Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
AsyncDiff is a new acceleration scheme that enables parallel computation in diffusion models. It significantly reduces latency without compromising quality by dividing the noise prediction model into multiple components and running them on separate devices.
Researchers have tackled the efficiency issues in autoregressive large language models by exploring linear attention methods combined with speculative decoding. This study introduces an augmentation technique for linear attention, making it compatible with speculative decoding and improving both training and performance.
Meta is upgrading its global data centers to support AI demands. It is planning to scale to 600,000 GPUs for AI training jobs. This involves innovative maintenance strategies and tools like OpsPlanner to ensure minimal disruptions and consistent performance while enabling rapid infrastructure scaling.
Sponsored
Social networks make it hard to delete your history. Redact makes it easy. Automatically remove your content from Twitter, Reddit,
Fast inference on the phone for the special Mistral 47B MoE model.
A library built on top of JAX and XLA that supports the development of large-scale deep learning models.
Perplexity is planning revenue-sharing deals with high-quality publishers for a recurring income stream.
Sponsored
Builders and GenAI developers, this one's for you!
Mistral, a Paris-based AI startup, turns 1 year old today. It announced a monster series B funding raise.
Elon Musk has dropped the lawsuit against OpenAI and Sam Altman that claims they have deviated from their original mission in favor of profit. He has the option to file again in the future.
Together has introduced Mixture of Agents (MoA), a novel method that combines multiple LLMs for superior performance that achieves a 65.1% score on AlpacaEval 2.0, surpassing GPT-4o. MoA uses a layered structure where initial responses from various models are refined by aggregators in successive layers, enhancing output quality through collaboration. Despite higher accuracy, MoA faces latency challenges. Future directions include optimizing model architecture and reducing latency.
Most compute for Transformers is spent in the Linear layers. This work builds on muP and Monarch matrices to create a structured representation that uses less compute and has better scaling laws than naive dense layers.
CARES is a comprehensive framework that evaluates the trustworthiness of Medical Large Vision Language Models (Med-LVLMs).
PHATGOOSE is a method that significantly improves an AI's ability to generalize and learn new tasks without prior exposure by smartly routing between different specialized language models for each part of a task.
Sponsored
In this webinar: AI thought leaders Steve Nouri and Arun Nandi, and AtScale CTO Dave Mariani, will discuss ways to use generative AI to increase data and analytics accessibility [12] for all enterprise users. Key topics will include NLQ, co-pilot features, and model
The Mistral.rs Rust-based inference framework has improvements to its speed, GGUF support, and more in the latest release.
The Human Gaussian Model (HGM) is a diffusion-guided framework for creating 3D human models from a single image.
LE3D is a method for real-time HDR view synthesis from RAW images. It is particularly effective for nighttime scenes.
Researchers have developed a novel framework that improves the grading of Diabetic Retinopathy (DR), a condition that can lead to visual impairment.
How overfit are popular LLMs on public benchmarks? New research from Scale AI SEAL revealed that Mistral & Phi are overfitting benchmarks, while GPT, Claude, Gemini, and Llama are not. The researchers produced a new eval GSM1k and evaluated public LLMs for overfitting on GSM8k.
The inaugural Mozilla Builders Accelerator is embracing Local AI, which brings AI models and applications directly onto personal devices to enhance privacy and accessibility. Key innovation areas include tools for developer productivity, local-based AI agents, dynamic UIs, adaptation of fine-tuning, retrieval-augmented generation, and advanced function calling. The initiative seeks participants to develop a decentralized, open-source AI ecosystem that prioritizes user empowerment.
Sponsored
Modern authentication for AI applications. Bot and abuse protection included in every plan. Industry-forward pricing, including only
Haize Labs announced on Twitter a suite of “haizing” (i.e. red-teaming and fuzzing) algorithms against industry-leading, safety-trained AI companies.
CaseMark, an AI startup revolutionizing legal workflows, has secured a pre-seed investment led by Gradient Ventures to expand its AI offerings for the legal industry.
OpenAI plans annual opportunities for shareholders to sell equity, but there are concerns over the company's restrictive approach.
Sponsored
If you want to see the future of AI-powered hospitality, look no further than Jurny [5]. And if you want to invest in that future, now's your chance—but act fast, this opportunity ends June 18 [4].
Apple announced at WWDC 2024 that it is integrating ChatGPT into a variety of native experiences on iOS devices. It plans to eventually expand to use voice and other capabilities.
Apple has unveiled its "Apple Intelligence" AI suite, which has features like a conversational Siri, AI-generated "Genmoji," and GPT-4o integration for complex requests. The capabilities will be exclusive to newer iPhones, iPads, and Macs. They will roll out in English with iOS 18, iPadOS 18, and macOS Sequoia this fall. Apple has emphasized privacy with on-device processing and "Private Cloud Compute" for off-device tasks and allowed independent privacy verification.
The Gboard team has outlined how they use SFT on a PaLM2-XS model to fix sentence and paragraph-level errors in written text on device. They found dramatic improvements and usage due to latency optimizations.
The team at Snap Research was able to reduce the size of the Stable Diffusion UNet model from 1.72 GB down to 219MB while increasing performance with their new quantization scheme. The quantization method is somewhat complex, but paints a strong path forward in running generative models on consumer hardware.
Apple introduced "Apple Intelligence" at WWDC 2024. Apple Intelligence is an AI system integrated into iOS 18, iPadOS 18, and macOS Sequoia that features advanced generative models for various everyday tasks like text refinement, notification summarization, and image creation. Emphasizing responsible AI development and user privacy, this system combines on-device and cloud capabilities to enhance user experiences across Apple products.
Sponsored
Data leaks happen every day — leading to spam, fraud, identity theft, and worse. Surfshark's Alternative ID [11] keeps your email address and name private, minimizes spam emails, and keeps your information out of data breaches. Sign up today [11] for 80% off + 3
A Jupyter Notebook that combines the experience of OpenAI's code interpreter with the familiar development environment of a Python notebook.
OVMR is a new method that enhances open-vocabulary recognition by combining textual descriptions and exemplar images.
The Predictive Dynamic Fusion (PDF) framework enhances multimodal learning by addressing reliability and stability issues.
AI image generation has progressed from creating images based on text descriptions since 2022. Using a child's game analogy, this article explains how these models refine noisy inputs to produce detailed and specific images, showcasing the rapid advancements and potential of AI in visual creativity.
Researchers have introduced a new framework that integrates active stereo principles into standard passive cameras without using a physical pattern projector.
Asana has introduced "AI teammates" to handle tasks like triaging requests and organizing project details proactively, aiming to improve efficiency and output quality. This new feature, demonstrated at Asana's Work Innovation Summit, is integrated into the workflow, acting like a human team member but remaining under human supervision.
Great model from Tsinghua KEM group trained on 10T tokens with support for 26 languages.
DIRECT-3D is a new diffusion-based 3D generative model that creates high-quality 3D assets from text prompts.
Your personal AI expert on all crypto & blockchain-related topics.
Sponsored
Odds are, you've received a suspicious phone call, text, or email asking for money. The scam might have been easy to spot — but with deepfakes and genAI, things are getting much weirder.
Apple is set to introduce its AI initiatives under the brand name "Apple Intelligence" in iOS 18 and other operating systems, focusing on daily life enhancements rather than creative tasks. Features will include AI-generated notification summaries, rich auto-replies in Messages and Mail, intelligent email categorization, an emoji creator, AI-powered photo editing, and transcriptions in Voice Memos.