@jsrailton
It really burns me that people are hyping and using ChatGPT and the like not understanding that "hallucinations" are not bugs, but that they only work by hallucinating: They are a model of 'what should the next word be to make something that "looks correct"' rather than "is correct", and no amount of patching is going to cause a hallucination machine to switch from "looks" to "is".
And those BEHIND the tools either know or should know this, but simply chose to ignore it.
@ncweaver @jsrailton ...frankly if people just read an article or two about how these models work , it's quite clear. For some applications (like suggesting an email or a letter) it's even necessary. Without wanting to belittle the risks...
@ErikJonker @ncweaver @jsrailton How they're being sold is important. They're being sold as magic truth machines.
@fuzzysteve @ErikJonker @ncweaver @jsrailton Where do you get those "Truth machine" ads? Everywhere I go I see "careful, this thing isn't factually correct" stickers.
@dpwiz @ErikJonker @ncweaver @jsrailton "This will confidently lie to you" isn't shown anywhere with Bing, for example. It's not about what people say. it's about what they imply when they rhapsodize about them.