From Molecules to Machines

Revolutionizing Science: ESM3 Decodes Proteins, Qualcomm Empowers Developers, and ALTER-LLM Brings GPT-4 to Robots

June 29, 2024

A few months back, Google’s DeepMind introduced AlphaFold 3, a groundbreaking advancement in AI-driven biological research. We covered it in this issue.

This was a huge breakthrough in detecting and manipulating protein structures. To a point where Medical research will be independent of human very soon.

A few days back, EvolutionaryScale's announced their newly developed ESM3 model which takes protein analysis to the next level. ESM3, trained on 2.78 billion protein sequences, can reason over protein sequence, structure, and function simultaneously, simulating 500 million years of evolution. These powerful tools are set to accelerate scientific research in biology, medicine, drug development, and materials science, offering unprecedented insights into the building blocks of life and paving the way for transformative discoveries.

Thanks to Google, “real” use cases of AI have finally started to emerge between all the deepfakes and Pasta eating videos.

For more, let's dive into these developments.

Before you dive in

2 New Skills have emerged over the last few months: AI Agents, Prompt Engineering.

To Learn these skills you need to learn 5 core concepts.

AI, ML, AGI, Agentic workflows, GenAI.

How do you learn these concepts step by step?

These 5 Courses at 90% off is all you need.

See you on the other side.

Stuff you should know

ESM3: A Protein Genius Model

EvolutionaryScale, a leading AI research lab, has developed ESM3, a generative AI model that can simultaneously reason over protein sequence, structure, and function. Trained on 2.78 billion protein sequences, ESM3 can simulate 500 million years of protein evolution and create novel variants. ESM3's capabilities can accelerate research and discovery in fields like drug development, materials science, and carbon capture. Scientists can fine-tune the model using their own data to enhance its performance. To make ESM3 widely accessible, EvolutionaryScale is partnering with AWS and NVIDIA to offer the model through their platforms. An open version is also being released for non-commercial use, empowering the scientific community to leverage this groundbreaking technology.

A Talking Robot

The ALTER-LLM project integrates GPT-4, a powerful language model, with Alter3, a humanoid robot, enabling it to generate spontaneous, human-like motions based on textual prompts. This innovative approach allows Alter3 to perform a wide range of actions without additional training, express emotions, and respond contextually to conversations. The system demonstrates GPT-4's comprehensive understanding of human movements and its potential in robotics. By bridging linguistic understanding with physical action, the project represents a significant advancement in humanoid robotics, achieving substantial autonomy and empathetic responses. The website showcases video recordings of Alter3's capabilities, highlighting the successful fusion of language models and robotic embodiment.

Qualcomm’s AI Vault

Qualcomm has made its AI models available to app developers through its AI Hub, following the introduction of its Snapdragon X Elite platforms at Computex. These models are optimized for Qualcomm's 45 TOPS Hexagon neural processing unit (NPU) and are designed to enhance AI capabilities in applications such as image classification, object detection, semantic segmentation, and generative AI. The AI Hub provides tools, documentation, and examples to facilitate the deployment of these models on edge devices powered by Snapdragon X Elite processors. This move aims to attract developers and catalyze demand for AI-enabled applications on Qualcomm's platforms, giving it a competitive edge against rivals like AMD and Intel.

Hand Picked Video

Anthropic has recently released a significant update to Claude, introducing feature called "Projects" that aims to enhance collaboration and productivity for users. A new feature that can potentially save hours!

Top AI Products from this week 

  • Olly.social - Huge update announced in v1.5 AI Update. Check it out!

  • Pygma - First - I'll analyze your Instagram profile data and save the vibe. Second - give you 7 days of content ideas that could be interesting to your audience.

  • Devv AI - Devv enhances developer productivity by combining LLMs with real-time data from Stack Overflow, GitHub, and DevDocs for accurate, up-to-date answers.

  • Featherless AI - Featherless is a platform to use the very latest open source AI models from Hugging Face. With hundreds of new models daily, you need dedicated tools to keep with the hype.

  • Zebracat - Zebracat is a text-to-video platform designed for marketers and entrepreneurs to create impactful videos quickly and easily.

  • Relay.app - Relay.app is the easiest-to-use Zapier alternative, now with 3 new flexible ways to incorporate AI in your automations.

  • Scene - Scene is a no-code website builder with AI assistance at every step—from refining briefs to generating wireframes and writing sample copy.

  • Lazy AI - Lazy AI turns the traditional software development process on its head. Now you can create and deploy full stack web applications with prompts!

  • Superhuman 2.0 - 1 billion professionals use email for 3 hours every day — more than any other tool! But unlike every other tool, email is not yet collaborative.

  • ElevenLabs Reader - The ElevenLabs Reader App narrates articles, PDFs, ePubs, newsletters, or any other text content. Simply choose a voice from our expansive library, upload your content, and listen on the go.

This week in AI

  • OpenAI Delays ChatGPT's Advanced Voice Mode - OpenAI has postponed the release of ChatGPT's advanced Voice Mode from June to July for initial testing, with a wider rollout potentially in the fall. The delay is due to ongoing improvements in content detection, user experience, and infrastructure scaling. This feature aims to enable real-time, emotion-infused AI conversations. The delay doesn't affect upcoming video and screen-sharing capabilities.

  • Reddit Implements Measures to Block Unauthorized AI Crawlers - Reddit is introducing new measures to protect its platform from unauthorized AI crawlers and data scraping. These changes include updating the Robots Exclusion Protocol, restricting unidentified bots, and enforcing compliance with their content policy. The move aims to prevent AI companies from using Reddit content without permission, while maintaining partnerships with authorized entities.

  • Amazon's AI Chatbot "Metis" to Rival ChatGPT - Amazon is developing a consumer-focused AI chatbot codenamed "Metis" to compete with ChatGPT. The web-based chatbot, powered by a new foundational model, could launch later this year. It's part of Amazon's broader AI strategy, which includes enterprise tool Amazon Q and shopping assistant Rufus. CEO Andy Jassy is reportedly directly involved, signaling Amazon's commitment to advancing in the generative AI space alongside other tech giants.

  • IMS-Toucan v3.0: Multilingual TTS Breakthrough - IMS-Toucan v3.0 is a groundbreaking text-to-speech model capable of synthesizing speech in over 7000 languages. This release features improved TTS quality, watermarking, and support for almost all ISO-639-3 languages. Developed collaboratively by academic institutions, it extrapolates from 462 languages to cover thousands more. This significant advancement in multilingual TTS technology offers unprecedented language coverage and quality, with a simplified user experience.