TLDR AI 2024-03-22

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.

PuzzleVQA is a dataset designed to test the abstract reasoning capabilities of large multimodal models like GPT-4V.

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.

Fed3DGS introduces a federated learning framework for scalable 3D reconstruction, utilizing 3D Gaussian splatting (3DGS) to enable decentralized scene building across vast areas.

This project introduces a framework that repurposes the CLIP model for medical anomaly detection by incorporating multi-level residual adapters and visual-language alignment loss functions.

Meta has released a native PyTorch library for fine-tuning language models.

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.

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.

Apple is in talks with Google to integrate Gemini generative AI technology into the iPhone and is considering using OpenAI's ChatGPT.