• Starter AI
  • Posts
  • Modular’s MAX platform, Groq launches GroqCloud, AI in robotics

Modular’s MAX platform, Groq launches GroqCloud, AI in robotics

Groq keeps turning heads in the industry.

Hello, Starters!

Developers are a key aspect of AI, so companies are always trying to create groundbreaking platforms for them to deploy many more advancements. Are you one of those devs?

Here’s what you’ll find today:

  • Modular introduces MAX

  • Groq & Definitive Intelligence are set to launch GroqCloud

  • AI is a walking coach for robots

  • Figure is making waves in the tech world

  • India lays new AI rules

  • And more.


How Large Language Model Applications are Built in 2024

One of the best tactics for building applications with LLMs is to use a vector database and RAG - Retrieval Augmented Generation.

This is where you take the user’s input prompt and then search through a vector database for similar text. Then, you take the matching passages (from the vector database) and the user’s prompt and enter that into your LLM (GPT-4, LLaMA, whatever) to generate text output.

Using Retrieval Augmented Generation has huge benefits. It can improve GPT-4 answers by 50%, even on data GPT-4 was trained on. Additionally, companies like Notion also use Retrieval Augmented Generation to personalize answers, so it can do things like question-answering on all your past notes.

Ram Sriharsha is the CTO of Pinecone (one of the most popular vector databases) and he wrote a great blog post about

  • How vector databases work

  • Why companies like Notion use Pinecone

  • How Pinecone built their serverless architecture

And more.

🖥️ Modular introduces MAX (12 min)

Modular has announced MAX Developer Edition. This toolkit aims to enhance developers' workloads and improve the deployment of AI applications for efficient and fast performance. It's backed by three main parts: MAX Engine, MAX Serving, and Mojo — Modular's programming language specifically designed for AI.

In addition to powering developers with the right capabilities to build high-performance AI applications, Modular's integration of Mojo as the foundation for MAX intends to provide a portable alternative to Python, C, and CUDA, unifying the entire AI stack.

Groq has recently acquired Definitive Intelligence intending to collaborate to create GroqCloud, a “developer playground” that includes integrated documentation, code samples, and self-serve access, providing developers with access to Groq's groundbreaking LPUs.

The LPUs, or Language Processing Units, have made Groq the talk of the town, with many developers asserting that their rapid processing capacities make the company a potential competitor to Nvidia.

A research paper titled "Humanoid Locomotion as Next Token Prediction," introduced by the University of California, Berkeley presents an innovative approach to teaching robots to walk like humans. By leveraging Transformer technology from LLMs, the robot predicts its next step based on the ones it has already taken.

In simpler terms, just like predictive text on our phones, where each step is like a word in a sentence, these "tokens" allow the robot to continue the sequence, enabling it to take a casual stroll through the park. This remarkable achievement was made with only 27 hours of walking data for training.

💸Figure has caught the attention of major tech companies like Microsoft and Nvidia. Securing $675 million in funding, the humanoid robotics company now has a value of $2.6 billion. Moreover, by partnering with OpenAI, they’re planning to develop AI models for humanoid robots.

India has recently made it a requirement for tech companies to obtain approval before releasing “unreliable” or trial-stage AI tools. This directive emphasises the need for labels highlighting potential inaccuracies, and without the government's permission, these tools cannot launch on the Indian internet.

What did you think of today's newsletter?

Login or Subscribe to participate in polls.

Thank you for reading!