Conversation
Notices
-
i'll be really scared when AI figures out natural language and consistently natural image generationi hope nueralblender memes don't become like the bad autocorrect and siri not understanding memes, that is quickly made obsolete
-
@georgia that's not going to happen until you have something near-human-equivalent
-
@georgia the problem with current AIs generating text/speech/images is filling them with meaningful content, because holding together a coherent story requires both broad and deep multi-domain knowledge of the world and how things fit together in it, and current AIs can't do that, even with massive datastores, because they don't build a rich world-model with powerful attention-focussing-agent (i.e. consciousness). gpt-3 or so does sort of ok when regurgitating / remixing short passages from material it's seen 1000 times before, but it can't do something like writing a coherent book, with focussed long-term memory to keep agents in the story stable and a world model to play with it in
-
@shmibs well humans speak language naturally and can tell that neural net generated images are abnormal looking. you don't think it would be possible for both to become as consistently reflective of what they emulate like voice recognition? for people not to be able to tell the difference between an AI's speech and a persons, or between an AI generated image and a real one?
-
@georgia sorry that's not coherent at all on my part; am distracted
-
@georgia not sure what you're saying, and maybe i'm stupid and confused and tired and stuff, but trying to get at was that ais now have basically brute-forced the low-hanging fruit for content generating and understanding, and what's left is long tails that require human-like, or at least human-comparable in some ways, AI to handle. machine translation for jp, for example, is always going to be bad until then
-
@shmibs i think you've misunderstood my post? individual human beings don't need to be proficient in multiple cants or jargons or what have you. an AI proficient in emulating natural language doesn't need to know everything. if what prints appears to be indistinguishable from human speech with high fidelity, that is the paradigm shift i'm referring to. just as humans generate language naturally and when faced with an unfamiliar subject do not return an error, so can a nueral net.
-
@georgia will happen in some contexts, but not under close scrutiny and not ever under other contextsif by "turing test" you mean broadly like "believe this content was human and not ai generated", pictures, videos, translations, whatever
-
@shmibs i'm just talking about passing the turing test consistently which unfortunately will happen
-
@georgia of course humans looking to fool people can always use an AI to help them generate things and then clean them up after though, fake videos etc XX
georgia
バツ子（痛いの痛いの飛んでけ;;
All senooken JP Social content and data are available under the