The Real Secret Behind AI Large Language Models
Why LLMs seem so impressive to us while actually revealing something extremely important about human intelligence
The last couple of years have seen LLMs come into their own, as ever-larger training sets running on ever-larger numbers of servers have enabled this form of artificial intelligence to produce seemingly impressive feats. Barely a day goes by without some sensational headline about how AI will soon reach or exceed “human levels of intelligence” and replace everyone from managers and Hollywood script writers to the most eminent scientists and engineers.
And yet, while it’s always amusing to watch a hype bubble expand dramatically before inevitably exploding thanks to the inexorable force of reality, AI LLMs do indeed tell us something extremely interesting about human intelligence.
Only, it’s not what the breathless hype suggests.
A few weeks ago I wrote an article about a horse called Clever Hans that was purportedly able to perform basic arithmetic. The reason so many people were fooled — including the horse’s owner, who didn’t realize he was giving off tiny cues that told the horse when to stop tapping its hoof — is because as a social primate species our brains are evolved to infer in others the…
