Recent submissions

Comment · 5 days ago
Comment · 16 days ago
Comment · 22 days ago
Static IPs completely destroy privacy when not being shared with many users. Our university has one outgoing static IP for hundreds of users :)
Comment · 1 month ago
"Most humans don't know about these features" - I agree with you there, but I think models should be able to do that. That's why, for the past year, I've been maintaining "changelog skills" that I feed into the models to fill in the gaps in their knowledge.

By the way, all modern models know which hospital the last pope was admitted to shortly before his death.

Comment · 1 month ago
Some whataboutism for those, who use AI to code: Modern model will NOT know about these nice new features.

I did some intensive testing with changelogs and came to the conclusion that knowledge is partitioned and comes with variable cutoff dates.

Common tools, like GIT, have their feature cutoff in like 2023 or even earlier. Enough for a model to work with the tool when there was no groundbreaking change.

Gossip has a fresher cutoff date. People ask models like an interactive Wikipedia.

Comment · 2 months ago
In my tests, I compiled the changelogs from various projects and checked how much functionality the model could recall over several test runs. In some cases, even with the latest models, the cutoff was around June 2024. Beyond that, it didn't recognize the features. However, it did know the cause of the last pope's death and even which hospital he was in (April 2025). However, the model did not know when Trump was sworn for the second time (Jan 2025) - but it did know THAT he was elected a second time (Nov 2024).
Comment · 2 months ago
From my own experiments with various models, I suspect that LLMs have distinct/partitioned cutoff dates; for example, historical literature doesn't change (Greek history, Shakespeare, Goethe), general knowledge (updated only in certain areas), technologies (updated regularly), software also remains surprisingly stable - for example, with GIT, a basic command set is sufficient to do 99% of the jobs - new features are unknown or unnecessary, and tabloid knowledge, which is always up-to-date (politics, Taylor Swift albums).
Comment · 4 months ago
Comment · 6 months ago
Hey LLM, invert those rules to get undetectable :)
Comment · 7 months ago
Comment · 7 months ago
Added to AGENTS.md :)

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton