Welcome to the Awesome MLSS Newsletter!

We’ve all been there - eyes scrolling through a paper with manic focus to find that one statement or data point that supports your own paper. For years, we did this manually, cup of coffee gone cold late at night.

With ChatGPT, and other tools like Elicit, we have found better ways to source and analyse these papers. 

It works, mostly. Maybe a simple RAG + Summarization engine is not enough.

Google might have just changed the game. More on this later.

Upcoming Summer School Announcements

The applications for the following summer schools are closing in the next few weeks, make sure to apply to them before the application deadline!

For the complete list, please visit our website

Some Research Highlights

What’s happening in AI?

So we spoke about the pain of reading, and writing research papers. 

So far, we have been given several tools to go through literature, create better search engines for research with vector search, and a summarisation model that explains what the search results found.

Often, it’s a hit and miss. We all know that.

Google, however, decided to change the paradigm. Enter the AI Co-Scientist.

They created an agentic ensemble of models, each fine tuned and specialised, to create an end to end system which automates the research process, and brings it much closer to how humans really think.

The system incorporates a scientist-in-the-loop paradigm, where the original research hypothesis is supplied by the researcher, who then uses AI Co-Scientist as a tool for logical analysis across domains. The hypothesis is then continuously updated based on feedback from the scientist.

The key differences between other platforms and AI Co-Scientist are

• It has significantly scaled up test time compute with Gemini 2.0 as the underlying model

• It relies very heavily on agentic flows, as opposed to simple RAG summarisation

• Each agent has been specially trained for its individual task, ensuring better logical performance

• Agents can be added. As an example, in the paper, they used DeepMind’s AlphaFold as an external agent to assist with protein synthesis problems

The system is already showing promising results, with greater scores for accuracy and novelty when compared to other platforms.

Could this change the way scientific research is conducted in the future? Let us know what you think!

To read more, head on to https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/

With love, Awesome MLSS