This program doesn't do anything? It is just a bunch of markdown files and a wrapper for LLMs with shell access?
And you're telling me that developing "skills" for this piece of shit is just writing a markdown file with instructions on how to do a thing?
People are mass buying overpriced Mac Mini computers just to play with A COLLECTION OF MARKDOWN FILES?
AI bubble has reached peak insanity.
RE: https://minidisc.tokyo/notes/aj7ffz404l
Timeline
Post
Remote status
Context
6
@SuperDicq I set up a vm and used that and it worked great, people buying a mac for this are retarded
@SuperDicq saw some idiot the other day call openclaw "distributed computing" lol it offloads 100% of its work to openai
@sun@shitposter.world @SuperDicq@minidisc.tokyo Well I'm guessing you can install ollama on an Android/Termux + bypass OpenAI... can run the pieces wherever you like. So it can be "distributed" however it suits you.
@nom @SuperDicq it runs like shit with local models
@sun@shitposter.world @nom@mk.spook.social The only reason for that is because these AI companies are proprietary hellholes who do not share their code so we could run the same shit at home if we wanted to.
@SuperDicq @nom no the models that work the best are just too big to fit on consumer gpus
Replies
6
@sun@shitposter.world @SuperDicq@minidisc.tokyo @nom@mk.spook.social that's what makes strix halo and apple's unified memory on the mac studio so great.
@lolitechengineer @SuperDicq @nom I have seen people using local models but claude/chatgpt blows the doors off them
@sun@shitposter.world @SuperDicq@minidisc.tokyo Not true, really, for those of us in the US who can extrapolate/have some modicum amount of imagination.
Definitely true if you've become a slave to the big ai modules / are trapped in lockstep greater Asia.
@nom @SuperDicq every time I'm told local models are as good as chatgpt I try it and it is if you only test it for a minute. if you do anything complicated they're just inferior
@sun@shitposter.world @SuperDicq@minidisc.tokyo Yes indeedie I think we all know that openai or claude on a great GPU are just dandy... but the local models are "good enough" to build horizontal product already.
Also, many robotic AI tasks can be done at home with fallback to the cloud. I mean it's pretty obvious.
A lot of systems are going to REQUIRE fallback to local models so please think about that when you're designing AI products in the near term for use at home.
@nom @SuperDicq I am an enthusiastic supporter of local models. I have a 24GiB nvidia 4090 for it. but most people are just using an api key on openclaw afaict and the reddits are filled with people saying they went back to chatgpt