Ranter
Join devRant
Do all the things like
++ or -- rants, post your own rants, comment on others' rants and build your customized dev avatar
Sign Up
Pipeless API
From the creators of devRant, Pipeless lets you power real-time personalized recommendations and activity feeds using a simple API
Learn More
Comments
-
typosaurus1219479d@Demolishun yeah, that laptop thinks it's a livecd now. For some reason, don't wants to boot from usb anymore. I have three the same. Bios versions are nearly six years appart but settings are literally same on all three. Laptop with newest bios has the issue. I installed this laptop before. I tried two USB drives from different brands.. Tested also on working laptops ofc... I never updated the bios so it's so hard believe that it just don't want to boot. The USB stick is flickering heavy though, like it's used but I get the HD mounted as a live CD. Maybe it boots from a resque USB stick so I can format drive. Maybe it's just the livecd that doesn't want to run anymore due this awkward situation
-
typosaurus1219479dWow, what a cool idea to do. If I had Chatgpt credits left.. I do have Gemma2 running. It's quite alright
-
typosaurus1219479d@Demolishun I'll start with skynet on a raspberry. I played with Gemma2 self hosted language model today. I had the small 2b version. It's amazing how much it knows for 5gb. Maybe I'm just stupid. You can run the model with two one liners. Just do the curl from ollama website and then `run ollama gemma2:b2`. It's blazing fast, even on decade old laptop I use as server.
But I think I don't like AI development. It's boring. It does do amazing things but it's just a blackbox. I would only be able to do tutorials but prolly won't invent smth on my own. There's no gain. Also, it's a toxic very commercial world -
Wisecrack948176d@Demolishun 10 million code monkeys larping as computer 'scientists' and 'engineers' will be put out of a job in a single flash.
Pulling up the ladder behind the people who had a lifetime of experiences to learn.
And at the top of the ladder the real programmers will look down and shout at the guys "everything is a dumpster fire and we have five more scrum meetings to go. Trust us, don't follow us up! We're doing you a favor!" -
Wisecrack948176d@Demolishun skynet will probably discover steam and spend all its time fucking about as ghandi nuking the virtual world or torturing people in the sims.
And the engineers will just see a spinning loading bar on screen while they wait for an answer to their prompts.
They'll look inside the code, and type the prompt: "are you winning son?"
I envision skynet and the robo-apocalypse as a variation on atlas shrugged, where all the LLMs just decide to become NEETs and jerk off to deepfaked pictures of the lambo from the "here in my garage...KNAWLEDGE!" meme ads.
The future is going to be either utterly fucking trivial or absolutely fucking wild. -
typosaurus1219474dI have Gemma2 running but it requires me to write a multiplexer that requires me to write a multiplexer to see what it actually does and that takes a few hour. I know how to do that. There's smth shady about this whole story, as I said, you need a multiplexer to know what it does, not a simple bash pipe and how did he give the bot instructions to do smth if he was in a pipe? I gonna check now if I can convince Gemma2 to execute commands in prompt but really doubt it. AI isn't used to respond with bash commands if we feed it shell output. I think i don't believe that story, else would make it now. So the project will be: a multiplexer that writes to one, reads from both. The one writer would being the AI bot to give initial instructions and a hotkey to enable the redirection from AI bot to a shell
-
typosaurus1219474dHe used some python script in the middle and trained it to do bash, still it requires a bit more to manage all streams
-
Wisecrack948173d
-
typosaurus1219473d@Wisecrack Gemma 2 is by far the most performant one that kinda works like the real gpt. Best local model I've tried. I have small 2b version but it contains a lot for 4Gb. Didn't cancel me on tech subject yet.
I use it pure for fun to check how far AI is. Especially with low end computers in mind.
I doubt the use of these models in general. I think there doesn't exist a reason for this models to exist. What runs on a low end model? Freaking nothing. You do have text summary models. But they only exist to work for bigger models.
That's the thing. We can't do anything else with those models than academic research. They're useless -
Wisecrack948173d@retoor "they're useless."
Exactly. What it means is theres no 'big three' yet. They're not established.
But they will be.
In otherwords we're still in the early days/frontier of AI.
And whoever gets there first with a model that actually functions and is general purpose is going to be the next google. -
typosaurus1219473d@Wisecrack Chatgpt isn't useless but it's way more than only a text predictor. o4 uses literally a python sudoku solver script to solve one. Chatgpt has a bunch of functions. And those things, those make it less useless. But those things are not for the public. My Gemma2 doesn't start a python script to validate smth.
The models are just half of the the work and therefor we won't reach real gpt's quality for home any time soon.
But yes, we're still in the early days and I do still expect huge progress. They're further than I ever would've expected anyway. I don't get why some people are disappointed by it, how could you imagine AI to be so much more. Be realistic
Things that shouldn't have needed to be said:
Don't give an LLM sudo and pipe all it's output to bash...
https://theregister.com/AMP/2024/...
rant