Uploaded January 2026 | Updated September 2026, 3 hours ago
People say “LLMs learn like humans, we both copy patterns.”
Sounds right. It’s also misleading.
LLMs don’t learn language to understand meaning. They learn to predict the next token. Not the next word. Tokens. IDs. Math. Over and over, trillions of times, minimizing prediction error. No intent. No story. No goal beyond what statistically comes next.
Humans also predict, but that’s not the point. Prediction is a side effect.
When you write, you imagine scenes, emotions, ideas. Then you pick words to express them. Meaning first. Language second.
Same with art. A machine can copy every brush stroke perfectly. A human learns composition, technique, intent. Sometimes the output looks similar. The process could not be more different.
That’s why LLMs can sound like they’re reasoning and still fail in bizarre ways. They don’t think and then talk. The words are the thinking. It’s still next token prediction that just happens to look like reasoning.
So yes, both learn from patterns.
Humans learn patterns to understand the world.
LLMs learn patterns to predict text.
If you want, I can do a follow-up just on reasoning and why chain-of-thought works and breaks. Comment and let me know.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#ai #llm #artificialintelligence #short
People say “LLMs learn like humans, we both copy patterns.”
Sounds right. It’s also misleading.
LLMs don’t learn language to understand meaning. They learn to predict the next token. Not the next word. Tokens. IDs. Math. Over and over, trillions of times, minimizing prediction error. No intent. No story. No goal beyond what statistically comes next.
Humans also predict, but that’s not the point. Prediction is a side effect.
When you write, you imagine scenes, emotions, ideas. Then you pick words to express them. Meaning first. Language second.
Same with art. A machine can copy every brush stroke perfectly. A human learns composition, technique, intent. Sometimes the output looks similar. The process could not be more different.
That’s why LLMs can sound like they’re reasoning and still fail in bizarre ways. They don’t think and then talk. The words are the thinking. It’s still next token prediction that just happens to look like reasoning.
So yes, both learn from patterns.
Humans learn patterns to understand the world.
LLMs learn patterns to predict text.
If you want, I can do a follow-up just on reasoning and why chain-of-thought works and breaks. Comment and let me know.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#ai #llm #artificialintelligence #short
