Kevin Shen

@newlanternai

[New Lantern @newlanternai]

"Not having to whip my neck around all day allows me more time to spend looking at the images and taking care of the patients.”
- Dr. Alexander Craft, MSK Radiologist & Partner at Radiology Associates, LLP

Reading cases on New Lantern has elevated Dr. Craft and Dr. Fan’s joy for radiology as they dial in on patient care instead of dragging-and-dropping thumbnails, hand-drawing measurements, and ultimately flipping back-and-forth between PACS, reporter, and worklist all day long.

What does this mean for your practice? Well, as Dr. Fan says, “In New Lantern, I clone myself,” yielding in 30%+ more RVU production in the same amount of time while increasing the quality of care and time spent looking at images.

Watch the whole conversation with Dr. Craft and Dr. Fan.

#radiology #medicalimaging #newlantern

[Andrej Karpathy @karpathy]

This is a new paradigm for interacting with Claude that is significantly more "inline" with all the other human activity org-wide. Once you do all of the under the hood engineering work to make this "just work" (e.g. across tools, integrations, compute environments, memory, security, etc.), Claude basically joins the team in a seamless way - you can talk to it as you would talk to a person and it can help with a very large variety of workloads.

Imo this is the 3rd major redesign of LLM UIUX. The first paradigm was that the LLM is a website you go to, the second was that it is an app you download to your computer. This third one is that it is a self-contained, persistent, asynchronous entity with org-wide tools and context, working alongside teams of humans. It really takes a while to wrap your head around it, but it works and it is awesome.

[Kevin Shen @_kevin_shen]

Many such cases, can't teach hunger

[Kevin Shen @_kevin_shen]

two founders, companies in the same space

[Kevin Shen @_kevin_shen]

Founder B:
- Attended private boarding school for high school
- Does victory lap lectures at HBS despite company raising a down round
- Bloated C-suite and inflated titles
- Delayed product and churning users

[Kevin Shen @_kevin_shen]

@docmilanfar Context-aware transcription model when

[Kevin Shen @_kevin_shen]

@brianryhuang @haizelabs jokes aside, props for having the courage to speak up

[Kevin Shen @_kevin_shen]

@brianryhuang @haizelabs "By haizing"

[jpark @ icml @jparkjmc]

life whitepill is realizing you can become 95% percentile in anything if you truly dedicate yourself

case study: my bio

[julian @JulianL093]

This is not a particularly good take and is indicative of a fundamental misunderstanding of what a top-tier technical college education is suppose to offer. Preparing to understand modern AI as a Harvard or Stanford undergrad is not about learning "prompt engineering", vibe coding, or building Slop Domain-Specific Wrapper Agent #1000, all of which can be picked up in a few days if not hours.

To the contrary, the best way for a smart 18-22 year-old to understand AI is to develop a very solid intuition for undergraduate and graduate level probability, linear algebra, and classical ML. If you actually know how foundational RL topics like Q-learning work, you are 95% of the way there, and if you can't even learn that from Harvard or Stanford then this is probably a skill issue on your end.

In @boazbaraktcs's excellent ML theory seminar in 2021, I don't think I wrote more than 200 lines of code cumulatively in the entire semester yet I learned an immense amount and credit that class for sparking my interest in modern AI. A year ago I couldn't coherently tell you what a transformer was, but it doesn't matter, because when you develop proper quantitative foundations in college you can figure it out in a couple of weeks. None of this stuff is really that complicated, people just like to pretend that it is.

[Deedy @deedydas]

Being a founder used to mean something.

It meant you had an idea so bold that you had to quit your awesome job to do it. An insight so large that only a separate entity could will it into existence. An mountain so difficult to climb that you're willing to commit 7+yrs to 24/7 days to make it happen.

Today, it's not that.

1) People treat "founder" as a promotion. A status symbol. "I'm CEO bitch" energy. A career path after Stanford.
2) Founders take retirement-level secondary ($10M+) sometimes without even having PMF at Series A.
3) Acquisitions / acquihires with little to no revenue can be pretty for founders with management payouts leaving employees hanging.
4) Ex-founders who have failed can make so much in angel checks or funds they run on the side, and fail up to great roles.
5) Playbooks for fundraising are so well known that if you're in the "in" club, you can do it on vaporware. Due diligence is light. Most VCs don't / can't even check your code. And have a lot of dry powder they need to deploy.

This isn't a critique, just an observation.

In many ways, it IS the best time to be a founder. But it also makes the signal to noise ratio so low for genuinely innovative founders who don't know how to "play the game".

[Kevin Shen @_kevin_shen]

@karpathy These findings are also 100% unsurprising if you've spent anytime around the tech stack radiologists have to work with today

[Kevin Shen @_kevin_shen]

@karpathy I think many academics and researchers would benefit from actually shadowing and watching radiologists work, instead of arrogantly assuming it's all straightforward image input -> diagnosis output

[Kevin Shen @_kevin_shen]

Radiologists will be the last doctors replaced, not the first

[Deena Mousa @deenamousa]

In 2016 Geoffrey Hinton said “we should stop training radiologists now" since AI would soon be better at their jobs.

He was right: models have outperformed radiologists on benchmarks for ~a decade.

Yet radiology jobs are at record highs, with an average salary of $520k.

Why?

[Kevin Shen @_kevin_shen]

@boneGPT @jeffzwang skill issue

[Kevin Shen @_kevin_shen]

@zackkanter Awesome to see the free market coming for rent-seeking legacy healthcare parasites

[Kevin Shen @_kevin_shen]

GPT-5: "It's toasted"

[Kevin Shen @_kevin_shen]

99% of discourse around LLM performance reminds me of this clip from Mad Man

https://t.co/hfQxATFUUX

[Kevin Shen @_kevin_shen]

@WillManidis Internet adoption, the masses discovered you can get instant distribution for anything. Reward function is clicks not utility