2023-08-18 23:30 UTC
@lizzard@social.tchncs.de @nicol@social.coop @rosamundi@mastodon.world
number one example for me is the principal antagonist in the Paranoia TTRPG, Friend Computer. Friend Computer issues lethal and contradictory commands not out of malice but through incompetence.
There are the Greenfly Terraformers of Alastair Reynolds' Revelation Space series. Rogue terraforming machines that turn everything into lush green habitats.
Trying to think of examples that aren't malfunctioning in some way. The real life examples are working perfectly correctly in that they're making or saving money and externalising costs.
Replies (1)
-
@WashingtonIrving@mastodon.world 2023-08-19 04:17
@floppyplopper@todon.nl @lizzard@social.tchncs.de @nicol@social.coop @rosamundi@mastodon.world These examples are exactly, what made my experience with ChatGPT frustrating. Absolute lack of context. You have to specify everything to utmost detail, because the Thing doesn't use / know context. For us it's obvious not to lie, thus kill people, thus anger people, thus cause them to switch me off, etc. And this is caused by bad memory management, thus lack of "instinct". ChatGPT lied to me no problem. After pointing it out, lied some more.