“No LLM-based AI will surpass 70% on the ARC-AGI-3 leaderboard, with a cost of $1000 or less, before June 2028.”

LLM-based AIs have no mechanism for true reasoning. Despite claims otherwise, they cannot think logically, nor solve tasks that require novelty, abstraction, planning, and dynamic interaction. The ARC-AGI-3 challenge tests for these capabilities, and I think it will continue to defeat state-of-the-art LLMs for at least two years.