#thinkaboutthepoorchatbots — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #thinkaboutthepoorchatbots, aggregated by home.social.
-
#SmallWeb and #SmallInternet are experiencing a renaissance. Making #SelfHosted, small and blindingly fast #web pages instead of platform-based lazy loading #JavaScript #CSS transition hellscapes has more benefits than just speed, #accessibility and search engine ranking.
There is a new kid on the block; #LLM #chatbots. They read the web much like #SearchEngines, #crawlers, #scrapers and #indexers do, and are easily foiled by programmatic web pages, interlaced #ads, interstitials, overlays and #paywalls.
The chatbots read the web like blind people do, through plain text. Make your site and your information accessible, and you will do great! Make the information of whatever you want to publish available in easy, clear tables instead of whatever that is that is in vogue now.
If you stop defaulting to hostile design, and consider accessibility, the information you want to publish will be available to chatbots when they advice their users.
Do you think chatbots will know about recent history and current events by reading the #news sites? Think again! If they even get through the paywalls, they are foiled by interlaced ads and other hostile design. Whatever we consider current events won't be read or learned by chatbots because we have made the information super highway into a huge pile of toxic waste where information goes to die in the complex tangle of copyrights, licenses and DRM.
The chatbots will not trawl social media either, although that would likely be a good source for data. Companies doing this sort of training won't want to take chances of personal data of non-celebrities ending up in the models.
I asked #LLaMA model for a text completion of a random thing. You know how the completion ended? "Download EBOOK ... Online free".