. . AI: Mistral's Hackathon A Huge Success and OpenAI Shares Sora Feedback (3.25.24)

This is your daily short summaries of the top news stories in AI, according to the AI community.

Friends, today's newsletter highlights the buzz surrounding Mistral's Hackathon, showcasing the 7B model's versatility across a range of applications, from DOOM gameplay to RNA sequencing.

Additionally, we observe more AI leaders transitioning to new opportunities, alongside a helpful guide on setting up Retrieval-Augmented Generation (RAG) models and Large Language Models (LLMs). Enjoy!

-Marshall Kirkpatrick, Editor

Mistral's 7B Model Fine-Tuned to Play DOOM, Sequence RNA, and More at Hackathon

First impacted: Researchers, Developers
Time to impact: Short

Mistral threw a hackathon in San Francisco this weekend with a staggering 2000 entrants! The amazing traction led one attendee to comment "Mistral AI is what OpenAI would be if it were actually open". The Hackathon featured participants using Mistral to do everything from playing DOOM to analyzing RNA sequencing data. Mistral 7B was fine-tuned to play DOOM using ASCII frame representations (imagine letters and symbols replacing the typical textures). It was trained to interpret a 32x64 grid where each pixel contained the object on the frame buffer, allowing the model to "see" and interact with the game environment. In another application, the model was fine-tuned to parse RNA sequencing data, which enabled it to identify the functions of micro-RNA sequences and investigate their relations to specific conditions. Another standout was the launch of an open-source AI wearable device called FRIEND by a dedicated team. This device boasts an impressive ability to run for more than 24 hours on a single charge. It's designed to listen to your conversations, provide feedback, understand commands through voice or chat, and even summarize conversations, all for just $20. [via @AlexReibman] Explore more of our coverage of: AI Models, Gaming AI, RNA Analysis. Share this story by email

OpenAI Shares Sora's Positive Feedback from Artists and Filmmakers

First impacted: Artists, Creators, Designers
Time to impact: Medium

OpenAI has shared a blog post about early users' experiences with Sora. The feedback they shared was positive and highlighted Sora's ability to enhance their creative process and create visuals that are both realistic and surreal, while also navigating around budget constraints. The post includes 6 demo videos; we recommend checking them out, especially 'Air Head'. [Sora: First Impressions] Explore more of our coverage of: AI in Multimedia, Creative Process Enhancement, Budget Constraints Alleviation. Share this story by email

Emad Mostaque Resigns as Stability AI CEO to Focus on Decentralized AI

First impacted: Decentralized AI community, Stability AI employees
Time to impact: Short

Emad Mostaque has stepped down as CEO of Stability AI and from his board position to focus on decentralized AI, according to a recent company announcement. Shan Shan Wong and Christian Laforte, the company's COO and CTO respectively, have been appointed as interim co-CEOs, as the company initiates a search for a permanent CEO to lead its next growth phase. [Stability AI Announcement — Stability AI] Explore more of our coverage of: CEO Resignation, Stability AI, Decentralized AI. Share this story by email

Induced AI Launches Data-Extracting API

First impacted: Data Scientists, Web Developers
Time to impact: Short

Induced AI is testing a new API that uses a URL and a natural language query to extract structured data from websites. The API eliminates the need for custom scraping scripts and can extract specific data. It works by spinning up a browser agent in a remote session and combing through unstructured websites to find the prompted information. However, there are some limitations; for instance, it currently does not support pagination, though this feature is anticipated in the near future. [Quickstart - Induced Docs] Explore more of our coverage of: AI API, Data Extraction, Structured Data. Share this story by email

LlamaIndex Launches LlamaParse and Query Pipeline

First impacted: App Developers, Data Analysts
Time to impact: Short

Jerry Liu, co-founder and CEO of LlamaIndex, has shared a guide on leveraging the llama_index, MistralAI, and LlamaParse for creating sophisticated RAG and agents. The guide details how to utilize the Mistral API or local models via ollama, incorporating features such as function calling, custom query pipelines, and advanced document parsing, along with showcasing reference applications like Local RAG CLI and LaVague. [Build Cool Stuff with Mistral! (RAG, Agents)] Explore more of our coverage of: LLM Applications, Data Extraction Tools, API Development. Share this story by email

That’s it! More AI news tomorrow!