Google uses tech behind ChatGPT to find new knowledge in major breakthrough

LLMs such as ChatGPT have shown great promise – but questions have been asked about whether they can actually find anything new

Andrew Griffin
Friday 15 December 2023 18:30 GMT
Google Android App Attack
Google Android App Attack (Copyright 2023 The Associated Press. All rights reserved.)

Google says it has used the technology behind ChatGPT to find new knowledge.

Ever since ChatGPT was released almost exactly a year ago, artificially intelligent chatbots have become hugely popular. They rely on technology known as large language models, which use a vast corpus of literature to create new text.

While they have been praised for their use in answering questions, bringing together information and helping with practical questions such as code, there has been some question over whether they can actually find out anything new. What’s more, they have been known to “hallucinate”, or make up information, meaning that even apparently new information could be wildly wrong.

Now Google says that it harnessed the best of LLM technology to find out new information in mathematical sciences. It has called the breakthrough “FunSearch” and it is described in a new Nature paper published this week.

The technology uses a pre-trained large language model akin to ChatGPT, built to provide creative solutions to questions in the form of computer code. But it works alongside an “evaluator” – another automated system that can watch for hallucinations and wrong ideas.

The two systems can then be paired with each other to “evolve” into new knowledge.

It marks the first time that a new discovery has been made for difficult open problems in science or maths using LLM technology, google said. In this case, it was used to find new solutions for the “cap set problem”, a specific open question in mathematics, but it could be used more broadly, Google said.

“Scientific progress has always relied on the ability to share new understanding. What makes FunSearch a particularly powerful scientific tool is that it outputs programs that reveal how its solutions are constructed, rather than just what the solutions are,” it said in its announcement.

We hope this can inspire further insights in the scientists who use FunSearch, driving a virtuous cycle of improvement and discovery.”

Join our commenting forum

Join thought-provoking conversations, follow other Independent readers and see their replies


Thank you for registering

Please refresh the page or navigate to another page on the site to be automatically logged inPlease refresh your browser to be logged in