Technology Got Extremely Good at Understanding Language Before It Got Even Slightly Good at Understanding the Person Typing It
Recent AI progress in language understanding has been genuinely dramatic — parsing nuance, context, tone, and intent within a single piece of text with a level of sophistication that would have seemed implausible a decade earlier. This progress is real, and it's specifically progress in understanding language, which turned out to be a substantially different achievement than understanding the person producing that language.
A system can parse a sentence's meaning precisely while having no model at all of the recurring pattern behind why that particular person tends to write sentences like that, in situations like this, again and again — because sentence-level understanding operates on the content in front of it, while person-level understanding requires a continuous record across many pieces of content that content-level parsing, however sophisticated, was never built to accumulate.
This explains a specific and common experience: interacting with an AI that clearly, technically understands exactly what was said, while somehow still feeling like it doesn't understand anything real about the person saying it — because those are, in fact, two separate capabilities, and remarkable progress on the first says very little about progress on the second.
Understanding language arrived first because it was, comparatively, the more tractable problem — bounded, well-defined, testable against a single piece of text. Understanding the person typing it is a different and harder problem, sitting one level up, and its progress has lagged accordingly — not because it matters less, but because it was always going to take longer to solve.
Part of Explain Ori →
Momar Lissa Ndiaye ("MLN") is the Founder & CEO of weyoga Inc., a Delaware company. — weyoga.ai · mln@weyoga.ai