Even Sam Altman Thinks AGI Talk Has Gotten Silly

Sam Altman told Alex Heath on the on the Sources podcast that AGI is an "irrelevant marketing term." Interesting to see this coming from someone who has spent so long at the center of the AGI conversation.

Here's the thing, LLMs are genuinely amazing. I personally use them in web applications, workflows, etc. But they're still probabilistic models that have a bad habit of spitting out unpredictable outputs for even the most basic tasks, let alone multi-step agentic workflows. Don't get me wrong, LLMs are technologically advanced and there are signifigant use cases. But getting from what we have now to some emergent "AGI" self is an incredible leap that is still unclear to me.

Yes, its true that performance scales with size (strong superposition, if you to want get technical), and LLM models seem to be getting smarter with every new frontier model release. But we're not actually any closer to replicating human understanding, reasoning, or consciousness. Besides that, we still don't have a solid definition of what AGI actually is, how it is defined, or how to measure against it, which is kind of a serious problem.

This isn't to say that we won't get there eventually. Maybe LeCun's world models or maybe some form of recursive self-improvement (RSI) will get us there eventually. But to be brutally honest, that probably won't happen before the business side collapses and the third AI winter sets in. I feel as though tossing around AGI is more like jumping the shark just to raise another billion than a near-term reality.

I'm certainly no expert, but perhaps the industry would have been better off it concentrated more on thoughtful product rollouts, human positive marketing (read: don't scare or antagonize your customers), and temper expectations. Honestly, I think we all would have been better off calling this Search 2.0 and just let the actual value speak for itself.