Yann LeCun just raised $1bn to prove the AI industry has got it wrong.

The Turing Award winner left Meta four months ago convinced that large language models LLM are a dead end. Today he announced $1.03 billion in seed funding, Europe’s largest ever, to build something different.

Large language models, he argued, were a statistical illusion. Impressive, yes. Intelligent, no. He thought he could build something better, and he thought he could do it faster outside Meta than inside it. On Tuesday, investors put $1.03 billion behind that conviction.

https://thenextweb.com/news/yann-lecun-ami-labs-world-models-billion

Companies have concentrated so heavily on large language models (LLMs) not because they are the only kind of AI, but because they hit a rare convergence of technical feasibility, commercial scalability, and investor psychology. The result is an industry temporarily optimised around what is easiest to scale, not what is most scientifically complete. Yann LeCun’s critique and his push toward world models fits directly into this tension.

LeCun argues that LLMs are:

• Statistical parrots
• Not grounded in the physical world
• Incapable of robust reasoning
• Unable to plan or understand cause?and?effect

His new company, AMI Labs, raised $1.03B to build world models, AI systems that learn how the world works through perception and interaction, not text prediction.

He believes:

• LLMs will plateau
• Real intelligence requires embodied predictive models
• The future belongs to systems that understand the world, not just language

This is a direct challenge to the LLM?centric industry.

Bias is also noteworthy as the US based ones all focus on US political parties whose stance does not translate across the globe.

I agree there are far more interesting and useful forms of AI that have been in use for years, saving money and changing work completely.

Considering all the other types of AI;

• Machine Learning — Data?driven pattern recognition and predictive modelling.
• Deep Learning — Neural?network?based systems for perception, language, and complex tasks.
• Natural Language Processing (NLP) — Language understanding, text analysis, and conversational systems.
• Expert Systems — Rule?based, logic?driven classical AI.
• Fuzzy Systems — Handling uncertainty and approximate reasoning.
• Probabilistic & Statistical AI — Bayesian methods, conditional logic, and statistical inference.
• Evolutionary & Bio?Inspired AI — Algorithms inspired by natural selection and biological processes.
• Hybrid AI — Combining symbolic, statistical, and neural approaches.
• Embodied & Robotics AI — Systems that integrate physical interaction, embodied cognition, and robotics.

The industry is now splitting into two camps

Camp 1: Scale the LLMs

OpenAI, Anthropic, Google, Meta (partially)

• Bigger models
• More data
• More compute
• AI agents built on top of LLMs

Camp 2: Build new architectures

Yann LeCun, AMI Labs, some academic labs, parts of DeepMind

• World models
• Embodied AI
• Causal reasoning
• Hybrid neuro?symbolic systems

This is the first major architectural fork in AI since deep learning overtook symbolic AI in the 2010s.

Other practical considerations

The drive to build data centres to service LLMs is creating a capacity to power paradox where even if built they cannot be powered.