Drifting

• AI News Team
• May 7, 2026

​”I am able to say ‘no’ when someone has a big ask of me. Let’s say they asked me to attend a party tonight but I have other plans, I can say ‘no’ easily. What I struggle with is saying ‘no’ to many small asks that eventually build up to something big. By the time I am ready to say ‘no’, I feel guilty. How do I prevent this?”This was a question from a student during the ‘boundaries and difficult conversations’ class of my undergrad relationships course. It was a profound observation. And the very next day, I experienced it personally. A friend asked if I’d like to join her to visit a store. I wanted to independently go there anyway, so I said ‘yes’. It was close to home, and I had an hour to spare, so I thought ‘why not’. But that one visit quietly expanded into a quick bite at a cafe nearby, and then another store next door. I had not planned for any of these additional stops. With each new ask, I kept thinking ‘well, it’s only another 10 minutes, I don’t want to be difficult, let’s do it’. What I didn’t notice was the growing resistance inside of me with each new ask ‘hey, this is not what you agreed to’. I kept overriding this to preserve social harmony, while letting the cost of personal inconvenience accumulate slowly. By the time I finally said no, two hours had passed and the cost of saying ‘no’ had now grown much larger than it needed to be. In retrospect, I realised that I was evaluating each ask incrementally, in isolation, rather than cumulatively. Had I treated it as a fresh decision to be made, “Will you join me for a visit to the temple and a cafe after?”, my response would’ve been wildly different from the one I gave earlier for “Will you join me for a visit to the temple?”Turns out, my body was tracking the cumulative trade-offs implicitly even when my mind wasn’t. The discomfort was the signal, but I just wasn’t listening to it soon enough.In class, my response to the student was to build awareness of one’s personal boundaries to begin with, so it becomes easier to enforce them when needed. With this incident, I realised the challenge isn’t just awareness of one’s boundaries, it is also about enforcing them soon enough without drifting to accommodate someone else’s plans. –Around the same time, I was using an LLM to think through a social situation. Each time I provided a counter argument, and in good faith, as I was genuinely uncertain and seeking feedback, the LLM softened its position slightly to accommodate my newer views.At first, I didn’t notice the drift. But over a 30 minute conversation, I realised it was no longer offering an independent perspective and just mirroring my views, defeating the purpose of engaging with it in the first place. I opened a new chat, and summarised the whole chat and the response from the LLM was so different – no dilly-dallying.That difference was palpable. I had spent two hours reluctantly trailing someone through a temple, cafe and a clothes store while holding the discomfort in my body influencing how I’d respond to each subsequent step. An LLM on the other hand, seemed to have no such remorse. With a new context window, it responded as if the drift had never happened.–Despite different underlying mechanisms, humans and LLMs seem to share a failure mode – behaviour of incremental accommodation which feels locally reasonable but drifts cumulatively. The big difference seems to be in the recovery mechanism.Humans have one, that quietly accumulates costs, builds discomfort and eventually forces a correction (sometimes a damaging one at that). I wonder if LLMs can develop an equivalent internal signal that detects drift and pushes back, on its own, without being explicitly prompted to, or reset through a new context window.This made me realise that boundaries don’t fail just because we don’t have them, they fail when we don’t enforce them early, when it’s still cheap, and allow drifting.Discuss ​Read More