Google says it uses the technology behind ChatGPT to find new information.
Since the launch of ChatGPT almost exactly one year ago, AI chatbots have become enormously popular. These bots are based on a technology known as large language models (LMMs), which use a vast corpus of literature to generate new texts.
While their use has been praised for answering questions, gathering information and helping to answer practical questions about things like code, there are questions about whether this technology can actually find anything new. Moreover, these tools are known to “hallucinate” – that is, to make up information – which means that even seemingly new information can be highly inaccurate.
Already, Google says it is making the most of GDM technology to find new insights in the mathematical sciences. This breakthrough, called “FunSearch”, is described in a new paper published this week in Nature.
Using a ChatGPT-like pre-trained broad language model, this technology was created to provide creative solutions to questions in the form of computer code. On the other hand, the “assessment tool”, another automated system that this system works with, remains vigilant against hallucinations and false ideas.
These two systems can then be paired with each other to “evolve” new information.
Google says that this is the first time a new discovery has been made using GDM technology in the field of challenging unsolved problems in science or mathematics. In this case, Google says the tool was used to find new solutions to a specific unsolved problem in mathematics, the “cap set problem” (i.e., the problem of how many elements can be put into a set such that no more than three of them form a line), but it can be used more broadly.
The company stated the following in its announcement:
Scientific progress has always relied on the ability to share new insights. What makes FunSearch a particularly powerful scientific tool is that it produces programs that reveal not just what the solutions are, but how those solutions are created.
We hope this will inspire further insights in scientists using FunSearch and create a virtuous cycle of development and discovery.
https://www.independent.co.uk/tech
