What's the diagnosis?

@garvs_72 congrats Dr 😎 looking forward to the residency content

I bamboozled them into giving me a medical degree this morning Dr. Mr. Garvin

@jasonryanmd @maozali_ You can share your screen with it, even talk to it about what is on your screen like a tutor. It’s awesome!

@maozali_ Do you just have it open and type in your question to it? Or do you have some way it can read your screen and see what you’re working on?

Is anyone out there using ChatGPT or other open access AI as they do NMBE and UWorld questions?

@sandw1ch___ like a moth to flame

@sandw1ch___ meg 2 is going to be a spiritual experience

@lilsketch 😭😎💀👍🚨❤️‍🩹

Prove you not toxic put all six recent emojis 🤑💻🚫💌💤🙃

one of the most heartbreaking stories in all of tech. rip

@coochielover121 is the roach still alive?