Dragonfly's zoom-and-select architecture is particularly helpful for biomedical images, which are large and the key features are localized. Dragonfly-Med achieves 92% accuracy on Path-VQA compared to 83% for Med-Gemini. It also achieves new SOTA on medical captioning. 🎯

Actually, really liked the Apple Intelligence announcement. It must be a very exciting time at Apple as they layer AI on top of the entire OS. A few of the major themes. Step 1 Multimodal I/O. Enable text/audio/image/video capability, both read and write. These are the native

This post got a lot of attention. I wanted to add more, here's my advice of how to go from no experience in ML to getting hired (from someone who got a FAANG job as an ML scientist without any CS or ML degrees) Classes/Resources to follow: Getting hired for an intro level ML job

I recently shared my thoughts on the potential impacts of Gen AI in healthcare. Check it out! gastroenterologyadvisor.com/features/gener… Gastroenterology Advisor

Claude 3.5 Sonnet, like Devin before it and GPTs before that, is giving us a hint of what practical AI agents might look like. We don't need sophisticated AI reasoning or action if systems are built to work with people - taking direction and asking for feedback is a good start.

New paper with students Barnard College on testing orthogonal thinking / abstract reasoning capabilities of Large Language Models using the fascinating yet frustratingly difficult The New York Times Connections game. #NLProc #LLMs #GPT4o #Claude3opus 🧵(1/n)

Applications are open for the new #AI Scholars Program, a one-year training in AI to empower you to drive positive change. It includes in-person and virtual education, literature review, mentorship and a team research project. Apply by July 12! hubs.ly/Q02DX0Ps0 #GITwitter

I’m excited to co-direct this amazing opportunity for early career and trainee GIs to become the next generation wave of AI innovators and leaders. We are rolling the program out on an accelerated schedule, so make sure to apply soon! (deadline: July 12)

When you look at LLM research groups focused on selling to enterprise (e.g., Salesforce, Cohere, and Databricks), there is a heavy emphasis upon: - Smaller models - Function calling - RAG For example, Cohere has the Command-R model, Databricks has their new agent framework

The main benefit of having extensive programming experience isn't the ability to write software (which is a bit of a commodity -- you could hire someone to do it), it's how it changes the way you think.

A Survey of Prompt Engineering Methods in LLMs Huge collection of prompt engineering methods for all sorts of NLP tasks.

Can LLM's allow us to leave standards behind? "Do We Need Data Standards in the Era of Large Language Models?" NEJM AI ai.nejm.org/doi/10.1056/AI… Gabriel Brat Josh Mandel, MD Matthew McDermott forward looking editorial commenting on "FHIR-GPT Enhances Health Interoperability with

As an llm practitioner, this talk is the most informative one I ever listened to so far in the area of the science of LLMs. I was deeply impressed. For my own students I would require them to memorize the entire content of the talk. I initially read the first part of this talk

I've been saying that the future of document RAG is multimodal indexing/retrieval, and ColPali by Manuel Faysse is big step towards that direction. ⚠️ No more text-extraction ✅ Use a fine-tuned vision model with a late interaction mechanism ✅ The model is small and outperforms

Absolutely huge for healthcare. Open-source models can run in any computing environment already approved for healthcare data!

Amazing! In addition to improved segmentation, this new version can run on video in real time and be directed with prompts! I’m looking forward to the GI adaptions!

The AI Scholars met for the first time last weekend! Launched by AI Institute for GI, this program seeks to meet the demand for high-quality GI care through #AI. They learned about AI in medicine, key AI principles & more. For program details: hubs.ly/Q02PnQg_0 #AIinGI

Since the introduction of electronic health records (EHRs), US healthcare has seen zero productivity gains, while nearly every other sector has improved output per worker. Healthcare remains stagnant. Imagine a restaurant implementing software for seating, orders, and billing