Good News - AGI Maybe 10 Years Away (Patreon)

Andrej Karpathy, one of the founding members of OpenAI, on Friday threw cold water on the idea that artificial general intelligence is around the corner. He also cast doubt on various assumptions about AI made by the industry’s biggest boosters, such as Anthropic’s Dario Amodei and OpenAI’s Sam Altman.

A few key points

Karpathy says large language models (like ChatGPT) don’t reason like humans. They only predict patterns from past data — meaning they can’t handle truly new ideas or situations they haven’t seen before. In short: they don’t generalize well. Animals and humans have built-in intelligence — for example, a zebra walks minutes after birth. AI, however, must learn everything from scratch through trial and error, which makes progress painfully slow and data-hungry.

The current training method (rewarding correct answers) is oversimplified. It doesn’t teach the AI to understand why something is right or wrong, just to chase rewards — what Karpathy called “sucking supervision through a straw.”

Even in tasks like writing code — one of AI’s best use cases — Karpathy says they mostly autocomplete or mimic common patterns. When you try something creative or unconventional, they fall apart.

If AI can’t generate truly new ideas, it also can’t automate AI research itself — which many boosters claim is key to reaching AGI. It might speed up testing or implementation, but not the creative breakthroughs.

Karpathy’s sober view contrasts with startup founders who have financial incentives to exaggerate AI’s abilities. Recent missteps, like OpenAI’s false claim that GPT-5 “solved” math problems it merely looked up, reinforce his point.