Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Now that Mickey is out of copyright, it is - in theory - ready to be used for generative training. This dataset is every frame of the original Steam Boat Willy animated short hosted on HuggingFace.
Google is restructuring its Ads business because of its new AI tools. The tools remove the need for a lot of work, increasing ad revenue profit margins. A growing number of advertisers are adopting Google's AI tools, eliminating the need for employees who specialize in selling ads for particular Google services. Details about the reorganization are expected to be announced next month.
Boost your site with smart AI feedback.
Rite Aid used facial recognition software that falsely identified shoplifters, leading to dramatic confrontations. The FTC is cracking down more on this technology.
Anthropic is reportedly seeking to raise $750M at a valuation of $15B in a round led by Menlo Ventures.
Apple has released two papers introducing new techniques for 3D avatars and efficient language model inference, which could allow AI systems to run on iPhones and iPads.
Gaussian Splatting has pushed 3D reconstruction forward. However, it often requires 200+ images of an object for accurate reconstruction. This work moves to a single monocular image and a learning-based technique that maps each pixel of the image into a 3d gaussian. The cool novel piece here is the introduction of the “Splatter Image,” which encodes 3D information for downstream rendering.
This paper explores the idea that models have hot neurons used for almost all inputs and cold neurons that are used much less often. By preloading the hot neurons to the GPU, you can save memory without sacrificing much throughput. A companion code library is available.
Researchers have developed a 'sequence-aware' loss function to improve Denoising Probabilistic Models (DPMs), addressing the issue of timestep correlation in image generation. This new approach not only offers a tighter estimation of loss but also shows significant improvements in image quality on datasets like CIFAR10 and CelebA, as evidenced by better FID and Inception Scores.
Sponsored
Tired of getting those calls? Take action before holiday spam drives you insane. Incogni [12] is a personal data removal service [12] that scrubs your sensitive info from the web (SSN, DOB, address, health information, contact details…)
A set of tools, resources, and examples for using the Llama series of models from Facebook.
The new CLIP-DINOiser method combines the zero-shot capabilities of the CLIP model with self-supervised features for improved semantic segmentation without annotations.
Llava is a strong multimodal model. This interactive design now allows chat style interleaving of text, image input, and image output.
AI has been in the spotlight throughout 2023. As we round out the year, what’s really happened? This article highlights the key trends that have defined AI in 2023.
Marketplaces have been a dominating business model over the last two decades. Here’s a breakdown from a16z sharing their thoughts on how AI will impact this company type. The central theme is about personalizing the experience for both sides of the marketplace.
UDiffText is a new method that enhances the quality of text in AI-generated images. By using a specialized text encoder and fine-tuning with a large dataset, UDiffText significantly reduces spelling errors and improves text accuracy.
Summarize hours of customer interviews in minutes.
Fast implementation of whisper that is dramatically faster than vanilla whisper for speech-to-text.
OpenAI has created a new safety advisory group and has granted the board veto power over all models.
Zoo Dev (previously Kitty Cad) has rebranded and released a new text-to-cad feature. It is a powerful platform that generates 3d assets that are usable for printing or parts.
Waymo's researchers say that its driverless cars have significantly reduced rates of police-reported and injury-causing crashes compared to human drivers. It is pushing broadly for international standardization for valid analysis of autonomous vehicle crash data.
This 850+ author paper outlines much of the Gemini language modeling effort from Google. It's understandably short on key architectural and dataset details, but gives attribution and evaluation details.
FontDiffuser is a diffusion-based method for automatic font generation, particularly excelling with complex characters and large style variations. It features a Multi-scale Content Aggregation block for better stroke preservation and a Style Contrastive Refinement module for style transfer.
The new 3D-LFM model uses transformers to reconstruct 3D structures from 2D landmarks without needing 'in-correspondence' 3D data. This approach, the first of its kind, efficiently handles varying point numbers, occlusions, and can generalize to unseen categories.
Advanced speech-to-text processing.
Easily train AlphaZero-like agents on any environment you want.
The MossFormer2 model, an advancement of the original MossFormer, offers improved monaural speech separation capabilities.
Diff-Text is a new training-free framework for creating photo-realistic images with text in any language. It enhances the Stable Diffusion model's multilingual capabilities by using sketched images as priors.
AI can play the game Labyrinth exponentially better than humans, with a best time of just 14.8 seconds. It can learn how to do so in just six hours. Labyrinth is an incredibly difficult physical board game that takes fine motor control, dexterity, and a lot of real-time problem-solving to win.
The LAION-5B dataset used by Stable Diffusion and other major AI products have been removed after a Stanford study found that it contained 3,226 suspected instances of child sexual abuse material, 1,008 of which were externally validated.
Meta's Chief Technology Officer Andrew Bosworth on the AI craze and where the technology will take the company.
The top-paying company for entry-level engineers is Jane Street with an L1 total compensation package of $350,000
Harvey AI, a legal service built on OpenAI technology, has raised another round of funding which values the young company at over $700m. The startup builds foundation models with OpenAI for law and legal practice applications.
A new company, whiterabbitneo, a reference to the Matrix, has released a 13B parameter language model trained for both offensive and defensive cyber security. It is trained to answer questions and provide information about computer security.
Anthropic updated ownership of output language in its usage terms. Also, it claims it will defend its customers in court over copyright claims against anthropic-generated output.
VideoPoet, a groundbreaking language model, is revolutionizing video creation with its unique ability to handle various tasks like text-to-video, video stylization, and even video-to-audio conversion. This approach stands out by combining multiple video generation techniques into one model.
This study presents a deep learning method for identifying deepfake faces in videos utilizing four pre-trained CNN models for high accuracy.
Introducing a multi-modal visual prompt tracking model, this project overcomes the limitations of single-modal object tracking by dynamically harnessing the strengths of various modalities like RGB and infrared.
Sponsored
For the next 12 hours, get 93% off Dollar Flight Club’s lifetime membership for $129 (normally $1,690) and try it risk-free (3-day money-back guarantee). Fly roundtrip to Paris from $299, Hawaii from $161, and more dream destinations discounted up to 90%. Start Exploring with $1500 off now [11].
LLMLingua uses a well-trained small language model after alignment to detect unimportant tokens in a prompt and enable inference with the compressed prompt in black-box LLMs, achieving up to 20x compression with minimal performance loss.
This project introduces Space-time Supersampling (STSS), a framework that significantly improves high-resolution, high-frame-rate content in real-time rendering.
Distil-Whisper is a distilled version of Whisper that is 6 times faster, 49% smaller, and performs within 1% WER on out-of-distribution evaluation sets.
Reliable Robotics Corp. has flown a small cargo plane without a human on board. The 12-minute flight, made with a plane on loan from FedEx, was Reliable Robotics' second automated flight. The startup is working to gain full approval from the FAA. Its system will restrict remote pilots to only supervising one aircraft at a time rather than managing multiple autonomous flights. Remote piloting will boost efficiency and allow planes to be repositioned more easily to match where demand is strongest.
This new model surpasses previous capabilities by simultaneously performing image segmentation, recognition, and captioning.
Microsoft Copilot can now compose songs thanks to an integration with GenAI music app Suno.