More than 50% of the reported reasoning abilities of LLMs might not be true reasoning. How do we evaluate models trained on the entire internet? I.e., what novel questions can we ask of something that has seen all written knowledge? Below: new eval, results, code, and paper.

1. Finetune an LLM on your training data. 2. Demo the performance on the same training data. 3. Make big claims. So typical that even annoying.

Our paper was just published in Nature Machine Intelligence In this work, we developed a foundation model for quantitative imaging/radiomic biomarkers. We then showed that features extracted from this model helped with diagnostic and prognostic applications. nature.com/articles/s4225…

Based on numerous requests, we are providing the open ShareIT link for UNI and CONCH. Please access it below: Open ShareIT Read Links: UNI: rdcu.be/dBMgh CONCH: rdcu.be/dBMf6 Journal Links for complete pdf: UNI: nature.com/articles/s4159… CONCH:

📅 Mark your calendar for the Apr. 3 #RadAIchat at 8 PM ET on "Deep Learning-based Decision Support for Neurosurgical Intervention" moderated by Suraj Pai and panelists Sven Haller Armaan Malhotra Christopher Witiw @Chriswsmith102 #radiomics #AI #DL #radres RSNA Charles Kahn, MD Hesham Elhalawani

Let's get started! T1. What role can #AI play across different stages of neurosurgical intervention? #RadAIchat

Radiology: Artificial Intelligence I think future directions will include more detailed prognostication, based on initial imaging and also more specific diagnostics (such as tumour subtyping etc.) 3/n #RadAIchat

T2. How can #AI tools like ASIST-TBI change how patients with traumatic brain injuries are managed? #RadAIchat

Radiology: Artificial Intelligence To address T2: AI tools like ASIST-TBI will serve as decision support tools to help radiologists and emergency department physicians triage patients based on the acuity of their injury and the potential need for emergency neurosurgery. 1/n #RadAIchat

A scientific conference but all submissions are in the form of a preprint at any stage of the research cycle (maybe collected in its own BioRxiv channel)— the meeting is presentation/discussion of all sorts that is essentially a peer review hackathon resulting in public comments

Announcing NeurIPS Preschool Track This year, we invite preschoolers to submit machine learning research papers.

Talking to many junior faculty members and students in AI lately. Many seem to be somewhat lost with all the seemingly fast progresses made by the industry. My suggestion to them is: It is industry's job to find how to do better, but academia is to find out how to do it right.

Daniel Dennett (28th March 1942 - 19th April 2024) RIP Dennett on how to criticise wisely

Researchers from the Artificial Intelligence in Medicine Program at Mass General Brigham have developed an AI model to uncover cancer biomarkers. This AI model has the potential to impact when cancer is detected and treated. buff.ly/3Vflc70 Hugo Aerts Suraj Pai