An interview with ex-DeepMind scientist David Silver on his startup Ineffable, building an AI "super learner" using RL as an alternative path to LLMs, and more

You can now find AI that learns complex tasks from scratch without needing massive pre-trained datasets.

David Silver, a key figure behind Google DeepMind's AlphaGo, has co-founded a new AI startup named Ineffable. This venture aims to develop a "super learner" that utilizes reinforcement learning (RL) as its primary method. Silver believes this approach offers an alternative to the current dominance of large language models (LLMs). His work on AlphaGo demonstrated the power of RL in achieving superhuman performance in complex games, and he aims to apply similar self-teaching principles to broader AI challenges. Silver expressed a sense of moral obligation to advance AI technology, suggesting a focus on fundamental learning capabilities rather than just scaling existing models.

Ineffable's focus on reinforcement learning represents a significant bet against the prevailing trend of scaling LLMs. By prioritizing self-teaching and reward-driven learning, Silver's startup seeks to unlock more generalizable intelligence, potentially bypassing the immense computational and data requirements of current LLM development. This could lead to AI systems that are more efficient to train and more adaptable to novel environments or tasks.

The success of Ineffable could redefine the path to advanced AI, offering a compelling alternative for researchers and developers disillusioned with the cost and limitations of LLMs. It suggests a future where AI's core capabilities are built on learning algorithms rather than brute-force data absorption, potentially impacting hardware design, research funding, and the very definition of artificial general intelligence.