these two AI lab CEOs just admitted it...

Quick Overview

Dario Amodei and Demis Hassabis confirmed in "The Day After AGI" interview that they are on track for AGI milestones, with Dario predicting a model capable of human-level noble laureate performance by 2026-2027 and Demis giving a 50% chance of matching human cognitive capabilities by the end of the decade, while also acknowledging significant economic disruption and the lack of broad societal preparation for these changes.

Key Points: Dario Amodei projects a model capable of human-level noble laureate performance across many fields is on track for 2026-2027, while Demis Hassabis states there is a 50% chance a system exhibits all human cognitive capabilities by the end of the decade. Dario noted that engineers in areas like coding are already transitioning to being editors rather than writers, predicting this shift will occur within 6 to 12 months, referencing how Claude code generation models now make most contributions to Claude code. The leaders agree that while verifiable output areas like coding and math advance quickly, areas requiring long experimental loops, such as complex chemical compounds or physics predictions, are slower due to missing architectural pieces like continuous learning loops (short-term/long-term memory). Dario stated that the revenue growth projection for AI models is insane, potentially going from $1B to $10B in 2025, suggesting the industry might become self-sustaining, though he also reinforced his past warning of a "white collar bloodbath" hitting entry-level jobs hardest. Demis expressed surprise that few economists are deeply thinking about the massive economic shifts coming, noting that the transition period might be rough because humans are not logical, long-term thinkers, and he trusts Dario and Ilia Sutskever to navigate the AI race responsibly. Dario strongly opposed selling advanced chips to China, arguing that limiting chip sales forces the race between the US and China to become a race between Google DeepMind and Anthropic, which they can control better, comparing chip sales to selling nuclear weapons for profit. Both CEOs are skeptical of AI doomers, believing risks like bioterrorism and autonomous machine danger are manageable if development pace is correct, though Demis acknowledges public fear is reasonable given job displacement and notes that post-scarcity meaning may come from pursuits like extreme sports, arts, and games.