Pre-trained models

On Sunday evening, we were driving to a relative’s place in Mahalakshmi Layout when I almost missed a turn. And then I was about to miss another turn and my wife said “how bad are you with directions? You don’t even know where to turn!”.

“Well, this is your area”, I told her (she grew up in Rajajinagar). “I had very little clue of this part of town till I married you, so it’s no surprise I don’t know how to go to your cousin’s place”.

“But they moved into this house like six months ago, and every time we’ve gone there together. So if I know the route, why can’t you”, she retorted.

This gave me a trigger to go off on a rant on pre-trained models, and I’m going to inflict that on you now.

For a long time, I didn’t understand what the big deal was on pre-trained machine learning models. “If it’s trained on some other data, how will it even work with my data”, I wondered. And then recently I started using GPT4 and other similar large language models. And I started reading blogposts on how with very little finetuning these models can do “gymnastics”.

Having grown up in North Bangalore, my wife has a “pretrained model” of that part of town in her head. This means she has sufficient domain knowledge, even if she doesn’t have any specific knowledge. Now, with a small amount of new specific information (the way to her cousins’s new house, for example), it is easy for her to fit in the specific information to her generic knowledge and get a clear idea on how to get there.

(PS: I’m not at all suggesting that my wife’s intelligence is artificial here)

On the other hand, my domain knowledge of North Bangalore is rather weak, despite having lived there for two years. For the longest time, Mallewaram was a Chakravyuha – I would know how to go there, but not how to get back. Given this lack of domain knowledge, the little information on the way to my wife’s cousin’s new house is not sufficient for me to find my way there.

It is similar with machines. LLMs and other pre-trained models have sufficient “generic domain knowledge” in lots of things, thanks to the large amounts of data they’ve been trained on. As a consequence, if you can train them on fairly small samples of specific data, they are able to generalise around this specific data and learn around them.

More pertinently, in real life, depending upon our “generic domain knowledge” of different domains, the amount of information that you and I will need to learn a certain amount about a certain domain can be very very different.

Everything is context-sensitive!

Up North

So it’s exactly a year since I moved to the north. North Bangalore that is. For exactly the last one year, I’ve called a place in Rajajinagar 2nd Stage (not to be confused with Rajajinagar 2nd Block) my home. It’s been a roller-coaster ride, I must say. Lots of ups, lots of downs. I must admit a year in I still haven’t settled in completely, and for the record, I’m already plotting a move back south.

The problem with the area where I live is that it is an area of houses built on 30 feet by 40 feet plots. While that itself is not a problem (with an FSI of 1.5, you can build a 1800 square foot house here, which is pretty massive), the problem is that this area was “built” in the early 1970s. Back then, nobody had cars, and given the economic situation nobody aspired to have a car. This, coupled with the smallish plots meant that people didn’t make a provision in their plots to park cars. Again – given the economic situation when this layout was built, it is a perfectly rational decision.

Now, forty years later, with rapid economic growth in the last 20 odd, everybody has a car. And since there’s no space to park them inside the houses, people park them on the streets. So on every street in my locality, you will find cars parked continuously on both sides of the road. Unfortunately people haven’t figured out a protocol of one-ways, so sometimes that leads to traffic deadlocks when cars approach from both sides of the road. Thankfully we are spared of such deadlocks since we live on a road that leads to a dead end (a real dead end, not a “Bangalore dead end”, which is just a T-junction).

Thinking about it, I find the land use here rather suboptimal. Most plots have two-storeyed houses, and given the size of the plots, the plots are fully built. So there is less than five feet space between my window and my neighbour’s – and on Sunday mornings we get woken up because Neighbour Uncle likes to talk loudly on the phone sitting in his porch (the neighbours reconstructed  their house a few years back, so they have a porch to park their car). Essentially there is little privacy.

There is little of other things also – like open spaces and trees, again a function of the plot sizes. I have a solution for this, but it is going to be hard to implement, and I don’t want to be the ones doing that. Yes, everyone has a plot of land to call their own. But they live in houses that aren’t too large and don’t have much privacy or open spaces or gardens.So I think it would be a profitable enterprise for a real estate developer to buy up the entire area (let’s say about two or three roads at least), merge the sites and build a high rise apartment here with all “facilities”. Even if the realtor were to pay above market rates (necessary since a lot of people wouldn’t want to sell) and compensate the incumbents handsomely in terms of houses, there is profit to be made. But then everyone wants their own “piece of land”.

Notwithstanding any of this, where I live is a great area. It is residential and solidly middle-class, and I can buy just about anything I want within a kilometer of my house. I only need to walk about 100m to catch an auto rickshaw, and it isn’t that hard to get an auto home at any time in the day. The biggest problem I find here, though, is infrastructure.

Malleswaram is a pre-car area. It was built in the 1890s to rehabilitate people from the old city (pete) area following an outbreak of plague. Consequently, the roads aren’t particularly wide. Which ever way I want to get out, I’m faced with narrow roads, and that combined with heavy traffic means travel times to the centre of town and beyond are large. And to make matters worse, Dr. Rajkumar Road (one boundary of “mainstream” Rajajinagar) is a highway, with buses leaving for all parts of Karnataka passing through this road.

I have a habit of living in cusps. My last residence was at the trijunction of Jayanagar, Basavanagudi and Banashankari. This one is on the border of Malleswaram and Rajajinagar. Maybe the next time I move, I should think of going into the “middle” of some area.