Technical details of Google DeepMind's Gemini Models in Medicine. In-person lecture

Hello, fellow human! Please check out the talks on the state-of-the-art AI research we’ve lined up for you.

If you are a San Francisco Bay Area local, please come to our first in-person event featuring Google DeepMind researchers. They'll be presenting Med-Gemini, a family of highly capable multimodal models specialized in medicine. These models have already shown the best results in most medical benchmarks and have established new state-of-the-art performance. 

Learn the details of the event here (special thanks to Nordic Innovation House for providing us with the venue).

In case you can’t make it to the event, the video recording will be available on our YouTube channel.

For the next two weeks, our virtual talks will be dedicated to AI safety and ethics.

Next Thursday, on July 25th, we are hosting a researcher from Oxford University who will present the research work: Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems. This ambitious work involves researchers from x.AI, MIT, Stanford University, UC Berkeley, Cornell University, Oxford University, Mila, Carnegie Mellon University, and Columbia University, among others, as well as prominent scientists like Yoshua Bengio and Max Tegmark.

In other words, this is truly a community effort to develop a meaningful framework for safe AI. Together with one of the authors of this work, we will discuss it in more detail and consider whether it is even possible to build guaranteed safe AI.

Details and registration are available here.

On Thursday, July 18th, we will be discussing with a researcher from Google DeepMind's Ethics Research Team the ethical risks associated with advanced AI systems, particularly in terms of their relationships with humans.

Check out the details here.

Lastly, I wanted to share a video recording of our latest lecture about a new optimization method for Learning from Human Feedback in Large Language Models. The new method has proved to be more cost-effective and performant than most widely adopted optimization methods like PPO (Proximal Policy Optimization algorithm developed by OpenAI).

Watch the lecture here.