Weeknote #22 [W26.09] - My mental map of Bangalore, and AI hype ruining my week

Moving houses and my mental map of Bangalore

I had been planning to move out of Electronic City ever since I changed jobs in October.

My previous office was in Electronic City, so living nearby made sense. I suppose that’s a bias my parents instilled in childhood. Growing up, we moved cities every few years, but we’d always live close to school (often within walking distance). I think because of that, the idea of staying somewhere far from work feels unnatural to me.

Last week, I finally finalized a new place near Hongasandra metro. (I haven’t moved yet, that’s happening this weekend.)

But even before moving, I have been thinking about how the move is going to significantly shrink my mental map of Bangalore. And that thought reminded me of Kevin Lynch’s The Image of the City, which describes how everyone has their own mental maps of their cities in their heads.

My mental map of Bangalore has changed quite a lot over the years.

We construct mental models of the world which help us navigate it. So, naturally, our mental maps of cities expand and shrink depending on our routines and where we spend our time.

In 2023, JP Nagar was a big part of it. But that was replaced by Indiranagar by the end of 2024 when Dialogues Cafe closed down and TPCC screenings moved to Underline Center. HSR Layout only appeared on the map recently, after I changed jobs a few months ago.

The AI hype is ruining my week(s)

I spent all of last Sunday at work because they had the bright idea of organizing an AI hackathon. There are going to be more in the coming weeks.

Sundays are supposed to be for sleep, and the whole thing has well and truly wrecked both my mood and spirit for this week. All I can say — from the bottom of my heart — is fuck AI and all the hype around it.

To be clear, I actually do like LLMs. I use them fairly often to help me think through ideas. And parts of both this blog (comments section) and the Cinema Next Door website (TMDB API integration) were fully vibe coded.

But liking it is one thing, and believing the hype around it is another.

A lot of hype around LLMs is predicated on the hope that they’ll keep getting better and keep replacing more human functions. That hope rests on brittle foundations.

LLMs are fundamentally unintelligent. Earlier, I talked about my mental map of Bangalore. Having reference frames of the world makes us able to simulate real world scenarios in our minds. I can imagine all the traffic, the smoke, noise, heat and humidity I’ll face when making a journey from one point to another in Bangalore. All that is accounted for in any decision I make. LLMs do not, and never will, have that ability.

While our brains can simulate multiple possible scenarios through our internal models, LLMs operate by only predicting the most likely next token in a sequence based on patterns learned from large amounts of text. We can keep throwing more compute and data at them, and they’ll get better at guessing the next token. But that will never approximate anything close to actual human intelligence. It is a dead end.

The more complex problem you throw at them, the harder they are going to fail, and the more human oversight they will need.

Which is fine. As long as we temper our expectations, tools like these can still augment human thinking in very useful ways…

Just don’t fucking come for my weekends.