Credit By: LinkedIn
Groundbreaking: Language Models Exceeding Information Repackaging
Using a large language model (LLM) like ChatGPT, AI researchers at Google DeepMind have made history by employing this technology to make the first-ever scientific discovery. This discovery raises the possibility that sophisticated language models could produce complete findings that are beyond human comprehension in addition to repurposing already-known knowledge.
FunSearch: An Innovative Approach to Problem Solving
To delve deeper into the potential of language models, DeepMind unveiled “FunSearch,” a technique for “searching in the function space.” This novel methodology deviated from conventional problem-solving techniques using computer programs to solve difficulties. Together with an “evaluator” who graded programs according to their effectiveness, the LLM functioned. The top-performing programs were merged and put back into the LLM to enhance it even more, leading to the progressive development of programs that could uncover new scientific findings.
Revealing Mathematical Wonders: FunSearch at Work
The researchers targeted two challenging puzzles with FunSearch. First off, FunSearch created programs that produced enormous cap sets that outperformed the most renowned mathematicians. This solved the complex problem of locating the most incredible collection of points in space such that no three points make a straight line. The second task asked students to solve a mathematical issue about effectively fitting objects of various sizes into containers. It was called the bin packing problem. FunSearch found a more effective method that did not leave tiny gaps unlikely to be filled.
Bringing Human and AI Cooperation Together in Mathematics
“This work potentially gives us another exciting tool for such collaborations, enabling mathematicians to search efficiently for clever and unexpected constructions,” said Cambridge University mathematics professor Sir Tim Gowers in response to a question about the significance of this work. Even better, these constructs can be understood by humans.
With this accomplishment, AI and human mathematicians may work together more effectively and discover innovative answers to challenging mathematical problems more quickly.
Follow ARP Media for more informative blogs.