DeepMind’s AI-First Science Quest Continues with AlphaFold 3

The new model can predict the structure of many life’s molecules such as proteins, DNA, RNA and several others.

Created Using Ideogram

Next Week in The Sequence:

  • Edge 395: We dive into task-decomposition for autonomous agents. Review Google’s ReAct( Reason + Action) paper and the Bazed framework for building agents in TypeScript.

  • Edge 396: With all the noise about Apple’s AI strategy, we dive into some of their recent research in Ferret-UI.

You can subscribed to The Sequence below:

TheSequence is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

📝 Editorial: DeepMind’s AI-First Science Quest Continues with AlphaFold 3

AI for science is one of my favorite forms of AI 😊. You can argue that the real test for AGI is when it’s able to create new science. No company in the world has been pushing the boundaries of AI for science like Google DeepMind. Among the scientific achievements of DeepMind models, none has achieved more notoriety than AlphaFold, the model that was able to predict protein structures from a sequence of amino acids. A few days ago, DeepMind published details about AlphaFold 3, which expands its prediction capabilities beyond just proteins to a broad spectrum of biomolecules.

Our understanding of life’s molecules is a core foundation of our understanding of biological life and certainly the cornerstone of drug discovery. AlphaFold 3 is able to predict the structure of large molecular structures such as proteins, RNA, DNA, or even small molecules such as ligands. Even more impressive is the fact that AlphaFold 3 can model the chemical interactions in those molecules which effectively control cell functioning. Starting with a list of molecules, AlphaFold 3 is able to generate a 3D structure that clearly visualizes its joint 3D structure, revealing its intricacies and interactions.

Science is nothing if not in the hands of scientists. One of the coolest things about AlphaFold 3 was the release of the AlphaFold Server, a free search tool that scientists can use to test new hypotheses and model biomolecular structures. AlphaFold 3 continues Google DeepMind’s incredible scientific achievements in areas such as mathematics, physics, biology, and several others. The AlphaFold 3 paper is absolutely fascinating and a recommended read.

🔥 How to Stop Hallucinations

Introducing Galileo Protect – a revolutionary GenAI firewall that intercepts hallucinations, prompt attacks, security threats, and more!

See Protect live in action, and learn how it works, what makes it the first-of-its-kind, and ways to stop hallucinations in real-time.

🔎 ML Research

AlphaFold 3

Google DeepMind published a paper detailing AlphaFold 3, the new version of its groundbreaking bioscience model that was able to predict protein structures. The new version builds on its predecessor to predict the structure of other molecules such as DNA, RNA, ligands and more —> Read more.

Prometheus 2

Researchers from several labs such as Carnegie Mellon University and Allen AI published a paper proposing Prometheus 2, an LLM specialized in evaluating other LLMs. Prometheus builds on its predecessor and achieve results that closely mirror GPT-4 and human judgements —> Read more.

LoftQ

Microsoft Research published a paper proposing LoftQ, a technique that streamlines the fine-tuning process in LLMs. LoftQ combines ideas from methods such as LoRA or QLora with techniques like quantization and adaptive initialization to build a highly optimal fine-tuning process —> Read more.

LoRA LAnd

AI researchers from Predibase published a paper outlining LoRA Land, a group of 310 fine-tuned LLMs that rival GPT-4 across different domains. LoRA Land provides adapters a combination of 10 based models fine-tuned for 31 different tasks —> Read more.

NeMo Aligner

NVIDIA Research published a paper introducing NeMo-Aligner, a toolkit for model alignment that scales to hundreds of GPUs for training. NeMo-Aligner combines techniques such as Reinforcement Learning from Human Feedback (RLHF), Direct Preference Optimization (DPO), SteerLM, and Self-Play Fine-Tuning (SPIN) in a single toolkit —> Read more.

Google and DeepMind at ICLR

Google DeepMind researchers published a comprehensive summary of some of the papers submitted to the ICLR 2024 conference. The papers cover categories such as agents, frontier computer vision and foundational learning —> Read more.

🤖 Cool AI Tech Releases

Bedrock Studio

Amazon released Bedrock Studio, a web interface for developers to collaborate in the implementation of generative AI applications —> Read more.

Model Spec

OpenAI published Model Spec, a set of guidelines for shaping the desired behavior in foundation models —> Read more.

🛠 Real World AI

PayPal Cosmos.AI

Paypal shared some details about Cosmos.AI, its internal AI/ML infrastructure platform —> Read more.

Inside Einstein for Developers

Salesforce’s VP of Software Engineering shared some details about the development of the Einstein for Developers platform —> Read more.

📡AI Radar

TheSequence is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

By Julio MarchiPosted on