Reading Log
Articles, papers, and posts I've read, in reverse chronological order.
Managed with rlog ↗Teaching kids GenAI using Scratch, course is structured super well but I think we should move past Scratch
Useful dataset to get better OCR on handwritten math
Literally psychosis, cult-like faux-productivity hives, thousands of people burning millions BILLIONS of tokens on a race to the slop
Now we're talking
Looking into why so much bursty traffic to my blog comes from Ashburn, VA. Turns out its a huge data centre town.
Hmmm
Back to plumbing?
Lalit super good at clocking and writing ab big tech internals.
Useful advice... for free!
Great post, gamblification study can be extended to absolutely everything under the sun at this point
Still super relevant nearly 20 years later
Small window into FB culture, crazy to see how hard he had to work to get back into writing code and avoiding meetings, maybe without this move there would be no claude code
Long but great. a16z paved the way for engineering-focused invesment even if others don't realise.
Mostly generic, nothing here stands out as something that made OpenAI special, author kinda shit at writing, can't identify what people want to read
Another post that mentions importance of having good taste for SWEs
Agents can't build context, they can only be fed context... author describes three options: heavy prompting and guidance (2024/early-2025), semi-independent agents (2025-now), continual learning (future)
Claude being allowed to terminate conversations itself, user can't even respond! Alignment + safety measures, no paper though
Don't interpret models in isolation, look at system as a whole
How models refuse harmful requests
Foundational model alignment paper: helpfulness vs harmlessness, fine-tuning, RLHF
LLMs reward existing top tier swe practices, need to explore managerial side!
Guide on how to structure a technical pitch deck (ghostty)
Inference, memory, hardware, chips, economics
Super useful article on inference speed vs cost tradeoffs
Great read, brilliant story telling but I wish it got into the "what we missed by not banking smarter" details
A decent conclusion and overview of the AI space, can't help but feel so much time was/is wasted trying every new trend instead of sticking to a proven model/workflow
1. Better reasoning -> less constraints on model 2. NEED a good semantic/memory layer
Conflate larger PRs and LOC with velocity/productivity.. come on now Greptile I thought your PR agent was anti-slop?
Groq foundational paper 2022 - read this to see where every author is, most went to Nvidia and Google
Chamath hosting Groq executives, useful for article
Groq on agentic workflows, this is how I'm spending the holidays...
2022 Groq presenting LPU and TSP papers
Interesting paper by Perplexity on adoption and usage of AI agents, exposes the growing class divide
Good breakdown of model context failure
Google Search breakdown
Paper on how Shazam works
Good post but no concrete examples just predictions
An egghead gives surprisingly accurate market and cultural predictions in 2011
Anecdotal at best, high effort is important throughout life, especially during preparation, the swimmer and the runners trained excrutiangly hard to breeze through their races
Hey this is quite good!
All it took was 1 year at Google in 2000 to fund a lifetime of failures, lessons learned though
Part of the process to get this extension to the people!
Excellent pros and cons of raycast development, still holds up 3 years later.
Claude definitely opts for complexity given an open-ended optimisation question. However prompt diff.
Super useful, used it to create this very extension I am typing in right now!
Useful discussion between msgspec and Pydantic founders on performance