Reaching a plateau doesn't imply that it's not early. It's still entirely plausible that we come up with a newer better model in ten years that gives us true AGI or runs on cheaper hardware or just gives us a closer approximation to human reasoning.
You got some compute hidden in your couch? There's plenty of reason to think there's not enough compute to achieve this, and there's little reason to think compute improves intelligence linearly.
don't you follow the news? Amazon is bidding on nuclear power plants. We will just build more energy sources. We have way too much leeway to go.
Also optimizations are being built both in hardware and software. There is no foreseen barrier. Maybe data to do training but now the models have pivoted to test / inference compute and reinforcement learning and that seems to have no barrier except more compute and energy.
That's what stargate is, UAE princes building datacenters in France is, etc… it's all in the news. So far, it seems like a straight line to AGI
Maybe a barrier will appear but doesn't seem like it atm
Ignoring how long it takes to build a nuclear power plant, and how we have limited resources to do so, is there anything to suggest they're doing that because of AI compute? From what I understand, they just want to be carbon neutral which is difficult to do when your compute needs increase exponentially.
For people who have the disciplinary background in neural networks and machine learning I imagine that replicating that paper into some type of framework would be straight forward right? Or am I mistaken?
The model itself yes. The changes from previous architectures are often quite small code-wise. Quite often just adding/changing few lines in a torch model.
Things like tweaking all the hyperparameters to make the training process actually work may be more tricky though.
I've done this (not thoroughly by any means) with OpenSnitch on the Ubuntu machine I have ollama installed on running the 32b R1 weights. No network traffic.
I'm not entirely sure if it is possible to do some type of code execution like that in just the weights themselves, though someone else who knows a bit more about this can weigh in here.
I'm not sure what the point of this really is. Many of those printers and POS systems have OpenCUPS support already, which is more or less "one command". If it requires a Raspberry Pi I have no idea what you are selling. Can you please elaborate on this?
Valid question. Printercow is aiming for a plug-and-play scenario where you can use whatever thermal printer you want with an extensive html-like template engine.
I've got a bunch of thermal printers in my work lab that definitely don't have this but they all support ESC/POS. They're all currently produced kiosk style printers.
There are privacy and data integrity issues with GitHub for both enterprise and personal use. Gitlab and other self-hosted alternatives are nearly exactly the same. I don't know if "by harder for me as a user" they mean the web-facing frontend, or something else. And even then something like Gitlab offers the exact same feature set as Github does on the web frontend.
On a sidenote this is the same "but x is so much more convenient" mentality that is driving open source projects to lock all their documentation behind something like a Discord chatroom instead of having a proper docs page or wiki.
> once it supports sneakernet as a transport, it will be perfect.
Sneakernet transport has a wide array of definitions, but my favorite part about NomadNet and Sideband is the fact you can print paper QR code messages and have a recipient or LXMF propagation node scan them and have your message delivered. I have a $13 thermal printer I was able to get messages printed with.
