Elektrine lite

← Feed

@flxtr@social.tchncs.de

2026-02-15 08:40 UTC

@Lazarou@mastodon.social I don't get how LLMs are needed for analytics? This field is basically about counting numbers, statistics magic and visualization. Isn't it? I mean, asking an LLM what kind of statistical sorcery would be needed to measure this and that and ask it for a small python script will probably work (if you double check everything). But what did they do, to come to a situation where an LLM makes up data for months? 😬

Replies (2)

  • @hashbangperl@hachyderm.io 2026-02-15 12:20

    @flxtr@social.tchncs.de @Lazarou@mastodon.social I've seen a ton of analysis reports done by agents and just trusted with little or no verification, looking for wrong database lock strategy, to reasons for errors in logs to data quality issues. It would not surprise me at all that someone pointed an agent at a database or something like crystal reports or even just a bunch of spreadsheets and asked it to summarise them based on a list of KPI, the results would genuinely be that bad without a bunch of hard work.

    Open ##2553898

  • @Kathmandu@stranger.social 2026-02-15 23:38

    @flxtr@social.tchncs.de People love simple interfaces. People have given up good coffee and good music in favor of lower-quality but single-button-push services. Asking the LLM is easier than doing the work, even routine work you know how to do. If the results aren't part of an immediate feedback loop, it could take a while for the results to stop *looking* plausible, or for reality to smack them on the forehead. @Lazarou@mastodon.social

    Open ##2553899