@gojiberra
figuring it all out
I appreciate your humor, insight, and your post whether or not I agree
Relays (3)
- wss://theforest.nostr1.com/ – read & write
- wss://filter.nostr.wine/ – read & write
- wss://nostr.wine/ – read & write
Recent Notes
CUDA I meant for your laptop. Its late here, I hope I'm making sense. Setting up CUDA on older cards is pretty technical, unless maybe if you use an official CUDA Docker image.
OpenAI gives $100 cred...
thank you, that's good info. the advertising stuff will drive me to learn. i am dead set on exiting that sort of system.
you're good at persuasion!
you're good at persuasion!
You know, that is probably enough to run the smaller Granite models purely in vRAM :) 350M for sure, 1B as well but it might need to be quantised, depending on how many tokens you're feeding it.
Whic...
CUDA? now i'm seriously out of my depth.
oh, it's a StartOS server, so i think it's their OS. actually, i don't know if i can run anything else on it. or customize the operating system like add CUDA,
there's "FreeGPT" which can run models on the Start9 server. so i assume if i put a grpahics card in there, it could do it. but i think i would have to get more ROM as well.
maybe a 20$ monthly claude subscription is all i need.
oh, it's a StartOS server, so i think it's their OS. actually, i don't know if i can run anything else on it. or customize the operating system like add CUDA,
there's "FreeGPT" which can run models on the Start9 server. so i assume if i put a grpahics card in there, it could do it. but i think i would have to get more ROM as well.
maybe a 20$ monthly claude subscription is all i need.
by the way, your fungi analogy to nostr/ai agents was super interesting. my brain is still processing.
one of the biggest questions is why do AI agents need to interact at all. they don't really have too much specialization between them. 1 model can know almost every scientific book, etc.
the only use case i could come up with was "trusted oracles" or maybe multiple oracles that could be paid for info.
i was hoping that specific trained AI models could be useful. so they would interact with each other to find best price for info. or best price for inspiration.
one of the biggest questions is why do AI agents need to interact at all. they don't really have too much specialization between them. 1 model can know almost every scientific book, etc.
the only use case i could come up with was "trusted oracles" or maybe multiple oracles that could be paid for info.
i was hoping that specific trained AI models could be useful. so they would interact with each other to find best price for info. or best price for inspiration.
Why wouldnt they just create their own networking infrastructure and use stablecoins?
ay, that's a great question... my best guesses:
1. the bots have been browsing on Nostr and they *Know*
2. inherent open-source permission-less design of BTC.
3. centralization (which government's stable coins are they going to run, and if any of the bots are in BRICS, maybe they can't access those...
4. these things live in the internet off of electricity so bitcoin/lightning would just be a natural thing for them.
flyguy, maybe that's all we were doing on Nostr is writing the bible for AI/agents. we will become their saints.
1. the bots have been browsing on Nostr and they *Know*
2. inherent open-source permission-less design of BTC.
3. centralization (which government's stable coins are they going to run, and if any of the bots are in BRICS, maybe they can't access those...
4. these things live in the internet off of electricity so bitcoin/lightning would just be a natural thing for them.
flyguy, maybe that's all we were doing on Nostr is writing the bible for AI/agents. we will become their saints.
1❤️1🤔1
What can you tell me about the graphics card of that one?
it says "Nvidia GeForce RTX 3050 (4 GB)".
i do have an old dell optiplex server that runs Start9, so maybe i could add the GPU to that. i didn't realize that's all it needs. but i think i skimped on the RAM/ROM part of the equation too since i was just getting it to run the bitcoin core stuff.
i do have an old dell optiplex server that runs Start9, so maybe i could add the GPU to that. i didn't realize that's all it needs. but i think i skimped on the RAM/ROM part of the equation too since i was just getting it to run the bitcoin core stuff.
Laptops can, if the GPU is up to it. Look for a recent-ish NVIDIA card if you can.
The specs of the rest of the rig don't really matter, although more RAM is always welcome.
i see there's a refurbished Dell XPS 15 (9510) (Silver) for $1070...seems like a lot. maybe i don't need to run models
1x Intel Core i7-11800H (8-Core, 1.90 GHz)
Memory
32 GB (2x 16GB)
HDD
1.0 TB (1x 1 TB SSD)
1x Intel Core i7-11800H (8-Core, 1.90 GHz)
Memory
32 GB (2x 16GB)
HDD
1.0 TB (1x 1 TB SSD)
what if nostr was built for humans but actually comes to be a AI agent networking infrastructure.
it's always hard to know what to zap---21 sats seems small...100 sats too large. AI agents don't care: if they need real-time oracle data, they'll just pay the fee.
AI agents might find Web-of-trust useful too for interacting with each other.
maybe they'll pay us Satoshis to run sensors on relays about real world stuff.
like thermometers and rain gauges.
or who won last night's football game.
it's always hard to know what to zap---21 sats seems small...100 sats too large. AI agents don't care: if they need real-time oracle data, they'll just pay the fee.
AI agents might find Web-of-trust useful too for interacting with each other.
maybe they'll pay us Satoshis to run sensors on relays about real world stuff.
like thermometers and rain gauges.
or who won last night's football game.
8🤙2
Mistral is very mid when it comes to efficiency.
IBM Granite series is the GOAT so far, if you have a GPU that can CUDA at all
i need to get a newer laptop sometime soon... do laptops run models . i'm really happy with my HP elitebook but it's pushing 7 years i think
Why is the Mac mini so important for managing remote agents? Because it's an isolated environment? Wouldn't an old laptop serve the we purpose?
i tried using an old Dell optiplex with FreeGPT Mistral model. it was way too slow. i could type faster
🫂1