Researchers created a new method using a very small computer model, called TRM, to solve difficult puzzles that even giant models struggle with. Instead of being huge and complex, TRM uses a tiny network that repeatedly thinks through the problem step-by-step to improve its answer. This simple but powerful "recursive" process allows the tiny model to perform much better on hard tasks like Sudoku and logic puzzles. TRM is more efficient and gets better results with far less computing power, showing that a smaller, focused approach can be more effective than a giant, all-purpose one.
Tiny AI as good as big AI
A new, tiny model solves hard puzzles better than huge ones by repeatedly thinking about the problem.