Comfy AI @ComfyAI_App

A blog post just published a week ago by Thinking Machines Lab provides an insightful and detailed discussion on the problem of "indeterminism" in LLM inference. Typically, when we use various large models and ask the same question multiple times, the answers will always be…

A paper published at the beginning of the year with Kaiming as the corresponding author introduces a new concept called "Fractal Generative Models." The results look pretty cool and seem to open up a new direction for generative models. The core idea is actually quite simple:…

A new study from Stanford University, recently published on arXiv and titled "Paper2Agent," could change the way we read papers and use code in the future. We all know how wide the gap is between reading a paper—especially a methods-focused one in computer science—and actually…

The most interesting part of this paper is that it points out a core pain point in the current evaluation of large medical models: everyone is competing for high scores on static exams like the USMLE, but this is vastly different from real, dynamic clinical consultation…

First of all, what is inductive bias? Inductive bias is essentially the set of assumptions that a machine learning algorithm makes about the patterns in the data. Every model has this built in. Let's take common machine learning algorithms as examples. For instance, linear…

Everyone is now striving to make large models smaller and more efficient so they can run directly on end-user devices like mobile phones. The Meta AI team has just open-sourced their latest MobileLLM-R1 model series. The largest model in this series, with 950M parameters, was…

I just saw a hot-off-the-press preprint on arXiv from MBZUAI that offers an interesting new direction for training large models. Currently, mainstream language model training, whether it's the classic next-token prediction (NTP) or the multi-token prediction (MTP) later proposed…

[ArxivSub] Daily Arxiv - Biomedical Data + Large Language Model 🏠More papers on homepage/collection 🔧Use ArxivSub to find papers 🌐arXiv ID: 2509.09064 📚Paper Title: Enhancing 3D Medical Image Understanding with Pretraining Aided by 2D Multimodal Large Language Models…

keyboard_arrow_left Previous keyboard_arrow_right Next

I found a review paper on arXivSub from last month titled "Speed Always Wins," which systematically sorts through various cutting-edge methods for breaking through the efficiency bottlenecks of the traditional Transformer architecture. For anyone interested in the underlying…

[ArxivSub] Daily Arxiv - Biomedical Data + Benchmark 🏠More papers on homepage/collection 🔧Use ArxivSub to find papers 🌐arXiv ID: 2509.08087 📚Paper Title: Performance Assessment Strategies for Generative AI Applications in Healthcare 💡What's About: This paper reviews and…

keyboard_arrow_left Previous keyboard_arrow_right Next

[ArxivSub] Daily Arxiv - Biomedical Data + Large Language Model 🏠More papers on homepage/collection 🔧Use ArxivSub to find papers 🌐arXiv ID: 2509.08087 📚Paper Title: Performance Assessment Strategies for Generative AI Applications in Healthcare 💡What's About: This paper…

keyboard_arrow_left Previous keyboard_arrow_right Next

I saw an old interview with Fei-Fei Li from A16Z on X, where she brought up a fascinating point: that the world originally had no language or writing. She’s right; when you walk through a forest, there aren't words floating on the trees, and the sky doesn't contain sentences.…

Recently, how to enable Large Language Models (LLMs) to "self-improve" through multiple iterations during the reasoning phase has become a hot research topic. For example, when solving a difficult problem, the model can think, check, and revise repeatedly to ultimately provide a…

[ArxivSub] Daily Arxiv - Biomedical Data + Benchmark 🏠More papers on homepage/collection 🔧Use ArxivSub to find papers 🌐arXiv ID: 2509.07260 📚Paper Title: HealthSLM-Bench: Benchmarking Small Language Models for Mobile and Wearable Healthcare Monitoring 💡What's About: This…