Search the archive

STABLE AUDIO OPEN (HUGGING FACE HUB)

Stability AI has released the weights to Stable Audio which was trained to generate sound effects on permissively licensed audio samples.

I'M BEARISH OPENAI

OpenAI's recent brain drain and focus on consumer products complicates its path to competing with big tech.

THE FUTURE OF FOUNDATION MODELS IS CLOSED-SOURCE

The future of AI foundation models will be closed-source due to the centralizing forces of data and compute, which favor economically viable and secure closed models over open-source alternatives.

Watch the Confluent webinar now

BRIGHTEDGE RELEASES POST GOOGLE I/O DATA ON THE IMPACT OF AI OVERVIEWS

New findings from BrightEdge Generative Parser reveal the primary industries impacted by AI Overviews, what triggers results, and where Google automatically anticipates and answers search queries.

CAPTIONS (PRODUCT)

Generate studio-grade videos — complete with custom lighting, varied camera angles, and different backgrounds — without the studio.

VIDEO GENERATION WITH EFFICIENT COMPRESSION

CV-VAE is a video VAE compatible with existing image and video models that offers spatio-temporally compressed latent spaces, enabling efficient training of new video models using pre-trained ones.

industries

A RIGHT TO WARN ABOUT ADVANCED AI

A group of current and former AI employees is calling for advanced AI companies to commit to principles that ensure transparency and protection for employees who raise risk-related concerns. They highlight the need for companies to avoid enforcing non-disparagement agreements, facilitate anonymous reporting processes, support open criticism, and prevent retaliation against whistleblowers.

WILL SCALING SOLVE ROBOTICS?

Over 900 people attended last year's Conference on Robot Learning, which featured 11 workshops and almost 200 accepted papers. One of the largest debates at the event was whether training a large neural network on a very large data set was a feasible way to solve robotics. This post presents the different sides of the argument to deepen people's understanding of the debate. Scaling has worked for other similar fields. However, it is impractical as there isn't much robotics data available and there's no clear way to get it. Even if scaling works as well as in other fields, it likely still won't solve robotics.

THE OPAQUE INVESTMENT EMPIRE MAKING OPENAI'S SAM ALTMAN RICH

Sam Altman is one of Silicon Valley's most prolific and aggressive individual investors. He manages an investment empire with holdings worth at least $2.8 billion as of early this year. Much of the portfolio isn't widely known. This article takes readers through what's known about Altman's investments.

MMLU PRO

MMLU is a common benchmark for reasoning tasks. It is often considered to be both the gold standard and something models have overfit. MMLU Pro is a new, harder, and cleaner benchmark for measuring language model reasoning.

TREE DIFFUSION: DIFFUSION MODELS FOR CODE

Fantastic diffusion paper that diffuses code for images. It can directly make edits as part of the diffusion process. It is slow, but can be combined easily with search to dramatically improve reasoning ability.

JAILBREAKING OF LARGE LANGUAGE MODELS

Researchers have introduced improved methods for optimization-based jailbreaking of large language models, building on the Greedy Coordinate Gradient (GCG) attack.

more relevant and accurate. READ THE BLOG NOW.

OMOST IMAGE SYNTHESIS

From the creator of ControlNet, Omost is a way to gain control over your image generation. It first rewrites prompts as a set of descriptive code. Then it uses that to render the final image. Importantly, you can edit the code before or after generation to slightly change the model output.

IMPROVED VIDEO SUPER RESOLUTION

Researchers have developed a training-free video interpolation method for generative video diffusion models. This new approach, compatible with various models, enhances frame rates without the need for extensive training or large datasets.

IMAGE COMPRESSION

Control-GIC is a new framework for generative image compression that allows fine-grained bitrate adjustment while maintaining high-quality results.

LLM INFERENCE SPEED OF LIGHT

Using theoretical speed of light modeling as grounding is really important for problems where the amount of computation and memory access is known a priori as it helps validate the quality of implementations and predict the impact of architectural changes.

PLENTIFUL, HIGH-PAYING JOBS IN THE AGE OF AI

It's possible that many of the jobs that humans do today will continue to be done by humans indefinitely, no matter how much better AIs are at those jobs, due to comparative advantage.

WHAT I LEARNED FROM LOOKING AT 900 MOST POPULAR OPEN SOURCE AI TOOLS

This review of open source AI repositories aims to give readers a big-picture view of the seemingly overwhelming AI ecosystem.

FACIA (PRODUCT)

Prevent fraud and spoofing attacks with advanced facial recognition technology.

EVEN THE RASPBERRY PI IS GETTING IN ON AI

Raspberry Pi is launching an AI chip integrated with its camera software that will allow AI applications like chatbots to run natively on the microcomputer.

USING AI TO DECODE DOC VOCALIZATIONS

University of Michigan researchers, in collaboration with Mexico's INAOE, have developed AI tools that analyze dog barks to determine playfulness or aggression, as well as identify the dog's breed, age, and sex.

AMD UNVEILS NEW AI CHIPS TO COMPETE WITH NVIDIA

AMD unveiled its latest AI processors, including the MI325X accelerator due in Q4 2024, at the Computex trade show. It also detailed plans to compete with Nvidia by releasing new AI chips annually. The MI350 series, expected in 2025, promises a 35-fold performance increase in inference compared to the MI300 series. The MI400 series is set for a 2026 release.

CLAUDE 3 OPUS AVAILABLE ON VERTEX AI

Google Cloud now offers Claude 3 Opus with tool use along with the smaller models as part of its Vertex AI offering.

MAMBA-2

Mambda is a powerful state space model. Its team has released a second version along with a long and detailed explanation of the model and the improvements.

FINEWEB: HIGHEST QUALITY WEB SCALE TEXT DATASET

Training a language model requires trillions of high quality tokens. Much of the information regarding the construction of these datasets is not published. This stellar blog post discusses different strategies for dataset cleaning and filtering. Its authors have released many of the best datasets out there for language model training.

LLMS IN HEALTHCARE

InvariantSelectPR is a method designed to improve the adaptability of Large Multimodal Models (LMMs) in domain-specific areas like healthcare.

IMPROVING LLM PERFORMANCE IN DATA-SCARCE DOMAINS

TrainAllInfAttn is a method that enhances the performance of large language models in specialized domains with limited data.

PORTRAIT ANIMATION WITH RAW DRIVING VIDEOS

MegActor is a new model that leverages raw driving videos for more realistic portrait animation. It achieves impressive results using a novel data generation framework and background encoding techniques by addressing identity leakage and background interference.

3D MESH GENERATION

MeshXL is a new model that generates high-quality 3D meshes.

FULLY LOCAL CODING ASSISTANT WITH OLLAMA AND CONTINUE

You can now run auto complete and chat fully on device with recent powerful improvements in coding with a number of open models.

AI WEATHER FORECASTING IS ALREADY MAKING HUGE STRIDES

The weather forecasting community is undergoing a revolution with AI models like WindBorne's WeatherMesh, which uses the rich ERA5 dataset to outperform traditional models while requiring significantly less computing power.

HANDWRITTEN MATH RECOGNITION

Implicit Character-Aided Learning (ICAL) improves handwritten mathematical expression recognition.

AI'S COMMUNICATION REVOLUTION

AI is leading to a revolution in communication spurred by OpenAI's GPT-4o, which integrates audio, vision, and text in real time. This shift enables more natural interactions with AI, transforming human-to-AI communication into a central mode of digital interaction and potentially leading to significant societal changes and new startups focused on AI-centric communication.

AMAZON'S PROJECT PI AI LOOKS FOR PRODUCT DEFECTS BEFORE THEY SHIP

Amazon's Project PI uses AI and computer vision to detect product defects or incorrect specifications before shipping, improving quality control and reducing returns.

BESTEVER (PRODUCT)

A powerful new creative AI suite to make ads.

ANOMALY DETECTION IN CHEST X-RAYS

This study explores the use of CLIP-based methods for anomaly detection in chest X-rays.

Invest in Jurny today ↗️

OPENAI IS RESTARTING ITS ROBOTICS RESEARCH GROUP

OpenAI is reviving its robotics research group after a three-year hiatus, aiming to develop multimodal robotics models and improve core AI models.

AI OVERVIEWS: ABOUT LAST WEEK

Google introduced AI Overviews to enhance search results, providing users with more accurate and relevant information, especially for complex queries. While some issues arose, such as erroneous results and misinterpreted content, Google has implemented over a dozen technical improvements to address these challenges, including better detection of nonsensical queries and limiting user-generated content in AI Overviews.

NVIDIA SAID TO BE PREPPING AI PC CHIP WITH ARM AND BLACKWELL CORES

Nvidia is reportedly preparing a system-on-chip that pairs Arm's Cortex-X5 core design with GPUs based on Nvidia's Blackwell architecture.

MUP FOR SPARSE MODELS

muP is a powerful tool used by all frontier model labs to transfer hyper parameters tuned on small models to larger, more expensive training runs. This paper explores how to do that for sparse models, thus dramatically improving training performance while reducing compute costs.