Guest: Richard Socher
Listen to the original ↗
The hosts and guest Richard Socher discuss Recursive’s $670 million investment in recursively self-improving AI and the timeline for artificial superintelligence. They also examine recent developments in AI, including the release of Sonnet 5.5, the Pentagon’s Project Meridian, and the potential for AI to accelerate scientific discovery.
Signal
- Discussed the potential obsolescence of the Turing Test, arguing that passing it now requires AI to throttle its intelligence to human levels, rendering the benchmark useless for measuring true capability.
- Examined the paradox of “flourishing” mathematics, suggesting that AI automation does not destroy the field but rather shifts human focus toward creative formalisms and high-level abstraction.
- Questioned the feasibility of banning recursive self-improvement, noting that enforcement would require a totalitarian surveillance state capable of monitoring private local inference on personal devices.
- Compared the release of Sonnet 5.5 to a cost-performance regression, highlighting that the model appears inferior to Opus 5.5 on cost grounds despite significant benchmark jumps.
- Outlined the strategic shift toward decision models like Jev, which prioritize fast, intuitive system-one responses for micro-coordination over slow, reasoning-heavy system-two generation.
Participants
- Hosts: Peter H. Diamandis, Salim Ismail, Dave Blundin, Alexander Wissner-Gross
- Guest: Richard Socher
Warning: This page is for information only. The review is AI-generated by Sparky, a local language model, and can contain mistakes. Listen to the original episode before acting on anything here.