In partnership with

Hello, Starters!

There are many uses for AI, from something as "simple" as creating funny images of cats in space to discovering new drugs and advancing medical research. Quite multifaceted, isn't it?

Here’s what you’ll find today:

• Google DeepMind introduces AlphaFold 3

• “Agent Hospital”: A simulation to teach medical AI agents

• OpenAI reveals its Model Specs

• Researchers use AI to understand whales

• AI helps country singer release new song

• And more.

🧬Google DeepMind introduces AlphaFold 3 (6 min)

AlphaFold 3 is the result of a joint effort between Google DeepMind and Isomorphic Labs. This AI model excels at predicting the structure and behaviour of molecules, such as proteins, DNA, RNA, and ligands. Based on AlphaFold 2, this new iteration showcases improvements in the prediction of protein interactions and extends its reach to a broader spectrum of biomolecules.

With the release of AlphaFold 3, both companies expect to transform our current understanding of the biological world and bring more innovations to science, especially in drug design and genomics research.

🏥 “Agent Hospital”: A simulation to teach medical AI agents (3 min)

Simulations can result in huge learning opportunities, especially for AI agents. That's what a group of researchers from Tsinghua University are proving with their approach, introducing an AI-powered simulation of a hospital that can serve two types of AI agents: medical professionals and residents who can also be potential patients. In this environment, AI agents focused on medical endeavours can be trained and evolve at their own pace accordingly.

The current experiments, with over 10,000 simulated patients, showcase the agents' improvement in examination, diagnosis, and treatment, with an accuracy of 88%. Despite the training not using real-world data, the agents surpass several benchmarks.

📄 OpenAI reveals its Model Specs (2 min)

As a means to promote greater transparency in AI and continue their commitment to responsible development, OpenAI has recently shared a framework to shape the behaviour of models like GPT-4. Named "Model Spec," it relies on three principles: assisting developers and users, benefiting humanity, and reflecting OpenAI's approach to social norms and laws.

The primary goal of sharing this set of rules is to gather public input to later refine the behaviour of models, as well as to showcase how their reasoning and intentions function.

An entirely new way to present ideas

Gamma’s AI creates beautiful presentations, websites, and more. No design or coding skills required. Try it free today.

🐋A research group from MIT's CSAIL, alongside the nonprofit Project CETI, is trying to decipher cetacean language using AI. The mix of algorithms includes pattern recognition and classification, which has led to the discovery that whale sounds are not "random" and involve complexities similar to human communication.

🎙️Randy Travis, an American country singer, has been given the chance to sing again thanks to AI. The musician lost his voice in 2013 due to a stroke, and an AI model trained on 42 of his vocal-isolated recordings has generated a new song called "Where That Came From."

⚡️Quick links

Microsoft’s AI Copilot is beginning to automate the coding industry

OpenAI partners with People publisher Dotdash Meredith

Logic Pro gets some serious AI—and a version bump—for Mac and iPad

Sam Altman doesn't think we are worried enough about how AI will impact the economy

LinkedIn shares insights into the challenges of putting LLMs to work

Thank you for reading!