Quoting Jeremy Howard
Jeremy Howard argues top AI labs should bar themselves from using their best models for frontier research while sharing them widely.
“Anthropic has chosen the opposite of the safe path: they are allowing themselves, the current top lab, to use their top model for frontier AI research. They've said they'll sabotage others who try.”
Jeremy Howard proposes that to slow recursive AI self-improvement, the leading lab should forbid itself from using its top model for frontier research while granting everyone else access, thereby halting frontier advancement and reducing power imbalance. He criticizes Anthropic for doing the opposite by reserving its best model for its own frontier work, though he personally favors democratizing AI rather than slowing it. It matters as a pointed critique of how concentration of frontier AI capability creates governance and power-balance concerns.