UC Riverside researchers just dropped a new method that lets artificial intelligence reason more like humans—without feeding it extra data. Their Test-Time Matching technique helps even smaller vision-language models outperform big names like GPT-4.1 on compositional reasoning tasks. If smarter evaluation and self-improvement can unlock hidden potential, do we need to keep making models bigger, or just use them better? #Tech #AI #MachineLearning