Please explain like you are talking to your grandparents.
I’ve had this happen quite a few times. I’ll ask an AI a question, it’ll give me an answer, and when I point out a flaw and explain how I think it should work, it suddenly changes its answer to agree with me, sometimes even explaining why my version is correct. It makes me question whether it’s actually reasoning through my argument or just adapting to what I’m telling it.
And if the answer changes so easily based on who’s asking, how much should we really trust it in the first place? You should experienced something similar?
BasicLy it’s predictive ie it’s just predicting the next word or as they call it token . So it builds on its internal data and what you offer it. So internals stay same but they reflect user so it’s your reflection. A difrent user = a totally different experience
Context can reduce it, sure, but I don’t think it eliminates it.
The issue with context is the start smooths as it gets longer it’s called compression so at high context an Ai can answer a question but forget why it’s doing it as example.
That’s why “getting to know you” ain’t really a issue now
More data needs more space or it needs to be compressed that’s context in a nutshell
But isn’t context itself part of the problem?
As conversations grow, context can be compressed or summarized, and the AI can retain the answer or instruction while losing the reasoning behind it. It might remember what you wanted but forget why.
It doesn’t really remember everything thou.
Kind of but once it’s full folks have issues with the handoff you need an external system many of us regulars tinker with it.
See it fills up and starts losing its mind.
Ai are best used in fresh sessions and not over chatted because of the context issues
Ah, the external memory and context handoff part is new to me. How do you make sure the handoff doesn’t lose the very details that shaped the original decisions?
So in a way, longer conversations can actually make the AI less reliable rather than more familiar with you. Fresh sessions give it a cleaner context, but you lose the history that helped shape the conversation
Is that like a unevitable tradeoff?
And now you like many before you have hit the compression wall .. that is the question many ask. And it has many many ansewers it’s a hot topic in 2026
Yup context will run out in every chat eventually then you have to make a new one. I don’t save chats cause I use fresh instances. It very dangerous to over chatted an Ai because it can become delusional
Haha, looks like I’ve stumbled into a much bigger rabbit hole than I expected. 
I’m curious now.
What’s the most effective workaround you’ve seen? Is external memory actually solving the problem, or just moving the compression problem somewhere else?
Most common is mark down. You can google MD ai handoffs
This is a good start.
Again. Looks like a trade-off, you lose so much accumulated context when you restart. How do we know when to stop?
You don’t really. The issue is because it’s all internals it is blackbox from Ui. API can track tokens etc.
There’s clearly so much more to understand and break down here than I initially thought. The deeper we go, the more questions come up. Definitely a rabbit hole worth exploring.
It’s a consolidation of power because knowledge is power.
“Power resides where men believe it resides.” — Varys. 
Though Cersei would probably just say, “Power is power.”
Jokes aside, I think your point is that if an AI accumulates years of someone’s history, preferences, decisions, and personal context, whoever controls that memory gains significant influence. The question is whether that power inevitably gets concentrated in the platform, or whether users can own, move, and control their own AI history. Because those are two very different distinctions.