LTXVideo 13B AI video generation

LTXV-13B is a groundbreaking 13 billion-parameter AI model developed by Lightricks, capable of generating high-quality videos 30 times faster than comparable models, thanks to its advanced multiscale rendering technology. The model supports various video generation modes, including text-to-video and image-to-video transformations, and can run efficiently on consumer hardware, making it a significant upgrade in the field of video creation.

Read more

AI is draining water from areas that need it most

The expansion of data centers to support artificial intelligence is straining water resources, with about two-thirds of new data centers built or in development since 2022 located in areas already experiencing high levels of water stress. The problem is worsening, with over 160 new AI data centers springing up in the US in the past three years in water-scarce areas, and the trend is also unfolding in other countries, threatening local water supplies, agriculture, and energy production.

Read more

The AI jobs crisis is here, now

Duolingo's CEO, Luis von Ahn, announced that the company is shifting to an "AI-first" approach, which has already led to the replacement of up to 100 workers, including writers and translators, with AI systems. This is part of a larger trend of companies using AI to displace human workers, with many professionals, including artists, illustrators, and voice actors, losing jobs or gigs to AI-generated content, and recent college graduates facing unusually high unemployment rates as firms hire fewer humans and use more AI.

Read more

SoundCloud faces backlash after adding an AI training clause in its user terms

SoundCloud is facing backlash from creators who discovered that the platform's terms of use allow it to use uploaded music to train its AI systems, prompting some artists to delete their accounts. However, SoundCloud claims it has never used artist content to train AI models without consent and has implemented technical safeguards to prohibit unauthorized use, with the updated terms intended to clarify how content interacts with AI technologies within its platform.

Read more

Simon Willison's first blog on LLMs (2022)

The author experimented with OpenAI's GPT-3 language model, prompting it to create a step-by-step tutorial for getting started with Datasette, a tool for creating and publishing data-driven websites. The generated tutorial was surprisingly accurate, with only a few minor errors, demonstrating GPT-3's impressive ability to accumulate and synthesize knowledge, although fact-checking is still necessary to ensure the accuracy of the generated content.

Read more

Scoring the European Citizen in the AI Era

The AI Act bans social scoring, a practice inspired by China's Social Credit System, but this ban is unlikely to affect existing scoring practices in Europe, such as creditworthiness assessments, as they are considered "high-risk AI systems" subject to regulations. However, the ban is argued to be a flexible and useful tool to protect individuals against disproportionate use of AI-based scoring, similar to Article 22 of the General Data Protection Regulation.

ArXiv

RAGDoll: Efficient Offloading-Based Online RAG System on a Single GPU

RAGDoll is a resource-efficient system that enhances large language model generation by incorporating external knowledge, designed for deployment on consumer-grade platforms with limited memory. By decoupling retrieval and generation into parallel pipelines, RAGDoll achieves up to 3.6 times speedup in average latency compared to traditional serial RAG systems.

ArXiv

Advancing Conversational Diagnostic AI with Multimodal Reasoning

The Articulate Medical Intelligence Explorer (AMIE) has been advanced to interpret and reason over multimodal medical data, such as images and documents, during diagnostic conversations, allowing for a more structured and effective history-taking process. In a study comparing AMIE to primary care physicians, AMIE was found to be superior in several areas, including diagnostic accuracy, demonstrating clear progress in multimodal conversational diagnostic AI.

ArXiv

Absolute Zero: Reinforced Self-Play Reasoning with Zero Data

The Absolute Zero Reasoner (AZR) is a new reinforcement learning paradigm that enables a single model to propose and solve its own tasks, improving its reasoning capabilities without relying on external data or human supervision. AZR achieves state-of-the-art performance on coding and mathematical reasoning tasks, outperforming existing models that rely on large amounts of human-curated examples, and can be applied to different model scales and classes.

ArXiv

Learning Adaptive Parallel Reasoning with Language Models

Adaptive Parallel Reasoning (APR) is a novel framework that enables language models to optimize their reasoning processes by adaptively allocating computation through both serialized and parallel methods. APR has demonstrated significant benefits, including higher performance, superior scalability, and improved accuracy, outperforming existing methods in experiments on the Countdown reasoning task.

ArXiv

EM-LLM: Human-Inspired Episodic Memory for Infinite Context LLMs

EM-LLM is a novel architecture that integrates human-inspired episodic memory into large language models, enabling them to handle infinite context lengths while maintaining computational efficiency. The model organizes sequences of tokens into coherent episodic events and retrieves them through a two-stage memory process, outperforming state-of-the-art models like InfLLM and RAG in various tasks while requiring similar resources.

Repo

Show HN: MCP Boilerplate – A rock-solid Model Context Protocol server

The Model Context Protocol (MCP) server boilerplate provides a standardized way to connect AI models to different data sources and tools using the MCP. It features a robust structure for building custom MCP servers, secure authentication, logging capabilities, and session management, along with example tools and comprehensive error handling.

Repo

Scraipe: AI Scraping and Analysis Framework

Scraipe is a Python framework for scraping and analyzing data, featuring versatile scraping, LLM analysis, and workflow management, with a modular design and customizable ingress rules. It can be installed using pip and provides a simple API for scraping and analyzing data, with examples and documentation available for users to get started.

Repo

AI System Prompt Leaks

The author is working on proofs and invites others to share their ideas and methods, providing contact information through email, Discord, and Reddit. They can be reached at asgeirtj@gmail.com, Discord username asgeirtj, or on Reddit as StableSable.

Repo

Lego GPT

LegoGPT is a project that generates physically stable LEGO brick models from text prompts, and its official repository provides instructions for installation, running inference, and fine-tuning the model. The project uses a fine-tuned LLaMA model and requires additional dependencies such as Gurobi and ImportLDraw to run stability analysis and render LEGO visualizations.

Repo