Post B91qRx5jwj3tKRM2Xw by JamesWidman@mastodon.social
 (DIR) More posts by JamesWidman@mastodon.social
 (DIR) Post #B90MoW1PZ8PKFbDFse by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       the far more interesting "AI civil war" is between the monolithic model advocates (who largely are bought into that architecture because of capital pressure) and the modular intelligence advocates (who are bought into that architecture because it is actually sustainable and does not need 9 gazillion datacenters)
       
 (DIR) Post #B90N2d0yMYxxzcjlB2 by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @dysfun correct
       
 (DIR) Post #B90NCklmNvxYwMSUgS by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @dysfun though this debate has been ongoing since before LLMs were ever a major thing.  also LLMs literally started as a shitpost.
       
 (DIR) Post #B90NJTxvGERmqmLa0O by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @dysfun they were always shit, they are basically MegaHAL: neural network remix
       
 (DIR) Post #B90NPDnyG412T0Qev2 by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @dysfun the end result, perhaps, but how they actually work is more like MegaHAL
       
 (DIR) Post #B90RP2SGrBGv59RmnA by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       anyway, i am a fan of composable intelligence.  it makes a lot more sense than trying to build large monoliths that need ever more compute to work.
       
 (DIR) Post #B90RzOE1ORA31g7EW0 by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       in short: language as I/O, cognition as structured data.  just like we humans do it.
       
 (DIR) Post #B90SHviDcz226454bY by drq@mastodon.ml
       0 likes, 0 repeats
       
       @ariadne I've always wondered. Let's take a popular task, like coding assistanse.I have a gajillion ideas on my mind and a day is only 24 hours long, so I need an assistant.Say, I work primarily in Go. Or JS. Or C++, or whatever, the point is I have a specialization on one or two languages.Do I really need a several hundred gigabyte model that has EVERYTHING in it, and a whole GPU farm to run it? Can I have a little model that can do Go REALLY well, and that I can run on my 9050 xtx or less?
       
 (DIR) Post #B90SPzwLnBQXWgXQ36 by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @drq that's my pointyour ask becomes structured data, and then that structured data can be deterministically transformed into the target language
       
 (DIR) Post #B90VsrL59LJb6vYlW4 by josh@vickerson.me
       0 likes, 0 repeats
       
       @ariadne but how else will you grift entire industries into making you a gazillionaire?
       
 (DIR) Post #B90W55Ip3jtjtQKfEO by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @josh i don't really have any desire to be wealthy.  i am at a level of wealth that i think allows me to meet my life goals.
       
 (DIR) Post #B90lcMfV0tOmcXGZhw by the_wub@mastodon.social
       0 likes, 0 repeats
       
       @ariadne I just read that as "I am a fan of compostable intelligence".I should go to sleep.
       
 (DIR) Post #B91qRx5jwj3tKRM2Xw by JamesWidman@mastodon.social
       0 likes, 0 repeats
       
       @ariadne @drq this idea is intriguing to me, and i wish to subscribe to the newsletter