Guys, AI is *literally* a two year old

AI is a 2-year-old.

That’s right, guys. AI is literally two years old.

ChatGPT came out in November 2022, but things didn’t truly get noticed at a society level until early 2024. Since then we’ve gone from “cute party trick” to “Dario Amodei is on Dwarkesh two days ago talking about how we’re at the end of the exponential and he expects a ‘country of geniuses in a data center’ in 1-3 years.” Meanwhile I’m reading daily hot takes about whether the AI economic transition will be “fast” or “slow,” whether “AGI” is even a coherent concept or some kind of Silicon Valley conspiracy theory, whether the displacement will be gradual or sudden.

How the hell can you figure out what sort of person someone will be at two years old?

Now imagine that two-year-old is GPT-5.3 and Opus 4.6 Extended and some even more galaxy-brained thing coming out next quarter. How can anyone be confident about anything with that kind of slope? We’ve already crossed from “line go up” to “line go up way faster and harder” for whatever definition of ‘line’, ‘up’, ‘faster’, and ‘harder’ you want to use. Is that not obvious?

If it’s not obvious, then the question we should be asking isn’t “will AI do X or not?” The question is to step up a meta-level and address a valid meta-accusation of an author/editor’s unconscious non-self-reflective bias:

What cognitive biases do humans possess that cause them to make such confident predictions about a phenomenon that’s two years old and accelerating?

Here’s my pet theory: contrarianism is hot.

It feels good to be different. To have an alternative opinion. To stand out. To go against the grain. To be meaningfully differentiated. That’s social value in Western individualist culture. The nail that sticks up gets… noticed. Gets engagement. Gets invited on podcasts to explain their hot take.

So people generate hot takes because our cultural and (probably) epigenetic history has converged on prioritizing individual exceptionalism and differentiation. We’re not operating in a context where falling in line with consensus is what grants status and going against it is social death. We’re operating in the opposite context.

This generates a market for contrarian positions regardless of their epistemic merit. Someone saying “actually it’s going to be slow and boring” now gets more engagement than someone saying “I dunno man, this thing is two years old and I’m genuinely uncertain.” Someone saying “AGI is a conspiracy theory” gets more engagement than “the concept is fuzzy but the underlying capability trajectory seems pretty clear.” That kind of “optimize for diversity of viewpoints” thing works in the classical regime when stuff is normal; but not in the relativistic regime when stuff is clearly not normal stop trying to go z-score +2 or −2 on your opinion to stand out, and just accept it’s not normal and *gestures wildly at everything* AHHH. We are now one score and six years into the present century, and in two of those 26 years we have leaped from something that doesn’t exist or is not in any use, to something that is materially helpful to many/most of the jobs, if not actively in the process of about to be doing them.

The fundamental issue isn’t “will line go up faster and harder, and is that bad for us?” The fundamental issue is that we have a systematically biased prediction-generation process that rewards being interestingly wrong over being boringly uncertain. It values value dispersion itself, not honesty and true calibration, which sometimes requires complete and absolute consensus, even if consensus is boring: yup, line is going way up way fast and that’s probably bad for us?

Maybe the epistemically humble position is just: guys, this is a toddler. We’re watching it grow at an unprecedented rate. Nobody knows what it’s going to be when it grows up. Come on. And anyone who’s confident in their predictions is probably getting more status from the posture of confidence itself than from being inherently correct.