Hacker News Reader

> by its nature

Declarations about what LLMs can't do "by their nature" have a terrible shelf life, historically.

Unless LLMs drastically shift incredibly quickly, said declaration is likely to hold, because of how LLMs work. I mean, they've worked the same way at their core for several years now.

They predict what comes next and are very good at it. That's all intelligence is. What comes next in a pattern? What will my senses perceive next when I drop this ball? What is the next hypothesis to test after I made these measurements in the previous experiment? What is the best next move in this game after the current trajectory?

How can we say that's all intelligence is when we can't even define intelligence ourselves?

We can define intelligence pretty easily. It's the ability to solve problems. The ability to solve problems depends on accurately predicting what your actions will do.

You're probably confusing intelligence with consciousness, which has no useful definition and is irrelevant for intelligence.

Maybe. And while this may be a useful superpower, it's hard to predict whether it would be an enjoyable one.

Their nature hasn't changed but their capabilities have. What we thought was impossible due to their "nature" keeps shifting.

Reply on news.ycombinator.com