Ineffable Intelligence raises $1.1 billion at $5.1 billion valuation to build AI “super learners”

Ineffable Intelligence Ltd, a British artificial intelligence startup founded a few months ago, has raised $1.1 billion in seed funding.

The investment values ​​the company at $5.1 billion. CNBC reported It was announced today that Lightspeed Ventures and Sequoia Capital led the transaction. Nvidia Corp., Google LLC, UK sovereign AI fund, DST Global, Index Ventures and others participated.

Ineffable Intelligence is led by renowned AI researcher David Silver. He previously worked at Alphabet Inc.’s machine learning division, DeepMind, for more than a decade. He was the lead developer of a groundbreaking product. AlphaGo A model that the same lab debuted in 2016.

AlphaGo defeated the world’s highest-ranked Go player in a series of games watched by more than 200 million people. DeepMind then used the technology behind the model to build a mathematically optimized AI. alpha proof. In 2024, AlphaProof became the first neural network to win a medal at the International Mathematics Olympiad.

AI models answer user questions based on existing knowledge gleaned from sources such as the public web. According to wiredIneffable Intelligence aims to build AI models that can acquire entirely new knowledge. The company believes such algorithms, called superlearners, can accelerate scientific research and engineering projects.

Ineffable Intelligence plans to use reinforcement learning to develop super learners. This is a common AI training method suitable for building large-scale language models, among other tasks.

The core idea behind reinforcement learning is to provide the AI ​​with a sample task and observe how well it completes it. That monitoring is typically performed by a second, less sophisticated neural network called a teacher model. Use it to provide feedback to the AI ​​during training and improve its output.

There are many implementations of reinforcement learning. One common approach is to equip the AI ​​model with a database of common situations encountered in production. This approach, known as model-based reinforcement learning, speeds up AI training.

Other reinforcement learning techniques do not use databases and instead prioritize model adaptability. Another method, RLHF, provides a teacher model that uses human feedback to guide the AI ​​training process. Human feedback greatly improves the quality of AI output.

Reinforcement learning is typically applied to a model that has already been trained through a process known as pre-training. Ineffable Intelligence plans to skip that step. Additionally, we put AI models into simulations and allow them to learn from each other.

AlphaGo used a similar approach. We played millions of Go games to develop new tactics that were not available in our training dataset.

Ineffable Intelligence’s funding comes two months after AMI Labs Inc., another early-stage AI startup with a prominent founder. safe Funding amount is $1.03 billion. CEO Yann LeCun said at the time that the company plans to develop a global model optimized for tasks such as improving aircraft parts design.

image: unsplash