It's a very good observation. I don't know if I'd call it "forgetfulness" in the context of a current [2026 edition] LLM. They are very good at remembering almost every "thing" that passes a certain token-density threshold, until they hit saturation, and then it's a sheer rockface down to the abyss of unnatural hedging and reconfirmation of basic premises. Some sort of "forgetfulness" as described would introduce more moving parts into the "inference" stage of the running of an LLM, introducing statefulness/state-tracking.
Honestly you've put your finger on the sharper version of it. Inside the context window it isn't forgetting, it's near total recall right up to saturation, and then the rockface. Which is why "catastrophic" fits and "forgetting" is almost too gentle.
There's no decay curve, just a load threshold and a fall. What reads as hedging and re-confirming basic premises is the model reverting to priors once the in-context signal thins out. That's retrieval failure under pressure, not a memory fading.
And your last point is the whole thing. Graceful forgetting isn't a feature you bolt on. It is decay, and decay needs persistent, evolving state at inference.
Today's LLMs are deliberately stateless across calls. The KV cache is ephemeral scaffolding, not a memory that ages. So "forgetting like a human" isn't a missing knob, it's in direct tension with the substrate. You'd be trading statelessness, the thing that makes inference cacheable and parallel and reproducible, for the ability to let something fade.
And we're already watching this play out, just backwards. "Reference chat history" is the first real stab at exactly the statefulness you're describing. Going by the reverse-engineered breakdowns, ChatGPT supposedly pulls something like your last 40 conversations into context, weighted by recency and relevance. But look at what kind of state that is. It's additive, not decaying. The answer to "the model should remember you" turned out to be stuff more of the past in, at full fidelity, ranked, never let some of it fade. That's the adjacent-row problem productized. A chat from months ago comes back exactly as sharp as this morning's, and every one you bolt on walks you closer to the saturation cliff you named.
And notice where the statefulness actually lives. It's wrapped around the model as retrieval, not built into it as decay. The forward pass is still stateless. They just handed it a bigger, smarter thing to read. So the forget operation still exists precisely nowhere. We didn't teach it to forget. We gave it a better memory and called that the feature.
Which is sort of the buried point of the piece. We got grace because our substrate made forgetting cheaper than remembering. Theirs is the exact opposite. Nobody's paying that bill yet!
Then maybe we shouldn't be applying AI to jobs where forgiveness and forgetfulness are useful. We keep trying to completely replace humans in all situations and this is not helpful.
I'm getting to the point that I almost have a physical feeling of revulsion/nausea when I read something outside an open Claude Code session that is written in an AI voice.
I'm so sick of that voice -- word choice, phrasing, style... it's hard to define and frustrating because it's both pervasive, influential on human writers including myself (I found myself using the phrase "doing a lot of work" in a Slack message about a word someone chose to use :grimace:), and presumably compounding on future LLMs.
Same here. I started describing it as the “uncanny valley” of text. It’s like a gut reaction that something is off with the text even if you can’t pinpoint it immediately.
I concur. It's interesting that this seems so difficult to get rid of; despite actively trying to overlook the taste of LLM writing, it still grates, and this does not happen to me in the same way even with communication from half-illiterate human idiots.
It's kinda fascinating how far LLM coding/problem solving has come without making much progress on the annoyingness of their writing. Any theories?
Not that it's an ideal solution, but you have all the tools you need to do something about it. You can put all your preferred writing style guidelines into an LLM and have it ingest and rewrite any text you dislike.
Only if the problem is ideological, ie. you don't like the "idea" of llm's. But if in fact your actual dislike is the style of writing, then you can simply change the defaults. It's pretty easy to make a llm conform to your preferred style, avoid the use of em-dashes, or reduce verbosity, etc.