Tag: #visual-reasoning12 results
He Kaiming's team published multiple CVPR papers advancing Flow Matching over diffusion models, achieving single-step FID 1.72 with Improved MeanFlow and 1.78 with JiT on ImageNet. JiT directly predicts clean images using large-patch ViT without VAE. VARC reaches human-level ARC visual reasoning without language.
This article provides a hands-on review of Claude's latest model, Fable 5. It highlights significant improvements in visual reasoning, specifically noting the model's ability to accurately count fingers in generated images.
Andrew Dai, a core contributor to Google's Gemini and PaLM, has left Google to found Elorian AI. The startup focuses on multi-modal visual reasoning models, moving beyond pure LLM architectures.
CVPR 2026 multimodal papers redefine vision intelligence: VideoAuto-R1 enables on-demand reasoning, LIVR performs latent visual reasoning sans language, and ARC reframed as vision task. Trends emphasize efficient, implicit inference over always-on chains. Evaluations move to open-ended tasks with structured data.
OpenAI introduces ChatGPT Images 2.0, a state-of-the-art image generation model. It features improved text rendering, multilingual support, and advanced visual reasoning capabilities.
Liu Zhuang and Danqi Chen released an open-source universal visual reasoning RL framework. It achieves SOTA performance using zero thinking data. Broad data drives scaling in visual reasoning RL.
Lang2Act enhances Vision-Language Models (VLMs) via self-emergent linguistic toolchains for fine-grained visual perception in VRAG, avoiding rigid external tools and info loss from image ops. It employs a two-stage RL framework: first to build a reusable action toolbox, second to exploit it for reasoning. Achieves >4% performance gains; code at GitHub.
Andrew Dai, a former Google Deepmind researcher, has launched a new AI startup named Elorian. The company is currently focused on developing visual reasoning capabilities and is actively pursuing hiring and funding initiatives.
Former Google DeepMind researcher Andrew Dai is launching a new startup focused on visual AI. He criticizes big lab AI models, saying they possess the intelligence of a 3-year-old when processing visual prompts.
AdaReasoner, a 7B model, achieves superior performance on visual reasoning tasks like puzzles by dynamically learning tool selection, timing, and usage. It introduces 'Agentic Vision' with iterative think-act-observe loops, outperforming larger models without massive scaling. Open-source code, models, and paper available on arXiv and GitHub.