Jacob Rosenthal

Before modern neuroimaging, there was pneumoencephalography: CSF was drained and replaced with air and an X-ray was taken. Patients were spun upside down so that the air bubbles would float to different places in the ventricles to get multiple views. This continued into the 70s!

Ursula Le Guin and Philip K Dick, two generational literary talents, were high school classmates

Some people today are discouraging others from learning programming on the grounds AI will automate it. This advice will be seen as some of the worst career advice ever given. I disagree with the Turing Award and Nobel prize winner who wrote, “It is far more likely that the programming occupation will become extinct [...] than that it will become all-powerful. More and more, computers will program themselves.”​ Statements discouraging people from learning to code are harmful! In the 1960s, when programming moved from punchcards (where a programmer had to laboriously make holes in physical cards to write code character by character) to keyboards with terminals, programming became easier. And that made it a better time than before to begin programming. Yet it was in this era that Nobel laureate Herb Simon wrote the words quoted in the first paragraph. Today’s arguments not to learn to code continue to echo his comment. As coding becomes easier, more people should code, not fewer! Over the past few decades, as programming has moved from assembly language to higher-level languages like C, from desktop to cloud, from raw text editors to IDEs to AI assisted coding where sometimes one barely even looks at the generated code (which some coders recently started to call vibe coding), it is getting easier with each step. I wrote previously that I see tech-savvy people coordinating AI tools to move toward being 10x professionals — individuals who have 10 times the impact of the average person in their field. I am increasingly convinced that the best way for many people to accomplish this is not to be just consumers of AI applications, but to learn enough coding to use AI-assisted coding tools effectively. One question I’m asked most often is what someone should do who is worried about job displacement by AI. My answer is: Learn about AI and take control of it, because one of the most important skills in the future will be the ability to tell a computer exactly what you want, so it can do that for you. Coding (or getting AI to code for you) is a great way to do that. When I was working on the course Generative AI for Everyone and needed to generate AI artwork for the background images, I worked with a collaborator who had studied art history and knew the language of art. He prompted Midjourney with terminology based on the historical style, palette, artist inspiration and so on — using the language of art — to get the result he wanted. I didn’t know this language, and my paltry attempts at prompting could not deliver as effective a result. Similarly, scientists, analysts, marketers, recruiters, and people of a wide range of professions who understand the language of software through their knowledge of coding can tell an LLM or an AI-enabled IDE what they want much more precisely, and get much better results. As these tools are continuing to make coding easier, this is the best time yet to learn to code, to learn the language of software, and learn to make computers do exactly what you want them to do. [Original text: deeplearning.ai/the-batch/issu… ]

Huge update to our preprint today on the superhuman performance of reasoning models in medical diagnosis! TL;DR – they don't just surpass humans in meaningful benchmarks, but in actual medical care from unstructured clinical data: A 🧵⬇️: x.com/AdamRodmanMD/s…

AI + expertise paradox. More technical and specialized medical tasks -> easier for AI to substitute & outperform _average_ human experts. Case in point is echocardiogram interpretation. Yet it is those tasks that are better paid/reimbursed vs general primary care. HT @venkmurthy

The only technical detail Google releases for this project is that they’re using a python version that’s been obsolete for 5+ years? 🤔

Queensborough Bridge by Edward Hopper (1913). Almost exactly 100 years later this spot would go on to become the home of @cornell_tech. I wonder what the future holds for Roosevelt Island in 2113?

When using GPT to help debug code recently, the first thing it suggests is always numerical precision error. It seems to have a poor feel for general magnitudes of numbers. No ability to gauge the difference between a "small" number like 0.006 and a SMALL number like 6e-20

Finally passed 2000 ELO in bullet on @lichess after more than 13k games played over the last 10 years. I won’t say it was time wasted, but it did make me reflect on these words from Paul Morphy, one of the great American chess masters 😅

This piece from more than 30 years ago is incredible - replace "computers" with "AI" and it could have been written yesterday. I predict that it will still be relevant 30 years in the future for whatever the next big technology leap will be. academic.oup.com/jamia/article-…

@Gopal_Kot Won’t AI decrease barrier to entry for apps, reducing the advantage of large labs? A couple undergrads + Claude can build apps now that would have taken whole teams a couple of years ago

If the AI arms race continues to be this tight long term, expect the large AI labs to pivot to the app layer.

When patients have questions about their health, they prefer answers from AI over answers from human physicians. Helps explain why patients are increasingly turning to chatbots for medical advice! jamanetwork.com/journals/jaman…

The people need to know if DeepSeek R1 can pass USMLE Step1!! 🥲

So is every“X beats human expert performance on Y” paper going to be re-run each time a splashy new model comes out?? Where: X = ChatGPT, GPT-4, o1, etc Y = medical MCQ benchmark of questionable clinical relevance

@frootparty 1 good paper >>>>> a bunch of bad papers

Completely misguided to think that these 3000 questions capture the entire "frontier of human knowledge." Still, very useful to curate challenging benchmarks like this. Models will benefit from training on these difficult questions (cf. hard example mining, boosting).

🌶️ Hot take: the value function for the number of papers a person publishes is not monotonically increasing! My own mental model is shaped more like a mesa, with x-axis scaled depending on seniority/circumstances. There is such a thing as having too many papers!

Many people in my med school skip lectures and instead invest their time in churning out clinical research projects. This is a very rational decision given the incentives for applying to residency but a net negative for medicine in my opinion. Need to change incentive structures!

There are obviously lots of issues with relying on LLMs for med decision making. But they also provide so much value by making medical knowledge more accessible! Balancing these interests to build trust and provide better care will be a big challenge for medicine in coming years

Areas that I think will be increasingly important for human doctors as AI continues to get better and better: - Communication / interviewing - Embodied tasks (e.g. physical exam, procedures) - Teamwork - Research (creating new knowledge) - Using AI tools effectively