A growing gap in how people see AI: casual users grumble, power users cheer

Two different realities

It has been reported that Andrej Karpathy drew attention to a widening split in public perception of AI. Casual users, the people who try free or older models in a browser for a minute or two, are pointing at obvious flaws — hallucinations, awkward answers, broken follow‑ups. Meanwhile, power users and researchers are marveling at what newer models can do: staggering gains in coding, math and other technical tasks. Two groups. Two very different experiences.

Why the gap matters

Why should anyone care? Because perception shapes policy, product choices and trust. If the headline experience is a flaky free bot, the public conversation will focus on failure modes and disappointment. If the technical community is running circles around those critics with models that actually solve hard problems, developers will push forward faster than regulators or everyday users can follow. The emotional core here is friction: excitement on one side, dismay on the other. Who’s living in the “real” AI world?

The hard part: communicating progress

This isn’t just turf war drama. It’s about communicating what’s new without sounding like snake oil. Better demos, clearer benchmarks and wider access to the advances would help — but that’s easier said than done when models are splintering into free legacy versions, premium closed systems, and rapidly iterating research forks. It has been reported that Karpathy’s post has revived that debate: how do we keep the conversation honest and useful for both casual and power users?

A gap like this can be bridged, but it requires effort from builders, press and platforms. Otherwise we’ll keep talking past each other — and miss the actual implications of the next big leap. Who wants that? Not many.

Sources: x.com