PhD Student in Computational Precision Health at @UCBerkeley
Joined December 2021
- Tweets14
- Following244
- Followers48
- Likes159
This paper has been a major part of my life for the past four years.
In our Nature paper, we show that AI applied to routine ECGs can identify patients at high risk of sudden cardiac death before tragedy strikes and reveal a visible ECG biomarker hiding in plain sight.
My friend the brilliant @kevin_volpp dropped dead
He survived (read abt it: tinyurl.com/ypcpdzpz)
But every year 300-400,000 in the US don't
This week: our @nature article (+A.Schubert J.Ross @m_sendhil M.Lingman) applies AI to this medical mystery
nature.com/articles/s41586-0…
Alex Schubert retweeted
Berkeley researchers developed a more accurate way to predict sudden cardiac death risk using AI and hundreds of thousands of EKGs.
bit.ly/4aTD7Zl
Alex Schubert retweeted
I see a lot of bad "AI predicts your death" papers. Most are a neural net mining spurious correlations from a biobank. This one in Nature this week is the rare exception, and it's genuinely excellent: an AI read sudden-cardiac-death risk off a routine ECG that doctors miss. 🧵
Alex Schubert retweeted
A new study published in Nature uses artificial intelligence to identify people at high risk for sudden cardiac death, and pinpoints a possible reason. trib.al/v7f83MJ
Alex Schubert retweeted
A very impressive discovery of a new ECG marker for sudden cardiac death validated in 3 different cohorts and linked to benefit of defibrillator, an outgrowth of human research ingenuity and AI deep learning nature.com/articles/s41586-0…
Alex Schubert retweeted
Everyone says the latest AI agents will be "job-ready" soon, especially after the release of Fable 5 this week. But is that really the case?
Over the past many months, my group and collaborators have been building Agents' Last Exam (ALE), a benchmark designed to test exactly that claim on real digital labor-market work.
My group and collaborators previously have created many of the benchmarks the field runs on, including MMLU, MATH, CyberGym, and ExploitGym. Today, I'm excited to share Agents' Last Exam (ALE): a rolling benchmark that measures whether AI agents can actually perform economically valuable work across a broad range of real-world domains.
With ALE, we evaluated Fable 5, GPT-5.5, Composer 2.5, and other frontier agent systems across more than 1,500 expert-sourced tasks spanning 55 occupations.
The result is both impressive and sobering.
Today's agents can solve a meaningful fraction of professional tasks. But when we look at the hardest tasks, the ones requiring sustained reasoning, deep domain expertise, and reliable execution over long horizons, they are still far from human-level performance.
On ALE's hardest tier, every frontier agent we tested, including Fable 5, achieved a 0% success rate.
The age of useful agents is here.
The age of truly job-ready agents is not.
We hope Agents' Last Exam (ALE) will serve as a new guidepost and north star for developing agents capable of reliably performing economically valuable work across a broad range of domains.
🧵
Alex Schubert retweeted
“AI agents will outperform humans at almost all jobs by 2026–2027.” - The forecast is everywhere.
So we built the exam to test that claim, on real labor-market aligned work. On the hardest tier, top agents pass 2.6%.
Meet Agents' Last Exam (ALE), a rolling benchmark measuring whether agents can actually do real jobs. 🧵👇
How can reinforcement learning better leverage ICU biomarker signals for decision-making? Our new paper introduces PrunedQL—a two-stage algorithm that identifies actions to avoid using these signals, then optimizes patient outcomes. Check it out!
ieeexplore.ieee.org/document…
We evaluate our method in off-policy and offline settings using simulated environments and real health records from intensive care units, showing strong performance compared to common offline RL techniques.
This work was co-led by Ali Shirali and me, along with @_ahmedmalaa.
For more information see: ieeexplore.ieee.org/document…
arXiv: arxiv.org/abs/2306.08044v3
Alex Schubert retweeted
Excited to share the camera ready of our #ICLR2024 paper "InstructCV: Instruction-Tuned Text-to-Image Diffusion Models as Vision Generalists" led by @yule_gan, Sungwoo Park and @alexmschubert
Paper: arxiv.org/abs/2310.00390
Code: github.com/AlaaLab/InstructC…
Alex Schubert retweeted
Excited to share a new preprint! InstructCV is a language interface for vision tasks trained by instruction-tuning a text-to-image generative model #StableDiffusion 🧵
📌 Paper: arxiv.org/abs/2310.00390
📌 Code: github.com/AlaaLab/InstructC…
📌 @Gradio demo: huggingface.co/spaces/alaa-l…
Alex Schubert retweeted
Generative models can synthesize medical images but assessing quality & clinical validity is challenging. In this work w/ @Harry_S_Sun @GGoldgof @atulbutte, we use human feedback for evaluation & finetuning... 🧵 (1/4)
Paper: arxiv.org/abs/2306.12438
Code: github.com/AlaaLab/pathologi…