Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Most video understanding models operate one frame at a time, which makes temporal questions somewhat challenging. STORM, which uses Mamba adapters, adds temporal attention operations. This post compares it against Qwen models.
The US military has awarded a significant contract to Scale AI and partners including Anduril and Microsoft to integrate AI agents into operations for decision-making in military workflows. The Thunderforge project aims to enhance strategic planning speed and accuracy while maintaining human oversight. The Pentagon plans to eventually deploy this AI system across all its combatant commands.
Sponsored
Learn how Ai2 dramatically increased compute capacity and performance on NVIDIA Hopper GPUs to support open source innovation using Cirrascale Cloud Services accelerated by NVIDIA. Read the case study
AMD has sent a number of large computers with MI300X GPUs in them for the tiny grad team to use to test its drivers.
Google has opened applications for its AI for Energy accelerator, a 10-week equity-free program supporting startups integrating AI into energy solutions.
A federal judge ruled that authors can move forward with their copyright lawsuit against Meta, which alleges unauthorized use of their books to train AI models.
Sponsored
Delve is the AI compliance platform that makes SOC 2, ISO 27001, HIPAA, etc. as simple as TurboTax.
While the naming might be slightly confusing, the product preview launch from Google shows the direction they want to pursue with AI-enabled search systems.
Anthropic has revamped its console to streamline AI deployment with Claude models. New features include shareable prompts, extended thinking for Claude 3.7 Sonnet, and enhanced tools for evaluating and optimizing AI-generated responses.
Larry Page is working on Dynatomics, an AI-driven manufacturing startup focused on generating optimized designs and automating factory production.
The Hazy Research team from Stanford has released a post and code for their ThunderKittens-enabled Multiheaded Latent Attention implementation. It comes out to be 30% or so faster than the official DeepSeek implementation.
DEMO³ introduces a demonstration-augmented RL framework that improves data efficiency in long-horizon robotic tasks by leveraging multi-stage dense reward learning and world model training.
Sponsored
You don't need to build a CSV importer. You don't need to deal with support tickets. You don't need to solve for edge cases and messy customer data. You need OneSchema [13]. Delight your customers with a frictionless import experience (CSV, XLSX, XML).
Babel is a multilingual LLM supporting 25 of the most spoken languages, covering 90% of the global population. It expands its parameter count through a layer extension technique that elevates its performance ceiling.
CSR optimizes sparse coding for adaptive representation, reducing computational cost while improving retrieval speed and accuracy across multiple benchmarks.
SCSegamba is a lightweight segmentation model for structural cracks that combines gated bottleneck convolution and a scanning strategy to enhance accuracy with minimal computational overhead.
A speech dataset with high-quality and diverse data.
Cohere's Aya Vision is a multimodal AI model that integrates both text and image understanding capabilities. The innovative model is designed to enhance content creation and analysis, enabling users to generate and interpret multimedia content more effectively. Aya Vision aims to push the boundaries of what's possible in AI-driven content generation and interaction.
AI's transformative role in scientific writing is undeniable but must be balanced with maintaining research integrity. Stakeholders advocate for ethical guidelines, AI detection tools, and global collaboration to ensure AI augments rather than undermines scholarship. By prioritizing transparency and accountability, the scientific community can integrate AI responsibly and uphold scholarly standards.
AI research lab, Exa, has made its search tool Websets available for public use.
SensorRy AI, a network that uses AI to detect fires when they're still small, was developed by a Stanford online high school student.
Apple is introducing AI-generated review summaries to the App Store with iOS 18.4, using LLMs to create short paragraphs that highlight key user feedback.
OpenAI has unveiled Next Gen AI, a suite of advanced tools designed to boost developers' efficiency in building AI applications. The new offering features enhanced model performance and streamlined integration options, facilitating a wide range of use cases. This initiative aims to empower developers to innovate faster and more effectively in the evolving AI landscape.
The two have done years of groundbreaking research and education in Reinforcement Learning.
The Qwen team has trained an open weight, Apache 2.0, licensed model that performs on par with DeepSeek R1 and better than many of the larger distill models. They found that by building on outcome based rewards with formal verification and test-case checks, the model can continuously improve at math and code. Also, by mixing in general instruction following data later in RL training, the model can still align with human preference.
Pipeline Parallelism is a great strategy when sharding models across many GPUs. However, it uses substantial memory. This paper, which comes with code, uses offloading to effectively reduce the memory required when the number of shards increases making PP a strong alternative to TP for large model training.
ToLo introduces a two-stage, training-free layout-to-image framework designed for high-overlap layouts.
DifIISR enhances infrared image super-resolution using diffusion models, integrating perceptual priors and thermal spectrum regulation to improve both visual fidelity and machine perception.
Sponsored
Ai2 is a nonprofit research institute dedicated to developing fully open source AI solutions. They needed to deploy a cluster of 1,024 NVIDIA Hopper GPUs efficiently while building out their data center. Cirrascale [10] delivered customized, scalable infrastructure designed
Strong voice cloning text-to-speech model built on Qwen. It supports emotive prompting along with the text input. Interestingly, the researchers found 8k tokens in the Codec to be sufficient for powerful speech generation.
FlowDec is a full-band audio codec leveraging conditional flow matching and non-adversarial training for high-fidelity 48 kHz audio compression.
Kiss3DGen repurposes 2D diffusion models for efficient 3D object generation, using multi-view images and normal maps to create high-quality meshes and textures.
Given the excitement around Claude playing Pokemon, this blog post is quite timely. It explores how to use RL to train a policy to play Pokemon. It is not a general purpose agent, but works quite well on the given task.
Israeli startup Perfect, which aims to streamline recruitment processes using proprietary AI, raised $23 million in seed funding. It claims to reduce recruiters' workloads by 25 hours weekly and has rapidly expanded its client base to 200 companies. Founded by Eylon Etshtein, the platform uses custom vector datasets, bypassing third-party LLMs, to provide accurate candidate insights.
A tech enthusiast used AI tools like ChatGPT and Claude to successfully create various apps, such as podcast transcription tools and task managers, despite being a hobbyist. This inspired him to launch a course, "Build Your Own Life Coach," to teach others how to build custom software using AI. The self-paced course has already enrolled nearly 150 students and aims to provide a comprehensive framework for software development with minimal coding experience.
Sofya leverages Llama models to streamline medical AI workflows, improving efficiency and reducing administrative overhead for healthcare providers in Latin America.
BodyGen introduces topology-aware self-attention and a temporal credit assignment mechanism to enhance the efficiency of robot morphology and control co-design.
Meta is in talks to build a $200 billion AI data center campus, with potential locations including Louisiana, Wyoming, or Texas.
CoreWeave has announced plans to acquire Weights & Biases for $1.7B. The integration aims to accelerate AI innovation by combining CoreWeave's cloud infrastructure with Weights & Biases' AI tools for model training and evaluation. The acquisition is expected to close in the first half of 2025, pending regulatory approval.
Amazon is developing an AI model with advanced reasoning capabilities under its Nova brand. Set to launch as early as June, the model is expected to use a hybrid reasoning architecture similar to Claude 3.7 Sonnet and aims to be more cost-efficient than competitors.
Cohere For AI has introduced Aya Vision, a vision model designed to enhance AI's multilingual and multimodal capabilities. It supports 23 languages.
This dataset was curated to have minimal overlap with existing video datasets while including themes and actions that users training models might be interested in for final video synthesis and understanding. All videos are collected from the official YouTube creator API and are CC licensed.
This novel reward dimension reduction method enhances learning efficiency in multi-objective reinforcement learning, scaling beyond conventional approaches.
Framework designed to substantially accelerate the generation process of ultra-long sequences, up to 100K tokens, while maintaining the target model's inherent quality.
L-MAP improves sequential decision-making in stochastic, high-dimensional continuous action spaces by learning macro-actions via a VQ-VAE model.