Sponsored by

Hello, Starters!

Generative AI is evolving every day, with companies continually adding more features to models that leave us in awe. Can you even imagine what it will be like in a year?

Here’s what you’ll find today:

• Covariant introduces RFM-1

• Pika Labs' latest audio model

• Consistent characters with Midjourney

• Nvidia gets a copyright suit

• Cohere announces Command-R

• And more.

🤖 Covariant introduces RFM-1 (4 min)

The robotics industry is undergoing a significant transformation thanks to AI, and Covariant's latest innovation serves as evidence of this growing trend. The company has introduced RFM-1 (Robotics Foundation Model 1), a large language model designed for robot language, and trained with Covariant's Brain AI platform.

The primary concept behind RFM-1 is to empower robots with the necessary data for human-like reasoning, ultimately enhancing humanoid or general-purpose systems across various industries such as manufacturing, food processing, agriculture, and even homes.

🔊 Pika Labs' latest audio model (2 min)

Video generation continues to be one of the main fields for AI development in the last few months. Keeping up with the race, Pika Labs has released a new generative audio model that produces sound effects from text prompts.

Most AI videos tend to be silent, and the fact that companies are speeding up to enhance their models to perform both audio and video at the same time showcases the rapid pace of generative AI. Could we be seeing AI-generated movies soon?

👾 Consistent characters with Midjourney (4 min)

Suppose you're crafting a story using an AI image generation service. In that case, you might struggle with your prompts, as most generators are built on "diffusion models," making them inconsistent often by creating new answers each time you send a request. Midjourney is changing this with a new feature that allows users to maintain a specific character through every query.

The upgrade is still in the early stages, but you can easily add '-cref' (character reference) at the end of your prompt, and Midjourney will try to match the character's features in a different setting.

What you need for better GenAI applications

Learn more from Pinecone Research on how Retrieval Augmented Generation (RAG) using Pinecone serverless increases relevant answers from GPT-4 by 50%.

- The larger the search data, the more “faithful” the results.

- RAG with massive on-demand data outperforms GPT-4 (without RAG) by 50%, even with the data it was explicitly trained on.

- RAG, with a lot of data, ensures state of the art performance no matter the LLM you choose, encouraging the use of different LLMs (e.g., open-source or private LLMs).

✋ Nvidia is currently facing a copyright lawsuit from a group of novelists who claim that the company's NeMo platform was trained with content scraped from a shadow library that hosted several pirated books.

🤖With a focus on enterprise AI, Cohere has launched Command-R, a language model that excels in retrieval-augmented generation and tool usage. It is designed to help companies integrate AI to streamline workloads.

⚡️Quick links

AICI: Prompts as (Wasm) Programs

Midjourney bans all Stability AI employees over alleged data scraping

Note taking with Rabbit's r1 

OpenAI says there is no “agreement at all” with Elon Musk.

AI talent war heats up in Europe

Thank you for reading!