Meta Introduces AudioCraft - A New Framework for Generative AI Audio

🧰 Tools for AI app builders

Introducing AudioCraft: A New Framework for Generative AI Audio
Meta has released AudioCraft, a framework that generates high-quality audio and music from text inputs. It includes three models: MusicGen, AudioGen, and EnCodec. The models are available for research purposes and to further understanding of the technology.

MiniChain: Simplify Coding with LLMs
MiniChain is a compact library designed to streamline coding with large language models. It's built to handle the core functionality of prompt chaining without the complexity found in larger libraries.

MiniChain supports multiple backends, including OpenAI, Hugging Face, and even Python itself.
This is a great alternative to LangChain or LlamaIndex. Another simple LLM chat SDK is the aptly named - SimpleAiChat.

Unofficial Claude javascript SDK
Claude Unofficial API Makes Communication Simple - Updated August 5th, 2023 This lightweight JavaScript library allows software developers to easily interact with the Claude AI chatbot platform to start new conversations, send followup messages, upload files, and manage conversations. In under 40 lines of code you can be up and running.

LanceDB: Zero-server vector database for developers
LanceDB provides an easy way for you as a developer to save time by simplifying the databases typically needed for large vector datasets and metadata. Its ecosystem integrations with tools you likely use like Pandas, LangChain and others mean you can start using LanceDB in your projects today.

Another High-Performance Open-Source Vector Database
Epsilla is an open-source vector database offering scalability, high performance, and cost-effectiveness. It supports Python and REST API, and uses advanced parallel graph traversal techniques for vector indexing, providing 10x faster vector search than HNSW.

Building search for the post-ChatGPT world
Metaphor has developed a neural search engine that uses LLMs to improve internet search quality. The company has launched the Metaphor API, enabling developers to connect their LLMs to the internet for high-quality information retrieval.

Unify LLM interaction in this SDK wrapper
EasyLLM provides "clients" which have a similar interface to OpenAI's API. This allows you to switch between different models with just one line of code.

Typescript LLM Framework
Ax is a comprehensive TypeScript framework for creating robust AI applications.
The library consists of multiple modular packages that can be adopted incrementally. This provides a scalable end-to-end solution for AI development.

Prompt Engineering Guide for Large Language Models
This guide on GutHub provides comprehensive resources on prompt engineering for large language models (LLMs), including latest papers, techniques, and tools. It also announces a new course on the subject, taught by AI expert Elvis Saravia.

Unleashing Generative AI with Jupyter AI
Jupyter AI offers a powerful platform to explore generative AI models in notebooks like JupyterLab, Google Colab, and VSCode. With '%%ai magic' command, it enables interaction with providers like AI21, Anthropic, and OpenAI, and ensures compatibility with different JupyterLab versions.

Together AI | The cloud service for developers to build with open-source AI
Together AI offers a platform for fine-tuning and running large AI models, including chat, language, code, and image models. It supports hyperparameter control, weight downloads, and hosts open-source models like RedPajama and LLaMA-2.

Open source AI chat in your IDE
Continue is an open-source autopilot for software development. It’s a VS Code extension that brings ChatGPT to your IDE. Features include answering coding questions, editing in natural language, and generating files from scratch.

📰 Generative AI news and trends

Anthropic releases new version of Claude Instant
Claude Instant 1.2, an AI model for tasks like text analysis and summarization, has been released with improvements in coding, reasoning, and safety. It shows better performance in math and coding evaluations, and can be accessed via API.

Nvidia and Hugging Face Partner to Offer Cloud-Based AI Training
Nvidia and Hugging Face have partnered to offer cloud-based AI training services. The new service, called Training Cluster as a Service, will use Nvidia's DGX Cloud infrastructure to power the training of new and custom generative AI models. Hugging Face's platform of over 250,000 models and 50,000 datasets will also be integrated into the service, providing a helpful starting point for any AI project. The collaboration makes sense for both companies, as Nvidia is looking to expand its cloud services business and Hugging Face is looking to make its platform more accessible to businesses.

⌨️ Put down the keyboard

Embracing AI in Programming: A Shift, Not an End
[read: 5min]
From O’Reilly Radar: AI tools like GitHub's Copilot won't replace programmers but will change the discipline, making coding more efficient and freeing up time for understanding user needs, designing, testing, and debugging. This shift towards AI-assisted programming is an opportunity for developers to focus on problem-solving and customer collaboration.

Key Patterns for Building Systems with LLMs
[read: 45min]
This post summarizes some of the most important patterns for integrating LLMs into systems and products, as discussed in Eugene Yan's write-up on the topic. Yan distills learnings from academic research and industry experience into seven main patterns, organized from improving performance to reducing costs and risks, and from closer to the data to closer to the user experience.