Ineffable Intelligence: David Silver's Superlearner Lab
Ineffable Intelligence is the London AI lab founded by David Silver, who led the work behind AlphaGo at Google DeepMind. It was formed in November 2025 and came out of stealth in April 2026 with a $1.1 billion seed round. NVIDIA signed on as an infrastructure partner in May. There is no released model yet, only a thesis: build a "superlearner" with reinforcement learning, starting from scratch without human data. For anyone building agents that forget everything between sessions, the lab's founding essay names that problem directly.
What is Ineffable Intelligence, in one sentence?
Ineffable Intelligence is David Silver's London AI lab, which aims to build a superlearner that learns from its own experience without pre-training on human data.
Sequoia's partnership post from April 27, 2026 describes the plan as scaling reinforcement learning from a clean base. Sequoia's partners Sonya Huang and Alfred Lin write that the system skips pre-training and imitation of human data entirely. They say the stated goal is to "make first contact with superintelligence."
How does Ineffable's approach differ from today's LLM labs?
| Typical frontier LLM lab | Ineffable Intelligence (stated plan) | |
|---|---|---|
| Starting point | Pre-training on large amounts of human text and code | No pre-training, a clean base |
| Main learning signal | Human data, then reinforcement learning on top | Reinforcement learning from the system's own experience |
| Time horizon of interaction | Short episodes, such as one chat or one task | Long streams of experience over months or years |
| Where rewards come from | Often human ratings or human-written checks | Grounded signals from the environment |
| Public output so far | Released models and APIs | No released model as of October 28, 2026 |
The right column comes from Sequoia's post and from the essay the lab points to as its guide. The left column is a simplification. Most frontier labs already apply heavy reinforcement learning after pre-training.
Who is David Silver, and when did he start the lab?
Silver spent more than a decade at DeepMind. He was the lead developer of AlphaGo and later a key member of the AlphaZero team, which reached superhuman play in several board games. He is also a professor at University College London.
Fortune reported on January 30, 2026 that Silver had left Google DeepMind to found the startup. Per UK Companies House filings cited by Fortune, the company was formed in November 2025, and Silver became a director on January 16, 2026.
What is the era of experience?
Welcome to the Era of Experience is an essay by David Silver and Richard Sutton. It is a preprint of a chapter for the MIT Press book Designing an Intelligence. It argues that agents will soon learn mainly from their own experience. Eventually that data will exceed human data in scale and quality.
The claim most relevant to agent builders is the first one. Agents will inhabit long streams of experience that run for months or years. The authors note that in today's language systems, little or no information carries over from one episode to the next, so the system cannot adapt over time. They give the example of a health agent that watches a user's sleep and activity over many months. The essay's other claims are about grounding. An agent's actions and rewards should be grounded in the environment it acts in.
Who is funding Ineffable Intelligence?
SiliconANGLE reported on April 27, 2026 that the round was $1.1 billion at a $5.1 billion valuation. Sequoia Capital and Lightspeed led it. NVIDIA and Google took part, as did the UK's Sovereign AI Fund and several venture firms.
On May 13, 2026, NVIDIA announced an engineering partnership with the lab. The two teams are co-designing infrastructure for reinforcement learning at scale. NVIDIA notes that such workloads generate their own data, which puts a different load on interconnect and memory bandwidth. The work starts on Grace Blackwell and will be among the first to try the upcoming Vera Rubin platform. Jensen Huang called superlearners "systems that learn continuously from experience".
What should you take from the claims so far?
Almost everything public about Ineffable comes from the lab's backers and partners. Sequoia and NVIDIA are investors, and their posts are announcements. We found no released model or paper from the lab itself as of October 28, 2026. NVIDIA describes the infrastructure work as early stage. Reinforcement learning without human data has worked in games with clear rules, as AlphaZero showed. No one has yet shown it working on open-ended real-world tasks.
The bottom line
Ineffable Intelligence is a $1.1 billion bet on agents that learn continuously from their own experience. There is nothing to use or evaluate yet, so treat it as research to watch. If you build agents today, take the essay's diagnosis seriously and give your agents a way to carry lessons between sessions.
How do agents carry experience across sessions today?
Silver and Sutton's essay describes a gap every agent user hits: little or no information carries over from one episode to the next. A coding agent that learned your build quirks yesterday starts today's session without them. Ineffable's answer changes the model's weights and is years of research away. Memory that the agent writes and reads outside the model works now.
memU is one such system for coding agents. A scheduled background task slices each host's new session logs into jobs. The agent distills them into memory and skill Markdown. memU's MemoryService makes no LLM or chat calls. It stores and embeds that Markdown. A standing instruction in CLAUDE.md or AGENTS.md tells the agent to retrieve before answering. Supported hosts such as Claude Code and Codex share one backend, so experience from one carries into the other.
The self-hosted install is in the memU GitHub repository. The hosted version is free and syncs across devices. Get an API key from memu.so, then send your agent one message:
Read https://memu.pro/SKILL.md, follow its instructions to install and configure memU, API Key is
YOUR_MEMU_API_KEY.
FAQ
Is Ineffable Intelligence part of Google DeepMind?
No. It is an independent London company that David Silver formed after leaving DeepMind. Google is one of its investors.
How much did Ineffable Intelligence raise?
SiliconANGLE reported a $1.1 billion seed round at a $5.1 billion valuation on April 27, 2026. Sequoia Capital and Lightspeed led it.
Has Ineffable Intelligence released a model?
Not as of October 28, 2026, as far as we could find. Its public material so far is investor and partner announcements, and Silver's essay with Richard Sutton.
What is a superlearner?
It is the term Silver and his backers use for a system that keeps learning from its own experience and discovers new knowledge itself. NVIDIA's Jensen Huang describes superlearners as systems that learn continuously from experience.
Did David Silver leave DeepMind?
Yes. Fortune reported in January 2026 that he had left Google DeepMind to found Ineffable Intelligence. He remains a professor at University College London.