I might have overcorrected
or not
“bro you’ve changed so much, what happened?” In the last few months, I’ve heard this from 3-4 different people I’ve known for so long. Is this good or bad? idk, mostly I’m practicing cultivating relationships - something that had been very difficult to do as a workaholic. now I might be possibly overcorrecting, but it’s evident that I’m making progress on this front and I’ll eventually reach some sort of equilibrium.
currently in a season of doing things out of my comfort zone. earlier this year I decided to take time off and just spend the rest of my time experimenting with random things and figuring out what I want to work on long term.
It’s a tough yet equally stupid decision to take. I don’t have any safety nets, never took handouts, I don’t have rich parents to fall back on. So when I do stuff like this, the consequences are usually that if something doesn’t work out I’m going to be homeless.
the only reason I have even the slightest confidence in myself is because the last time I quit my job at deloitte to go full-time into game development, in 3 months I got offered an exclusive contract with the largest publisher in the world. It’s a dangerous confidence booster when u go all in once and it works out.
anyway, now it feels like I’ve been doing way too much recreational/social stuff, and not enough “tangibly productive” stuff — which is okay. I’ve been working non-stop for the last 5 years, ima cut myself some slack for these couple of months while this phase lasts.
Life recently
I went out.. a lot.
I stayed over at zo house, web3 community house thingy - this place was fun
worked on snitchGPT, a vision API based web app that lets you take a picture of people violating traffic rules, automatically recognises what crime is being committed, identifies the license plates of the vehicle, and reports it to the bangalore traffic police.
absolutely annihilated people in FC24 and mortal combat on the ps5. played a bunch of board games - played catan for the first time.
It was also diwali!
In the diwali spirit, went to a bunch of festive parties.
a bollywood party with college buds I haven’t seen in a longgg time.
a crazy house party with mostly random people - played a few games among other interesting stuff. discovered this game called exploding kittens.
a game night - went out for bowling, laser tag and a couple of social events.
connected with some very interesting people, made some serendipitous discoveries, overall a very cheerful and social time lately. also been cooking up some stuff which I will talk about soon
random insights
crisis accelerates eventualities
when you're faced with a crisis, do you sit back and wait for things to magically sort themselves out? nah. that's not how it works, you're forced to confront the situation head-on and deal with it. and guess what? that often leads to outcomes unfolding at lightning speed.
almost funny how you can do things when you don't have a choice. humans are deeply flawed beings. we are not optimized for peak efficiency, but rather for the least effort. you'll work harder not to lose something than you will to gain something new.
Work Updates
currently I have paused the development of skive, and I am focusing more on machine learning inside of Unity. figured training AI to take part in a tournament would be very cool.
I created a sandbox of AI agents, and taught them how to press a button and eat food. ngl it’s a lot more challenging than I initially thought. reinforcement learning might not be the best approach for training such a complex AI for my game. for a simple task it takes a couple million steps to train the agent, and even just changing a few parameters negates months of work (in terms of training time)
I’ve also been digging into imitation learning as an alternative to this. through this approach, I can “record” my actions to train the agent. I just have to take manual control of the inputs for the AI, demonstrate it through heuristics, and then let the AI observe my actions, understand what I am doing, and learn how I am doing it. a much more efficient approach where the agents already have a reference point (my demonstration) for the goals and rewards. the issue is most of the time just a single demonstration isn’t enough for the AI to learn, so it does require me to do the same action in different ways a few hundred times to get it right.
the setup for just the ai/ml package on macOS was a nightmare, it took me over a week just to get python working with silicon and unity, given that there are basically zero insights on the M1’s compatibility with the ml package for unity. there are always 10+ open issues at all times with the core package, and when one thing breaks the entire setup has to be done all over again. to say it's frustrating is an understatement.
stuff that keeps me going
twitter drama
literally just browsing twitter for entertainment now
welp
It’s no big deal guys, it’s only the fastest growing most impactful AI company in existence, not like it influences human history or something
thanks for reading ig
Fun read overall