The 92 % stat looks really interesting! It’s rarely the spectacular crash that knocks a cluster over. Instead, the “harmless” retry leaks state until everything breaks at 2 a.m on one fateful Friday. Evidently, we should budget more engineering hours for mediocre, silent failures than for outright disasters. That’s where the bodies are buried.
Every polished template looks the same, but each handrolled site is weird in its own way. I’ll happily take wabi-sabi HTML for personal projects over yet another Tailwind landing page!
It spikes HIF-1α → EPO for a day or two, but meta-analyses still doesn't show a real performance bump, let alone safety at 8 km. Feels less like innovation and more like mountaineering’s own carbon-plate shoes, except the failure mode here is cerebral edema, not a slow marathon time.
LLMs are amazing at writing code and terrible at owning it.
Every line you accept without understanding is borrowed comprehension, which you’ll repay during maintenance with high interest. It feels like free velocity. But it's probably more like tech debt at ~40 % annual interest. As a tribe, we have to figure out how to use AI to automate typing and NOT thinking.
Interesting parallel, but as it is packaged today - "AI" looks much lighter on technical customization than quant.
What can you really meaningfully change on the large models today? a bit of RAG, some prompting - sure. But that sounds like a much lighter lift compared to the tasks done by large teams of top-tier CS grads in older quants firms.
I suspect a lot of HN readers aren't familiar with cricket. But as a fellow cricket fan, I really appreciate the opportunity to nerd about the gentleman's game here! :-)
Question: Did you get a chance to check if geography has any meaningful impact on how impactful the toss is? e.g. at some venues in the subcontinent where dew kicks in the evening, or the greentop venues in England where the first morning becomes almost unplayable in front of swinging pacers?
This is very cool! Given how messy and busy many websites have become, we really need a robust markdown converter that lets readers focus on reading the content. Nice to see something stepping up where Readability left off.
This makes me wonder - what gross margin might Cursor, and the similar AI IDEs, might be operating at right now?
Traditionally, most SaaS was designed to have 80%+ gross margin (cost after direct costs such as cloud hosting, support etc.). Surely cursor etc. would have much smaller margin with 500+ premium-model requests @20$?
From my experience of working with YT influencers in tech/ software development: many of their audience do prefer watching longer videos on YT.
They go to tiktok/ instagram/ YT shorts to binge-watch short videos. And come to YT either to have learn something from longform content (e.g. deeper tutorials), or find the interesting parts on their own, or just to have something longer to run in the background. Most Youtubers worth their salt would have fairly impressive engagement on their longer videos.
While I thoroughly enjoyed reading this piece of internet-rant, I've to argue that dB is still probably the best we have on this!
In RF engineering, expressing signal levels in dBm or gains in dB means you can add values instead of multiplying, which definitely appeared like a huge convenience for my college assignments! A filter with -3 dB loss and an amplifier with +20 dB gain? Just add. You can also use this short notation to represent a variety of things, such as power, gain, attenuation, SPL, etc.
I guess, engineers don’t use dB because they’re masochists (though many of them surely are). They use it because in the messy world of signals, it works. And because nobody knows anything that might work better!
Current AI coding tools don't seem to have any real impact on true development productivity/ efficiency yet.
Apparently you cannot just one-shot everything to production. who'd have thunk? :-)