Will LLMs ever reach AGI?

People disagree about whether LLMs will ever reach AGI.

The No case

• LLMs predict text from patterns. Critics argue this is different from reasoning, planning or truly understanding a problem.
• They mainly learn from text and images rather than real-world experience, touch and movement.
• They can struggle with completely new situations that do not match their training data.
• Making models bigger and adding more data may not be enough. Quality training data is limited, and next-token prediction may never become genuine understanding.

The Yes case

• LLMs have shown some abilities they were not directly trained to perform. In 2025, Google's AI system found new, faster ways to do matrix multiplication, a core maths operation behind all AI.
• Newer systems combine LLMs with memory, tools and real-world sensors, so text prediction may be only one part of future systems.
• Some AGI predictions have moved closer as AI systems have improved faster than expected.