Ex-DeepMind scientist David Silver on Ineffable’s “super learner” using RL as an alternative to LLMs
deepmind google startup
| Source: Techmeme | Original article
Former DeepMind AlphaGo architect David Silver discusses his startup Ineffable, which is developing an AI “super learner” that uses reinforcement learning as an alternative to large language models.
David Silver, the DeepMind researcher who led the AlphaGo breakthrough, sat down with the Financial Times to discuss his new venture, Ineffable. The startup is pursuing what Silver calls a “super learner” – an artificial‑intelligence system that relies on reinforcement learning (RL) rather than the massive text corpora that underpin today’s large language models (LLMs).
Silver argues that RL offers an alternative route to general‑purpose intelligence, allowing machines to teach themselves through trial and error instead of ingesting billions of words. By framing learning as a series of reward‑driven interactions, Ineffable hopes to build agents that can acquire skills more efficiently and with less dependence on curated data. The interview touches on the technical challenges of scaling RL, the team’s early prototypes, and how the approach could complement or even compete with the dominant LLM paradigm.
The significance lies in broadening the AI research agenda. While LLMs dominate headlines and commercial deployments, a successful RL‑based “super learner” could reshape how developers think about data, compute costs, and model generalisation. It also raises questions about safety and controllability, given RL’s propensity for unexpected strategies.
What to watch next: Ineffable’s upcoming benchmarks against established LLMs, potential funding rounds, and partnerships with hardware providers that can support the intensive simulation environments RL demands. Industry observers will be keen to see whether Silver’s vision can translate into a viable, scalable alternative to the current text‑centric AI wave.
Sources
Back to AIPULSEN