🧬 AlphaFold 3 Enhances Drug Research

PLUS: IBM Releases Code Models

Welcome back!

DeepMind’s new AlphaFold 3 sets the highest benchmarks in molecular modeling. This leap forward in AI’s ability to predict the intricate interactions between diverse biomolecules holds transformative potential for drug discovery. AlphaFold 3 is more than a tool, it’s a gateway to the future of biological research. Let’s unpack this...

Today’s Summary:

  • DeepMind launches AlphaFold 3

  • IBM releases Granite Code Models

  • Open-source AI challenges big players

  • Microsoft debuts AI for U.S. intel

  • AI deepfakes storm social media

  • 3 new tools


AlphaFold 3: The Next Leap in AI-Powered Molecular Modeling

The Summary: DeepMind's AlphaFold 3 achieves a new breakthrough in accurately predicting the structure of diverse biomolecular complexes, ranging from proteins to DNA and small molecule drugs.

While previous versions could already predict protein structures with high accuracy, AlphaFold 3 takes a major leap forward by accurately modeling interactions between different types of biomolecules, crucial for drug discovery and biological research.

Key details:

  • The model is at least 50% more accurate than existing methods in predicting interactions between proteins-ligands, proteins-nucleic acids, antibody-antigen binding, and other molecular complexes.

  • AlphaFold 3's accuracy on the PoseBusters benchmark surpasses physics-based tools for the first time.

  • DeepMind provides free access through the AlphaFold Server, while Isomorphic Labs collaborates with pharma companies to accelerate drug design using this technology.

Why it matters: AlphaFold 3 represents a major milestone in using AI to decipher the intricate interactions between life's molecules. Its increased accuracy across biomolecular complexes could allow new breakthroughs in drug discovery, enzyme design, genetic disease research, and elucidating biological mechanisms. With free tools and commercial collaborations, it brings transformative potential for therapeutic development and advancing our understanding of the biological world.

“With AlphaFold Server, it’s not only about predicting structures anymore, it’s about generously giving access: allowing researchers to ask daring questions and accelerate discoveries.

Céline Bouchoux, The Francis Crick Institute

IBM Launches Granite Code Models

The Summary: IBM has open-sourced its Granite family of code models, ranging from 3 to 34 billion parameters, under the Apache 2.0 license. These models are designed to make coding more accessible and productive for developers.

They have been optimized for enterprise software development workflows and perform well across various coding tasks. IBM believes these models can enable a future where writing code is as easy as conversing with an assistant.

Image: IBM

Key details:

  • Trained on 3-4 trillion tokens from 116 programming languages and 500 billion tokens from a mixture of code and natural language data.

  • The models match state-of-the-art performance among open-source code LLMs on tasks like code generation, fixing, and explanation.

  • Granite-8B-Code outperforms CodeGemma-7B, Mistral-7B, LLama-3-8B models of similar sizes in various coding tasks.

Why it matters: Open-sourcing the Granite code models is important to democratize AI-powered coding tools and make them accessible to a wider developer community. These models can automate tedious tasks and modernize app development. IBM's commitment to open innovation and transparency around model training is a good signal aligning with enterprise needs for trustworthy AI solutions.


Open Source AI Rapidly Approaching Proprietary Models

The Summary: A fascinating interactive chart on Huggingface offers a real-time visualization of the narrowing gap between open-source and proprietary AI models.

The data suggests that while proprietary models currently maintain a lead in capabilities, the progress of open-source AI appears to be faster.

Image: Huggingface

Key details:

  • The chart visualizes the evolving “smartness” of open-source vs proprietary AI models based on Elo ratings.

  • Data suggests that proprietary models' lead may be temporary, as open-source AI progresses faster.

  • The open-source LLaMa3 model scores higher than larger models like GPT-3.5 and Claude 2.

  • The trend promises further democratization of advanced AI to individuals and businesses.

Why it matters: The rapid progress of open-source AI models represents a significant shift in the AI landscape. As these models become increasingly capable, they could potentially level the playing field, making advanced AI technologies accessible to a broader range of users.


Quick news


🥇 New tools

That’s all for today!

If you liked this newsletter, share it with your friends and colleagues by sending them your invite link : {{rp_refer_url}}.