I'm running out of time.
I have a year left before I will have to inevitably leave the UK. I don't want to return to Singapore so soon.
I need to start thinking about my relocation elsewhere. Maybe China or the US? Could consider a masters too...
from my experience using GPT-5, it performs unexpectedly well on some obscure and niche problems in places where o4-mini would have flopped. is a good lalamo
i think the only thing stopping me from getting claude code, gemini-cli or codex, and letting it take the wheel is
1. i am quite picky about the code it writes. llms usually aren't concise or elegant with the code it writes.
2. i still enjoy writing it
But for a hackathon?
This was my main text for my ml class. It's surprisingly more up to date with the ml landscape compared to many texts and is mathematically rigorous, but some of the topics there aren't necessary (VC dimension) and some of them should be expanded on more (adam, attention). The
i think people severely underestimate just how much of a dark forest the internet is. there are threat vectors lurking around with quiet resentment, waiting for you to mess up. be careful guys!
Looking back from the start of September last year, I've realised I'm starting to get a servicable working mental model of databases, computers, networking, and OS. It's not a 4 year CS course, but it's already paying huge dividend in how I think about the code I write
should I use CUDA through Rust bindings or should I try to be employable and use C (I am not using C++)? The tradeoff is that Rust doesn't have first class support for CUDA, so I will have to give that if I use Rust
if you wish to do math with the machine, you must speak to it. you must listen and understand it, before you can tell it what to do. understand however that at the end of the day, it is not your friend. it is a cold matmul calculator that brings your math abstraction into