This is what it means to fight the war to the last Ukrainian.

This is further illustration that raw model capability isn’t really the most interesting thing at this point. The model and the harness should be viewed as a single system. I expect that we’ll be moving towards neurosymbolic systems where LLMs act as a heuristic engine within the boundaries of a symbolic logic system.

mainly cause they’ve been running a global empire and it didn’t matter where the inputs came from, everything is optimized for profit rather than resilience

Possibly, I haven’t looked at how easy it is to get your hands on one of those.

The thing is that you have different grades of oil, and US refineries aren’t actually geared towards the type of oil the US produces.

If you absolutely need one now, a mac is probably the cheapest way to run them because of the unified memory. With any x86 solution you have to get a separate video card with at least 32gb vram to run a decent local model. However, if you wait a bit then you can probably get a dedicated chip a lot cheaper in the near future https://wccftech.com/alibabas-tsmc-built-5nm-risc-v-chip-xuantie-c950-now-runs-qwen-3-8-27b-model-natively-unlocking-massive-vertical-integration-tailwinds/

I expect doing ASICs for models will work even if they keep improving. It’ll be like regular chips getting new versions. You buy a chip with a specific model etched into it, and if it does what you need great. Next year, a new version comes out. So, it’s actually a feature since it allows companies to keep selling new chips.

It does look like we are entering diminishing returns territory though. The biggest evidence for this is that Chinese companies have now basically caught up to Anthropic and OpenAI. If the progress at the frontier was still happening at the same rate, then the gap wouldn’t be closing so quickly. There’s also a lot less noticeable difference between stuff like Claude 4.6 and Claude 5. When they went from 3.x to 4.x it was very noticeable. And at least for agentic coding, most of the improvement seems to come from the harness now. I expect improvements will continue, but at a much more gradual pace. It’s also possible people will figure out a new architecture that’s superior to LLMs, or works with them. World models are one promising area already being explored.

пошел на хуй либерал

that’s little consolation for humans though

That’s kind of what I’m expecting going forward too. Local models will get good enough in a year or two for most tasks, and then you just have a specialized chip like the GPU to run them.

Alibaba just announced a chip specifically for running local models. We’ll see what it ends up going for. https://wccftech.com/alibabas-tsmc-built-5nm-risc-v-chip-xuantie-c950-now-runs-qwen-3-8-27b-model-natively-unlocking-massive-vertical-integration-tailwinds/

Training happens once per model, but inference is an ongoing process. So, there’s going to be a huge amount of energy saving if we move to using local models.

yeah, you go ahead and sign up for the Ukrainian foreign legion fash

I don’t know if you were aware, but you can just not write imbecilic comments on public forums.

This is another example of how capitalist incentives lead the whole system to ultimately undermine itself. Nobody cares about what happens in 10-20 years, the only thing that matters are quarterly profits. So, nobody wants to invest in training new experts. And this has been going on long before LLMs as well, it’s just that this tech makes the problem even more acute.

the piefed thing was pretty funny

Having used LLMs for development extensively, I’ve come to realize just how important it is to have deep knowledge of the problem domain to use them effectively. If you don’t understand which algorithms are correct, what data structures make sense to model the problem, etc., then you have no way to evaluate what the LLM spits out. And it seems like universities are doing a really poor job adjusting the curriculum to ensure students actually learn fundamentals, and are able to apply them. I recall even back when I was in the university there was already a trend where the focus was primarily on coding and getting people ready for the industry rather than the theory part. And theory is really the part you need to be learning.

Yeah, the physical price is what people actually have to pay per barrel of oil, and that’s been driving up all the input costs. And you can play games with the superstructure only so long when the material base is collapsing.

haha I keep forgetting how insane prices are now

How’s the output quality on the lowest quant?