TLDR AI 2024-06-04

AMD AI Chips 💾, Claude 3 Opus & Vertex AI 🤝, Mamba-2 🐍

Headlines & Launches

AMD Unveils New AI Chips To Compete With Nvidia (2 minute read)

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 (3 minute read)

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

Mamba-2 (23 minute read)

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.
Research & Innovation

Improving LLM Performance in Data-Scarce Domains (16 minute read)

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

LLMs in Healthcare (18 minute read)

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

FineWeb: Highest Quality Web Scale Text Dataset (18 minute read)

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.
Engineering & Resources

Portrait Animation with Raw Driving Videos (3 minute read)

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 (17 minute read)

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

Fully local coding assistant with Ollama and Continue (6 minute read)

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

Handwritten Math Recognition (24 minute read)

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

AI Weather Forecasting Is Already Making Huge Strides (4 minute read)

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.

AI's Communication Revolution (14 minute read)

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.
Quick Links

Bestever (Product)

A powerful new creative AI suite to make ads.

Anomaly Detection in Chest X-Rays (GitHub Repo)

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

Amazon's Project Pi AI Looks For Product Defects Before They Ship (1 minute read)

Amazon's Project PI uses AI and computer vision to detect product defects or incorrect specifications before shipping, improving quality control and reducing returns.
The most important AI, ML, and data science news in a free daily email.
Join 500,000 readers for