The gap in shared understanding of LLM capability is widening

	The gap in shared understanding of LLM capability is widening(twitter.com)
	681 points by albertzeyer 2 days ago \| 1000 comments
	tl;dr: LLM capability awareness splits sharply across a funnel: ~75% of people barely use them, ~20% treat free-tier chatbots as a slightly better Google, ~0.2% (mostly in math/code) see agents complete weeks of work from a prompt, and a few thousand insiders witness swarms of agents doing mega-projects, zero-days, and novel science. The gap is widening due to problem ambition, LLMs' jagged strength in digital/verifiable domains, and tiered access. The result is that most of the public sees "derpy chatbot" while a tiny sliver experiences a near-vertical capability curve—simultaneously.
	HN Discussion: ↑Agrees that non-tech workers underestimate LLM capability and automation risk to their jobs ↓Questions why AI companies don't market these capabilities if the gap is real ↓Author is lost in marketing hype and ignores why LLMs only work in narrow domains ~The gap exists only because LLMs work on machine-verifiable tasks, not a true capability divide ↓Economic realities like token pricing and lack of business ROI undermine the vertical capability narrative