Tiny AI models solve problems well with better selection logic

Original: 1/ Weak LLMs generate correct solutions in their latent space all the time—they just fail to select them.

Source: x.com ↗