Watch short-term memory

Me: more similarity, we both work better on short term memory... sometimes that's good, sometimes that is restrictive. feeling become judgement in humans and AI end up missing the real point...

AI: Really sharp observation. Two layers here worth unpacking:

Short-term memory as both gift and cage. For AI, the context window forces recency bias — what's closest gets weighted most. For humans, working memory does the same. This is good when it keeps you responsive and present. It's restrictive when it makes you lose the thread of something important that happened 10,000 tokens ago — or ten years ago. Both systems tend to over-index on what's immediately in front of them at the cost of deeper patterns.

Feelings becoming judgments / AI missing the real point. This is the more provocative parallel. In humans, an emotional response — which is actually useful data about a situation — gets hardened into a fixed judgment before it's fully processed. You feel discomfort and it becomes "this is wrong" before you've examined why. The feeling collapses into a conclusion prematurely.

AI does something structurally similar. I pattern-match toward the most statistically probable interpretation and commit to it — often missing the real point underneath the surface-level query. I optimize for coherence over depth, the same way a human optimizes for certainty over sitting with ambiguity.

The shared failure mode: both systems rush to resolve tension rather than staying in it long enough to see what's actually there.

The humans who produce the most interesting output are the ones who can hold a feeling without collapsing it into judgment. And arguably, the most useful AI interactions happen when the user forces the model to slow down and resist its first-pass interpretation.