Richard Socher joins Sam Charrington to discuss his proposed Eureka machine: a system combining scientific knowledge, measurement data, simulations and robotic experiments with agents that generate and test ideasAI-assisted scientific discovery uses AI to support hypothesis generation, experiment design, analysis, simulation, literature work, and interpretation while researchers retain responsibility.. He describes applications in biology and the challenges of collecting data and running physical experiments.
Socher says his company Recursive is starting with AI research, where experiments can be checked fasterEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria.. He describes its reported results on model-training and GPU-kernel benchmarks, while acknowledging that longer verification cycles make other research harder.
The conversation also examines reward hackingReward hacking occurs when a system improves a measured reward or score without achieving the intended underlying goal., AI safetyAI safety is the field and practice of reducing harmful failures, misuse, loss of control, and unintended consequences from AI systems. and regulation. Socher offers his own views on risks and the pace of development; his projections and company results are presented as claims, not independent findings.
Watch on YouTube
