Deploying AI to Train AI
Kona and Aleph - new approaches to AI and Agentic AI - from Logical Intelligence, which just brought on Yann LeCun as Founding Chair.
My takeaway from the Google DeepMind and Anthropic Davos talks was that if the frontier model builders (in this case Gemini and Claude) can figure out how to have AI train LLMs, we will see a step function in performance improvement. Otherwise, it will be more of the same - incremental improvement - from model release to release.
Yann LeCun became founding chair at Logical Intelligence (HERE). You may test their reasoning model - Kona 1.0 - HERE as it competes against LLMs in solving puzzles and Sudoku.
Logical Intelligence is piloting an AI Agent - Aleph - which automates formal verification end-to-end and produces machine-checkable proofs that critical logic behaves correctly across every execution path.
CEORater news article from January 24th: “CEORater AI Chat: Tell Me About Elon Musk”
