Decoupling Internal Representational Changes in Fine-Tuned LLMs: What Really Drives Task Performance?

Deep dive into how fine-tuning reshapes LLM internals and the surprising decoupling of representational changes from causal importance.

Deep dive into how fine-tuning reshapes LLM internals and the surprising decoupling of representational changes from causal importance.
A 60-second tour of VideoEval scoring 10 AI-generated clips: prompt adherence, color saturation, motion ghosting, pass/fail verdicts, and the human review queue. Real demo, real numbers.