Do you hate all lessons from humanity's past or just the most important ones? If it takes work from a specific subset of the population and isn't compensated, then my friend, what you advocate for is slavery...
Ah yes, I forgot, Linus Torvalds and the thousands of others that built Linux over time are all slaves. Guess someone should probably go rescue them.
Torvalds hasn't coded in years and lives off the achievements of others. The amount of people that tonguebathe his asshole is astounding. He brings very little to the Linux project, and actually hinders it.
It's a little bold to assume that 15% of the US population means the entire US wants this. We founded this country against unwarranted government interference in our personal lives, it's why the fourth and fifth amendment exist.
You could say the same about China, where only 7% of people can vote. In the US, 72% of people can vote.
In the US over half of the population is said to be concerned that there's not enough legal protections to prevent the government from snooping on telecoms and internet data. Only roughly 14% of the US population want to go to the more extreme of having surveillance in every household. 74% of correspondents to a poll by the Center for Democracy and Technology were concerned about agencies handling and consolidating personal data.
None of them were compelled, and nobody is stopping you from running your own LLM generously provided by others. Doesn't mean when linux came out people nationalized Apple and Microsoft.
The risk I'm talking about isn't nationalization of companies, its corporate monopolization of frontier intelligence capabilities through capital consolidation and regulatory capture. "Just run your own LLM" ignores the asymmetry of frontier intelligence. You can build an operating system in your garage with just time and cheap hardware. You cannot go build GPT-5. And that's the problem with keeping it proprietary. I…
And everybody should have free access to the latest medicine, healthiest food, premiere education, fastest transportation, best compute, etc. When you're doing stuff on the bleeding edge, investing HEAVILY into R&D, there's no way to distribute the product to everyone without killing the rate of progress. Free models aren't too far behind, why do only the best, most risky ventures need to be kneecapped? Nobody is stopping you from building your own, better model. I can't help but to see this as you seeing someone else who made something better and want to take it for yourself while heavily discounting the enormous amounts of work and risk it took to make it.
See my comment to parent. I've been using local LLMs for practical, personal tasks for a few months now very successfuly. You can run fantastic local models if you have either: - M-series Apple device with ideally >= 24GB of VRAM - RTX [345]090 GPU I'm fortunate enough to have both and use an M-series laptop as basically a persistent server (I don't use it much and when traveling typically just use my work laptop). M…
What about RTX 3080? Too little VRAM?
you can run Qwen 3.6 35B-A3B (3 billion active parameters + some GB for context) that can easily fit into 10 GB of ram while the not currently active experts are offloaded to the cpu ram with llama-cpp.
I literally wrote about running quantized models and how much more affordable it could be in the very next sentence . Please don't reply if you can't be bothered to read the entire comment, it's not that long.
I read the comment, thanks. I just disagree with your cost estimate. Even for a small business that needs high throughput they could probably do it for far less than $300k if they aren’t just blindly buying the first big nvidia setup they can.
But your estimate was explicitly based on running the 4-bit version. The full uncompressed Deepseek R1 model is 1.4TB. You'd need 16 h100s for this, and considering the new price for an 8 node cluster is about $350k, I think my estimate of $300k was pretty generous, assuming you'd buy used or liquidation or something.