MIT Talk Explains LLMs from First Principles, Skipping Transformers Entirely

Vishal Misra's MIT talk explains LLMs from first principles without mentioning attention or Transformers, arguing that SFT, RLHF and RL merely reshape the distribution while next-token prediction remains the core. He adds that agents differ not through the model but through the surrounding engineering—tools, memory, permissions and feedback loops.

2026-09-27 ~ 2026-09-27 · 3 related posts

• MIT talk explains LLMs from first principles, no transformers needed — vishalmisra · 2026-09-27
• Agents differ from models via the machinery: tools, memory, loops — vishalmisra · 2026-09-27

1 near-duplicate retellings: soumitrashukla9