Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Sakana AI creates state-of-the-art Japanese language, vision, and image generation models. It introduced an evolutionary model merge that aims to evolve foundation models without expensive pretraining. The model merge has been released along with an explanation of the method.
Cohere’s newly launched RAG-optimized Command-R model, designed for businesses to get into large-scale production, is coming to the recently launched NVIDIA API catalog.
The U.S. Department of Commerce has proposed granting Intel up to $8.5 billion under the CHIPS and Science Act to support semiconductor projects in multiple states. This funding aims to bolster U.S. chip manufacturing, develop critical technologies like AI, and create thousands of jobs. Intel's planned $100 billion investment in expansion is one of the largest in U.S. semiconductor history.
Meta Reality Labs has trained a model that takes visual input and translates it into a 3D representation of a scene. The 70m parameter model runs quickly on-device and exhibits extreme stability.
Diffusion State Space Models (DiS) are a new type of diffusion model that use a state space backbone instead of the traditional U-Net for image data. These models can handle long-range dependencies and are efficient in generating high-quality images with less computational effort.
PuzzleVQA is a dataset designed to test the abstract reasoning capabilities of large multimodal models like GPT-4V.
Sponsored
Every day, Sourcegraph’s Cody [11] helps developers write >150,000 lines of code. Here’s how: AUTOCOMPLETE: Cody autocompletes single lines or whole functions in any programming language [11], configuration file, or documentation.
Meta has released a native PyTorch library for fine-tuning language models.
Fed3DGS introduces a federated learning framework for scalable 3D reconstruction, utilizing 3D Gaussian splatting (3DGS) to enable decentralized scene building across vast areas.
The GPT-4 fine-tuning API allows you to create a custom version of GPT-4 that can be 3x better at 1.5x the cost.
Anthropic's latest AI, Claude, achieves high scores on Mensa tests, signaling proximity to self-improving AI and posing both opportunities and existential risks. Machine learning is already displacing jobs, as evident at Klarna, where a customer service AI replaced 700 agents, hinting at an accelerating trend in automation. Duolingo's recent layoffs due to AI's translation capabilities further underscore this shift, emphasizing the growing impact of AI on the future of work.
Two years after publishing a skeptical AI article, Gary Marcus revisits his predictions and stands by his initial skepticism. Despite advancements like GPT-4, fundamental issues like genuine comprehension and trustworthy AI remain unresolved. Marcus concludes that scaling up data and computing power alone won't achieve AGI, and interdisciplinary collaboration is crucial.
Apple is in talks with Google to integrate Gemini generative AI technology into the iPhone and is considering using OpenAI's ChatGPT.
Copy.ai, an AI-powered writing tool, managed to acquire an impressive 10 million users within just four years.
The European Parliament has approved the first major AI regulation framework, setting global standards for artificial intelligence governance.
Sponsored
AI lab, Bland AI [4] has released a hyper-realistic sounding AI phone agent, and it’s blowing everyone's minds.
(P.S. TLDR readers can sign up here and access something even crazier... [5])
Stability AI's research scientist Robin Rombach, crucial to developing the Stable Diffusion model, is leaving the company, marking a significant departure amidst a year of technical team changes.
Waabi's Copilot4D is a pioneering foundation model that leverages LiDAR data to understand and predict the 3D dynamics of the environment over time, advancing the capabilities of autonomous machines.
NLX, an enterprise conversational AI platform, has raised additional funding from Cercano, Comcast, and others. The platform is used to build chat, voice, video, and conversational systems.
DreamDA offers a new approach to data augmentation, utilizing diffusion models to synthesize diverse, high-quality images that closely match the original data distribution.
Chain-of-Spot (CoS) introduces an Interactive Reasoning technique that significantly enhances how Large Vision-Language Models (LVLMs) process and understand images. By focusing on key regions of interest within images in response to specific questions or instructions, CoS enables LVLMs to access detailed visual information without compromising image resolution.
StableVITON is a novel approach to image-based virtual try-on. This method focuses on maintaining clothing details while leveraging the generative power of pre-trained diffusion models. StableVITON learns semantic correspondences between clothes and the human body in a pre-trained model's latent space.
Sponsored
Phospho is an open source text analytics platform [13] for LLM apps. Easily log user inputs and LLM app outputs to get an overview of what is happening in real time. Automatically detect and extract semantic events in each messages. Build automated workflows via webhooks or API calls when specific events are detected. Self deploy
Triton is a library for writing CUDA kernels in a Python-like way. It is gaining popularity. This repository has a set of puzzles in increasing difficulty that encourage learning the tool.
FRESCO is a novel approach that combines intra-frame and inter-frame correspondences to significantly improve the spatial-temporal consistency in video translation tasks.
This project enhances the capabilities of diffusion models for tasks like image editing and restoration by introducing Generalized Consistency Trajectory Models (GCTMs). These models streamline the process, making it possible to modify images with remarkable precision and efficiency by translating between any two distributions with just one step.
Tsinghua University and UC Berkeley researchers have achieved a significant breakthrough in matrix multiplication, presenting an innovative technique that has already spurred further enhancements. This advancement in a core computational operation could lead to substantial time, power, and cost savings across various applications. This represents the most considerable progress in reducing the computational complexity of matrix multiplication since the previous milestone in 2010.
This is a tutorial on creating a pixel-art-like image processing tool with interactive web elements like sliders for customization using a binning algorithm in JavaScript. The binning technique uses parameters such as bin size and gap to convert images into stylized, pixelated artwork by averaging pixel brightness within bins. The implementation involves manipulating pixel data on HTML canvas elements and optimizing looping structures for efficiency.
LLMs like Claude can generate effective fuzzers for parsing code, automating a process that traditionally requires significant human effort. While LLMs are typically not precise enough for static analysis, they appear well-suited for creating fuzzers due to the stochastic nature of fuzzing. A hybrid approach that combines LLM-driven static analysis and targeted fuzzing could be promising for identifying and exploiting vulnerabilities in code.
Sponsored
This webinar by Kolena and Rad AI will introduce cutting-edge testing methodologies that reduce time-to-market for AI/ML, while ensuring
JARS allows anyone to create episodes of their favorite shows with friends.
OpenAI could release GPT-5 in the summer.
Beijing Internet Court ruled that an AI-created image was copyright-protected artwork due to the human creator's intellectual input, awarding damages in a landmark case with significant implications for the AI industry and copyright disputes worldwide.
Mustafa Suleyman and Karén Simonyan are leaving Inflection to start Microsoft AI. Sean White is joining as the new CEO. A few senior team members from Inflection are also joining Microsoft AI.
Lilac joins Databricks to continue building their system for unsupervised language dataset construction.
A collaboration between IBM and NASA has resulted in the development of advanced transformer-based language models trained on scientific literature for improved performance in natural language understanding tasks within the scientific domain. These models are open-sourced on Hugging Face.
An article from a former OpenAI scientist about Sora and the potential costs of generating videos.
Retrieval Augmented Generation (RAG) is a way to inject knowledge into a potentially stale language model. Unfortunately, the current paradigm of “frozen RAG”, where only part of the pipeline is trained and the actual model itself isn’t updated, performs poorly outside of demos. This blog outlines the next generation of RAG where everything is tuned specifically for the task. An open model like Mistral 7B can outperform traditional GPT-4 RAG in this system.
Stable Diffusion 3 is a powerful image generation model. This paper introduces Latent Adversarial Diffusion Distillation, which reduces the number of diffusion steps down to 4 while maintaining image generation quality.
Sponsored
OctoAI [10] is the GenAI inference API that makes the most popular models faster, cheaper, and more reliable at scale. You can switch to OctoAI [10] in three lines of code and future-proof your app to leverage the newest LLMs that keep improving.
Quiet-Star is a reasoning framework that improves language models' abilities to generate correct output. Its code has been released along with an 8 step per token model.
Poster Dataset Distillation is a new method that condenses entire datasets into a single 'poster', enabling AI models to learn from fewer than one image per class.
Researchers have developed a new framework to help vision-language models learn continuously without forgetting previous knowledge using a system that expands the model with special adapters for new tasks.
MineDreamer is an AI agent that excels at following complex instructions in the Minecraft world thanks to its innovative use of advanced language and vision models.