Today is the day. I left the office for the last time an hour ago. Time sheets filed, corporate laptop handed in, goodbyes said.
I'm now working full time on Marginalia Search.
marginalia.nu/log/83_full_ti…
If programmer speed and efficiency was truly such a significant competitive factor, we wouldn't be packing them like sardines them in noisy and stuffy open floor plan offices.
Did some local LLM-based labeling of the HN comment corpus, tasking the model to classify how likely it is that each comment is:
* Using AI-like grammar (e.g. it's not X, it's Y)
* Using AI-like markup (e.g. em-dashes)
* Trying to shill something
* Trying to influence public opinion
Still a bit work in progress, but here are some preview data.
Did some statistics, and it seems posts from newly registered accounts on HN are nearly 10x more likely to use EM-dashes, arrows, and similar typography than established accounts.
marginalia.nu/weird-ai-crap/…
I have a hunch, but cannot prove, that prediction markets are the driving force behind a lot of we disinformation online, since they essentially monetarily incentivize making people misjudge the state of the world.
There's been a huge uptick in this sort of brigade like behavior around current events. First noted it around LK99, that failed room temperature semiconductor in 2023, but it just keeps happening.
RFC 7231: The `Retry-After` header should contain "[...]value [that] is a non-negative decimal integer, representing time in seconds.
HTTP server implementers, for some reason: You know what would go well in `Retry-After`? Some frickin' decimal points!
So I've been cooking a bit with the ranking penalties.
Search result relevance in the default, unfiltered view should now be a lot better and much more human-centric. I don't want to promise SEO spam is completely gone, but it damn near is.
Pushed some bug fixes to the index ranking algorithms, that had the side-effects of making the search results a bit worse because fixing the bug made certain SEO spam more effective.
Have a fix in the pipe, but it won't be ready until tomorrow earliest, a few days maybe.