I tried rubberducking today to fix a bug. It has felt a bit weird to me before although I have done the same thing in some other form, like while pair programming and code review. This was the first time that I remember consciously practicing rubber ducking. And it worked very well. I didn’t even complete one paragraph and I figured out what was wrong. This was the whole thing that I wrote: ...
I am in the train right now, about half an hour to go until Bangalore. The interesting thing about this journey was that I did my work while sitting in the train. It was a side upper seat in a 2AC. I had wanted one of the larger seats in the bigger part of the compartment but the side upper has its own perks. The group-of-four seats (or group-of-six in 3AC coaches) put you snap in front of another person. If it’s a family (as my kind neighours were), there is very little privacy you have to yourself. The last time I was in such a group seat, I was surrounded by a family with luggage and kids. I interacted a bit with them but my awkwardness at small talk and the langauge barrier meant that it wasn’t too much fun. I did meet a cute girl across from me on the same trip and small talked a bit, but she seemed as awkward as me if not even more, and we spoke online for a bit after the journey. But it’s not every time that it happens and I like this quiet space. This train also has a light bulb and charging point dedicated to my seat! I was about to run out of battery twice on my phone and once on my laptop and I could just put it to charge. ...
So this happened on 15 May. I’m writing this post on the 17th. How did I know that it was 15th? I went all the way back in my email inbox to the first one and I got the Google’s “Let’s get you started” email on May 15 2022. I had taken a 2 week break soon after, as I remember, to find a flat in Pune. That was in the first two weeks of June. Now that I think about it, what was I doing taking two whole weeks off to find a flat, and in Pune. I haven’t done that for Bangalore and I get by. ...
A few days ago I came across Swaroop CH’s Twitter profile and started looking at his blog. It had entries from 20 years ago and more. I loved it. The communities that he was part of then are the communities that I am a part of now, and I could even recognize some of the names that he mentioned. His blog was split up into lots of small posts, similar to what people these days would use Twitter for. He had blogged about some of the early meetups that he went to, including the forming of BangPypers, early versions of Barcamp Bangalore, and other important career events for him. I loved it and thought to myself that I want to have something like this 20 years later too. I use Twitter a bit too much, but whatever I put there is eventually lost. I want to maintain this list as a slightly more sanitised collection of important career and personal events for me, that isn’t as smeared by rubbish as my Twitter timeline. I also worry about Twitter shutting down. haha. 2023 joke. kids, there used to be something called Twitter… ...
I got access to ChatGPT’s plugins features a couple of days ago. I tried using some of them. Diagram It felt very useful, where I could ask ChatGPT to explain some idea to me with diagrams. I also tried WebPilot which could go through links. I wasn’t too impressed by the other ones I tried, but it was intriguing to see in the list plugins for companies that could directly provide services from the chat interface (such as restaurant reservation, house brokering, fashion shopping, domain registration, etc.). I could see that it’s not such a bad UI for end users. They can type (speak?) in natural language and get the results they want without navigating through complex UIs. This reminds of Google Actions (and corresponding APIs in Alexa, Siri, …) where the interface for users is to speak something to take some action rather than navigate through UIs. ChatGPT is much better at interpreting what the user meant. I had implemented a Google Action back in 2018-2019, which was my first decent hobby project. It was a blindfold chess app and I had made some clever hacks in parsing the user’s text that allowed it to pretend that it was smart and could understand moves said in any one of the many different ways. But it was a challenge to implement it, and it wasn’t perfect. With things like ChatGPT, we can directly interpret what someone meant and encode it into stricter forms. I can see how it can be a very good interface for doing things smartly. ...
There is no true interpretation of anything; interpretation is a vehicle of human comprehension. The value of interpretation is in enabling others to fruitfully think about an idea. - Andreas Buja I’m quoting this from the book “The Elements of Statistical Learning.” This sounds fascinating, and seems prescient now that we are in the age of ChatGPT and LLMs, huge statistical models that interpret things for us.