LeCun JEPA Theorem Shows a 109M Model Can Out-Align 72B Language Models
A new paper from Randall Balestriero and Yann LeCun turns the Platonic Representation Hypothesis into a theorem for JEPAs and reports that a 109M-parameter text model aligns with vision encoders better than language models up to 72B parameters, using 42 times fewer training samples.
Read more
