Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
OpenAI is reportedly planning a major announcement on May 9, possibly unveiling a new search engine that could challenge Google. This engine may use Microsoft Bing's infrastructure, aligning with OpenAI CEO Sam Altman's vision of revolutionizing information discovery beyond the current Google model. If true, this could significantly disrupt the search engine market and how users find information online.
Dr. Brenden Lake of NYU aims to bridge human cognitive studies and AI by recording his daughter's POV to develop a language model, challenging the current AI approach that relies heavily on large datasets. This work explores how AI can learn from the sensory input akin to a toddler's experience. The research could lead to AI capable of conceptual mapping through association, similar to human language acquisition.
A novel dataset and trained visual language model that enables higher quality instruction following over multiple images.
FeNNol is a cutting-edge library that simplifies the creation and deployment of hybrid neural network potentials for molecular simulations.
Spider is a novel unified model designed to enhance the understanding of context-dependent (CD) concepts such as camouflaged objects and medical lesions, which depend heavily on visual context.
Sponsored
OctoAI's OctoStack enables you to bring GenAI models to your datastore, ensuring privacy and security while maximizing the value of your company's true moat — your business data.
Many modern performance benchmarks rely on GPT-4 as a judge of generation quality. Prometheus is a model built on top of Mistral that performs extremely well on this task.
Researchers have developed a new algorithm called RaffeSDG that improves the accuracy of medical imaging models when analyzing data from different sources.
Bonito is a model and toolkit designed to take unstructured text as input and create certain types of instruction datasets like question answering, instructions, and summarization.
Recent AI-powered hardware launches, like the Humane Pin and Rabbit R1, have faced criticism, yet there's still significant venture capital and interest in the sector, with prominent figures like Sam Altman eyeing substantial investments. The allure lies in AI's potential to revolutionize consumer hardware by utilizing sensors, silicon, and interfaces creatively. However, the challenge to offer a compelling alternative to versatile smartphones remains, with AI still needing to mature and hardware startups struggling to compete with established tech giants.
Penzai, a JAX library, enables easy manipulation and understanding of trained models through legible, functional Pytree structures. It includes a diverse set of tools for model visualization, debugging, and component analysis. Installation and usage are straightforward, with detailed tutorials available for learning to build and manipulate neural networks using Penzai.
This deep dive tutorial walks through how to build a system that reports on car traffic density. It uses modern computer vision to count vehicles over time.
A great new dataset from Google that contains detailed and comprehensive labels.
The U.S. Air Force is aggressively moving towards an AI-enabled fleet of over 1,000 unmanned warplanes, with operational capability expected by 2028.
AI-powered talent sourcing.
Sponsored
If you missed Netlify's Compose Web + AI virtual event, now's your chance to watch every session on-demand, including:
Apple CEO Tim Cook is anticipated to tease new AI features during the "Let Loose" event next week, with further details expected at the WWDC in June. The new iPad Pro models may receive the M4 chip, hinting at potential advanced AI capabilities in Apple's upcoming products.
Eight prominent U.S. newspapers owned by investment giant Alden Global Capital are suing OpenAI and Microsoft for copyright infringement in a complaint filed Tuesday in the Southern District of New York. On top of a similar case filed by the New York Times against both companies, the new suits add heft to publishers' claims. Until now, the Times was the only major newspaper to take legal action against AI firms for copyright infringement.
Mustafa Suleyman, co-founder of DeepMind, has been named the chief executive of Microsoft AI. He will contribute to Microsoft's expansion into AI consumer products, while his former colleague and DeepMind's other co-founder, Demis Hassabis, will lead AI research at Google. The journey of these two influential figures reflects the personal and competitive undercurrents driving the race to develop the next major computing platform.
This study discusses how combining algorithmic adjustments with tailored hardware can enhance ViTs' efficiency, particularly through model quantization.
Spikformer V2 combines the self-attention mechanism with the biological efficiency of Spiking Neural Networks (SNNs). This innovative model uses a Spiking Self-Attention mechanism and a Convolutional Stem, enhancing its ability to process visual features while being energy-efficient.
Full-Frequency Dynamic Convolution (FFDConv) is a new method that enhances 2D convolution for sound event detection. By generating unique frequency kernels for each band, FFDConv improves the accuracy of detecting sound events, especially in terms of their frequency characteristics.
Sponsored
Domino Data Lab is running a series of free in-person events [14] for data science and IT leaders to network, innovate, and share perspectives. Hear from industry leaders from EY, Moderna, Bridgewater, Moody's Analytics, & more — register now to shape the discussion on responsible AI practices that can accelerate impact.
The Segment Anything Model (SAM) from Meta AI, a notable foundation model in computer vision, excels at image segmentation but struggles in certain specific areas. This project presents ASAM, an advancement on SAM that uses adversarial tuning to boost its performance.
This project introduces SUNDAE, a new approach that enhances memory efficiency through spectral pruning and neural compensation.
InstructDr is a model designed to excel in various visual document understanding tasks like question answering and information extraction. InstructDr can adapt to new tasks and datasets by combining document images with large language models, outperforming existing models.
This study reveals a method to significantly extend the context length of language models up to 128K, emphasizing the importance of both the amount and variety of training data.
NYC's ascendancy in AI highlights the city's robust talent pool and growth as an AI hub. The NYC tech scene has drawn AI unicorns and tech workers. It is fuelled by resources like elite universities and a $400 million AI Research Consortium fund.
In recent years, the majority of AI applications have adopted traditional subscription-based pricing models, with a focus on per-user charges, reflecting their role as digital assistants rather than replacements for human workers. Innovative pricing strategies, such as outcome-based models, are emerging among newer AI companies, potentially enhancing customer adoption and revenue by charging only for successful results.
Sponsored
Learn how to build a virtual banking assistant with access to real time data using Amazon Bedrock and Confluent Cloud. Watch the tech
Microsoft updated Azure OpenAI Service's terms to prohibit U.S. police from using its generative AI for facial recognition, clarifying restrictions on law enforcement applications globally.
The $800 million investment aims to create over 100 jobs and is expected to be operational by the end of 2026.
StreamMultiDiffusion is a framework that enables real-time region-based text-to-image generation.
Coreweave is a GPU provider that rents out super clusters. This recent funding brings them to a $19B valuation.
Field AI, founded by former NASA JPL leader Ali Agha, specializes in autonomous robots that operate in unstructured environments without needing maps, GPS, or human supervision. Using advanced AI models developed from DARPA challenges, the company's robots can adapt to new and changing conditions, significantly enhancing their utility in industrial and construction settings.
AWS has launched Amazon Q, a generative AI assistant aimed at improving software development and decision-making by leveraging a company's internal data. Amazon Q facilitates coding, testing, and app development for developers, while offering data-driven support for business users through natural language interaction. The service also includes Amazon Q Apps, enabling the creation of custom AI applications without coding expertise.
Multi-Layer Perceptrons are used widely in AI today, including in the Transformer between Attention layers. However, they use fixed activation functions. This paper suggests using learned activation functions on edges using the Kolmogorov-Arnold representation (functions can be represented by a superposition of simpler functions). In this case, the researchers replace weights with splines. The architecture is much more complicated, but has some interesting properties that potentially make it more interpretable.
Reka trains large foundation models and has quickly caught up to some of the best players despite having a fraction of the funding. It has released a subset of its internal evaluation suite, which it uses to determine how strong its models are.
This survey paper delves into the world of Retrieval-Augmented Language Models (RALMs), highlighting their evolution, structure, and diverse applications in NLP tasks such as translation and dialogue systems.
Lightplane Renderer and Splatter components is a new method that dramatically reduces memory usage in 2D-3D mappings. The Lightplane Renderer skillfully generates images from neural 3D fields, while the Lightplane Splatter efficiently projects these images into 3D Hash structures.
The new Mamba model, trained using contrastive language-image pretraining (CLIP), shows impressive efficiency and performance in zero-shot image classification.
Researchers have developed an innovative 3D generation algorithm, dubbed MicroDreamer, which significantly accelerates the process by reducing the number of function evaluations required.
A mysterious AI model named gpt2-chatbot, displaying capabilities akin to GPT-4.5, has emerged on lmsys.org, prompting speculation of it being an unofficial OpenAI test for their next iteration. Key identifiers such as response quality, OpenAI-specific traits, and rate limits suggest a high level of sophistication, potentially hinting at a discreet benchmarking initiative by OpenAI. Investigations and discussions around gpt2-chatbot's origins and capabilities are ongoing in the AI community.
High-profile AI startups like Inflection AI, Stability AI, and Anthropic are facing financial pressures as they struggle with the high costs of developing generative AI models. While OpenAI, backed by Microsoft, has shown revenue growth, competitors like Anthropic and Stability AI grapple with substantial gaps between revenue and operating expenses. Microsoft's investment in AI hints at the tech industry's belief in AI's long-term profitability, despite the current challenges in monetizing these expensive technologies.
Sponsored
Visually assemble intelligent experiences on an intuitive canvas — and use built-in AI to make them richer and more personalized. Explore