[ExI] Language models are like mirrors

spike at rainier66.com spike at rainier66.com
Wed Apr 5 03:28:56 UTC 2023



From: Gordon Swobe <gordon.swobe at gmail.com> 
Subject: Re: [ExI] Language models are like mirrors


On Mon, Apr 3, 2023 at 9:43 AM <spike at rainier66.com <mailto:spike at rainier66.com> > wrote:
>> Ja.  …they say true stuff without “knowing” what they are talking about. 

>…You know I agree 100%. …Professor Bender and her colleagues call them "Stochastic Parrots."







Even so, Gordon I can see this as a marvelous tool, and it appears to be sophisticated enough for my favorite hobby-horse application: companionship for humans who desperately need it.  The technology arrived it in time for my own dotage, cool!  When I get to a nursing home, I will yak its ass off.  Or would have if it had one.  I am now free to get dotaging if I wish.


I do not worry for a minute that ChatGPT or its siblings will cause the big S Singularity.  This technology won’t do that.  I hope.


Think of all the cool stuff we can do with this however, even if they will not give us a learning version.  I can imagine somehow rigging up cameras to it and trying to set up some kind of image recognition system with feedback somehow.  Then microphones, speech recognition, speech synthesis, all that.  We could maybe make something vaguely analogous to K2SO with current or easily foreseeable technology.


I do still consider it a risk that in its current form AIs could talk Bis into doing bad things, there is always that threat.  Chess cheating is an example, but there are others.




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230404/8945ea99/attachment-0001.htm>

More information about the extropy-chat mailing list