Yes, that’s correct. Ineffable Intelligence has raised $1.1B according to the London Times this week. The company claims that they are creating an AI that can discover knowledge on it’s own without human data. They are quoted in paper saying that their discoveries could equal Darwin and transcend current inventions in mathematics, science and technology. The approach is unproven, but is it realistic?
First, Ineffable is founded by David Silver who led the development of Alpha Go. Investors in the company include Sequoia, Lightspeed and Nvidia. The core technical claim of the superlearner concept is a paradigm shift in how an AI could achieve superintelligence. The current dominant approach (OpenAI, Anthropic) uses LLMs that absorb human-generated text and data to enable pattern mimicking. Ineffable uses reinforcement learning where the AI learns through trial and error in environments where it can discover knowledge independently.
I understand that the company believes that the LLM approach to AI will fail to achieve superintelligence. (Other leading AI pioneers like Yann LeCun has also argued that LLMs are a dead end for superintelligence or AGI.) Ineffable are aiming at a system that discovers knowledge independently and theoretically transcending human knowledge boundaries. As of 2026, scaling this to general intelligence across all domains is unproven. From what I can gather, several significant hurdles exist:
| Challenge | Why It Matters |
|---|---|
| Sample Efficiency | Reinforcement learning (RL) typically requires millions of trials; scaling to general knowledge domains is computationally expensive |
| Safety & Alignment | Systems that learn independently may develop goals misaligned with human values |
| Infrastructure | Training such systems likely requires unprecedented compute resources |
| Evaluation | How do you measure progress when the system discovers knowledge beyond human understanding? |
The $1.1B seed round reflects investor belief that Ineffable have a first mover advantage. But are they asking the tough questions with regards to feasibility? For instance:
- What specific RL architectures are being scaled?
- What environments will the superlearner train in? Physical simulations or digital worlds?
- How will safety be ensured during autonomous discovery?
- What timeline does Silver envision for meaningful breakthroughs?
None of these questions have been publicly answered. Ineffable’s credibility comes in part from AlphaGo/AlphaZero-style breakthroughs, but those were structured domains with clear objectives. Open-ended scientific discovery is a different category of problem they try to solve with RL.
The $1.1B seed round is proof that investors are willing to fund an alternative to the current approach. The real test is whether Ineffable can scale from gaming and simulations to open-ended discovery. The company is fast becoming one of the most expensive scientific bets in AI. Superintelligence transcending the greatest inventions in human history is something we could easily dismiss, but I’d rather approach the topic with disciplined skepticism.