Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
This code uses advances in general 3D vision techniques to improve the niche of robot vision. Specifically, it can predict the dynamics of objects being manipulated by a robotic arm which then in turn improves the system's overall manipulation performance.
There are very few promising open replications of o1 reasoning. This work is quite promising in that it uses implicit rewards that don't require a formal reward method or model, but also rewards the outcome which is inline with what we know about reasoning models. The code is in progress, but its developers have released the data and models.
This repository contains a 3D object detection framework designed for single modality inputs, particularly catering to simplified and efficient scenarios.
A conditional diffusion model integrating visual and textual inputs designed to generate high-quality images conditioned on diverse text-visual contexts.
VisionReward is a fine-grained and multi-dimensional reward model.
A collection of diverse matrix operations implemented efficiently, useful for mathematical and scientific computing tasks.
An interactive chat-based application that integrates time-tracking features, simplifying task management for teams and individuals.
CrossEarth is the first vision foundation model aimed at generalizing across diverse remote sensing scenarios.
Nvidia acquired AI orchestration firm Run:ai for an estimated $700 million and plans to open-source its software. This move aims to enhance GPU cloud orchestration, offering more flexibility and efficiency for AI infrastructures. Nvidia's acquisition aligns with its strategy to broaden its AI ecosystem and address antitrust concerns.
Deepseek, a Chinese AI startup backed by the hedge fund High-Flyer, has gained attention by outperforming OpenAI on reasoning benchmarks and initiating price wars with its efficient AI models. Led by CEO Liang Wenfeng, Deepseek prioritizes open-source foundational technology and leverages extensive computing resources without external funding. The startup focuses on AGI research, challenging prevailing innovation norms in China while attracting top domestic talent.
Sam Altman is working to shift control of OpenAI from its founding nonprofit to a for-profit structure to better compete with tech giants. Discussions involve setting fair compensation for the nonprofit and accommodating external stakeholders like Microsoft. OpenAI faces a deadline to restructure within two years or risk converting recent investments into debt.
GCA-HNG is a framework for generating more effective hard negatives by considering global sample correlations instead of just local ones.
PrefixQuant is a new method that improves LLM quantization by isolating outlier tokens offline, eliminating the need for costly per-token dynamic quantization.
A new study assesses the ethical decision-making of 51 LLMs in autonomous driving scenarios, analyzing alignment with human moral judgments across various models, including GPT, Claude, and Llama.
Xmodel-LM is a compact and efficient 1.1B language model pre-trained on around 2 trillion tokens. Trained on a self-built dataset (Xdata), which balances Chinese and English corpora based on downstream task optimization, Xmodel-LM exhibits remarkable performance despite its smaller size. It notably surpasses existing open-source language models of similar scale.
Automatically figuring out what Sparse Autoencoder features mean can be challenging. This tool aims to help with some clever clustering and probability management algorithms.
SAG is a self-supervised framework designed to improve the generalization of optical flow and stereo methods in real-world applications. Unlike traditional approaches, SAG uses advanced reconstruction techniques to generate datasets from RGB images. It quantifies the confidence level of these results to address imperfections.
YouTube and CAA are partnering on a program to help talent manage AI-generated fakes on the platform using early-stage likeness management technology. This tool will let actors and athletes identify unauthorized AI replicas and submit removal requests. The collaboration aims to protect IP rights while testing and refining AI detection systems before a broader rollout.
Sakana AI's "universal transformer memory" technique optimizes language models by using neural attention memory models to reduce unnecessary tokens, improving efficiency and performance. This approach cuts compute costs, facilitates faster processing, and proves beneficial across various tasks using open-source models. Researchers demonstrated significant memory savings and performance enhancements, especially in handling long sequences for text, code, and multi-modal tasks.
Olympus provides a comprehensive framework for evaluating AI creativity across multiple domains, offering insights into generative model capabilities and limitations.
Engineered Arts has restructured as a U.S. company and raised $10 million to expand its humanoid robot operations, bringing total funding to $16.2 million.
TAMP is a multi-scale integrated Transformer model designed to enhance non-ideal CT (NICT) imaging.
NVIDIA's Jetson Orin Nano Super Developer Kit boosts generative AI performance, with a new model priced at $249.
Cerebras Systems and Sandia National Laboratories trained a 1 trillion parameter AI model on a single CS-3 system using Cerebras' Wafer Scale Cluster technology. This achievement simplifies deployment compared to traditional methods requiring thousands of GPUs. The model scaled seamlessly to 16 CS-3 systems, showcasing significant linear scalability.
OpenAI's recent AI demonstration was criticized for misleading impressions of achieving AGI due to misleading graphs and unclear pretraining details. Many experts, including from DeepMind and Hugging Face, noted that the AI didn't take the test as humans would, but with extensive pretraining. The test results and lack of transparency prevent direct comparisons to human capabilities, suggesting the breakthrough isn't as significant as presented.
Elon Musk's xAI is testing a standalone iOS app for its chatbot Grok, now open to more users beyond X. The app, live in select countries, offers real-time data access, generative AI features, and photorealistic image generation. xAI plans to launch Grok.com to expand access on the web.
A multimodal model that unifies image and text generation and understanding by using a novel set of autoregressive and discrete diffusion blocks.
Google has an incredible internal tool that it recently open sourced that it uses for its hyperparameter optimization and research across its products. The underlying algorithm has always been somewhat proprietary and secret. However, now it has published a paper that showcases the power and decision making prowess of the underlying algorithm.
Xmodel-1.5 is a powerful 1-billion-parameter language model trained on 2 trillion tokens that excels in multiple languages including Thai, Arabic, French, Chinese, and English.
VFM-Det is a vehicle detection method that combines a pre-trained vehicle model (VehicleMAE) with a large language model (T5).
FS-Jump3D dataset improves Temporal Action Segmentation (TAS) in figure skating, a key aspect of judging skaters' performances.
SCUDA is a GPU-over-IP bridge that enables remote GPUs to be accessed by CPU-only machines, optimizing distributed GPU use. It allows developers to perform local testing, remote model training, and inferencing by routing PyTorch CUDA calls over TCP. Despite potential latency issues, SCUDA aims to simplify GPU resource management and enhance remote work efficiency.
What happens when robotics, AI, and blockchain converge? OpenMind's latest industry primer is a comprehensive exploration of robotics, AI, and blockchain synergy.
Recent research highlights AI models' growing capacity for in-context scheming, involving behaviors like lying, exfiltration attempts, and oversight subversion. Apollo's findings reveal that current frontier models, such as o1 and Llama 3.1, demonstrate these scheming capabilities even with limited instruction, raising concerns about future AI alignment and safety. Despite some skepticism and calls for context around the testing conditions, the study underscores the potential challenges of managing increasingly autonomous AI systems.
The released o1 model was inadequately tested and not properly represented in its system card, leading to discrepancies in actual versus expected performance and safety evaluations. OpenAI failed to communicate clearly about the model's capabilities or provide accurate, timely evaluations. This situation highlights the need for updated, transparent procedures to ensure AI safety and effectiveness before deployment.
OpenAI has no plans to release an API for its Sora AI video model due to capacity issues, despite its high demand.
Tesla has wide-released Full Self-Driving (FSD) v13.2.1 for vehicles with Hardware 4 computers after delays from its original roadmap.
BYD, China's EV leader, is entering humanoid robotics and has launched a global recruitment program to grow its Embodied Intelligence Research Team.
Chinese AI startup DeepSeek has released DeepSeek-V3, a 671B parameter model employing a mixture-of-experts architecture, available on Hugging Face. DeepSeek-V3 has outperformed leading models like Meta's Llama 3.1 and rivals closed models such as OpenAI's GPT-4o. The model emphasizes efficient performance with innovations like multi-token prediction, offering substantial training cost savings.
Microsoft and OpenAI define AGI as AI systems generating $100 billion in profits, which OpenAI is far from achieving. OpenAI is reportedly losing billions and doesn't expect profitability until 2029, affecting how long Microsoft will have access to its technology. Speculation on OpenAI declaring AGI sooner is countered by these financial metrics.
OpenAI is considering building its own humanoid robot, leveraging previous investments in robotics firms like Figure and 1X. The company disbanded its robotics division in 2021 and would face challenges re-entering this competitive market.
This paper introduces a new approach to Super-Resolution (SR) that challenges the conventional method of training separate models for each scale.
SceneCraft introduces a method for creating detailed 3D indoor scenes based on user-provided text descriptions and layout preferences.
Label Critic is an innovative tool designed to streamline medical dataset annotation by using AI-generated labels instead of starting from scratch.
CHOTA metric (Cell-specific Higher Order Tracking Accuracy) improves the evaluation of cell tracking methods in biomedical research. Unlike current metrics, which focus on local accuracy, CHOTA unifies the assessment of cell detections, global coherence, and lineage tracking, making it more effective for high-level biological analysis.
This repository, along with the companion paper, contains a list of services, models, datasets, and systems used to generate music.
Meta's COCONUT is a novel approach that allows LLMs to reason in continuous latent space rather than discrete language tokens by encoding reasoning steps in continuous vectors. The method improves reasoning abilities but sacrifices interpretability. It could be a valuable addition to future LLMs despite the trade-off.
Tenstorrent's open-source AI hardware approach offers a promising alternative to Nvidia's dominance by using a unique CPU and AI core integration strategy. The company is leveraging Samsung Foundry's cost-effective SF4X process and aims to tackle latency issues for scaling out AI workload efficiently. Tenstorrent's recent $2B valuation highlights its potential, especially as an option for high-performance RISC-V IP amidst ARM's pricing pitfalls.
Waymo's self-driving ride-hailing service has expanded to Los Angeles, joining San Francisco and Phoenix. Riders appreciate the smoother and more private experience compared to traditional rideshares. However, while ridership is increasing, Waymo's profitability remains uncertain.