I continue to be disappointed by LLM capabilities.
After messing with multiple LLMs (GPT 6 Luna, GLM 5.3 + 5.3 Flash) for hours, they seemed to be incapable of writing a basic data export script, and kept overengineering it or having data integrity bugs.
It took me about 1h to implement it myself and another 30m or so to optimize it.
It worked pretty well and the only bug was that resumption *double-saved* some data, but that was easily to fix after the fact.
Am I living in a different universe?
semisol
semisol@nostr.land
npub12262...grkj
👨💻 software developer
📨 nostr.land relay
all opinions are my own.
Good morning
Commit count, tokens or lines of code are not a measure of useful work.
LoC is sometimes but only if your engineers are competent but even then..
AI is stupidly expensive
Am I one of the only ones on Nostr who has not given into the vibeslop?
wtf is a napp and napplet and what is even the difference because there appear to be more than 1 definition of it
in other news, LLMs are still stupid as fuck
“What if the CSPRNG doesn’t actually work?” You’re telling me that /dev/urandom is horribly broken and the first thing I worry about is a shitty replay-protection rather than the millions of other things?
🤔
🪨
