dachworker

Tenstorrent was designed with the expectation that the AI workloads would change in nature (different architectures). For LLM workloads stick to NVIDIA/AMD.

I just found out about them. To me it seems their core product is "boring" industrial robots with some AI fluff on top, which imho is smart because you can capaitalize on the hype without having it be pivotal to their entire business model. What I find interesting is the robot-as-a-service model.

Huh? I thought that was their entire selling point. That you can have a factory that needs retooling and instead of having to carefully plan and buy robots yourself, you can hire these guys instead.

Personal vehicles are atypical as a market in that a huge percentage of people buy used rather than new. So when you look at new vehicle price, you have to factor in that the top 20% (not sure about actual percentage) of people are buying new, then corporate entities are financing a huge segment, like maybe 40%, and the rest of us buy used vehicles downstream from them. €25k is barely above entry level in Europe, but it can only be afforded by most people because of this system.

I have recently installed Fedora Sway Spin and it's so nice. I did change the configs since I have been using i3/sway for a long time. But otherwise, the default setup is just very sensible.

I guess, the only thing it's missing is an AI agent component, but for now I don't know how to integrate one and trust it that it won't leak my personal data, so I prefer not to mess with that stuff.

IMHO robots should either be quarantined or they should have physical safety like for example with table saws where if it detects flesh it just halts. There's absolutely no way I would trust a robot based on a .md. That's lunacy.

Makes sense, no? Test time compute is something you can vary, so it makes sense that you start covertly reducing it once the model has already made it's splash.

I think in recent times we realized that culture is still very important and just doing realpolitik alone does not work. Turkey and North Africa are way too foreign and bring more problem than benefits.

Japan and Taiwan are culturally very foreign unlike Australia and Canada. If we add them we might as well add the whole globe.

I always thought I was a nerd, but I was never into super heroes or fantasy books/movies and I was never into playing Pokemon or Magic the Gathering. Star Wars didn't appeal to me and neither did Lord Of The Rings nor Narnia nor Harry Potter. I wasn't even that big into video games, although I did spend quite a lot of time playing some of the big titles (Vice City for example) but I spent way more time flying in MS Flight Simulator and FlightGear.

When I was little I was fascinated by big engines and in general driven by the curiosity of wanting to know how everything worked. I would disassemble every piece of electronics in the household. I was into model airplanes and gliders. I was obsessed with building electric motors. I was also really big into steam machines. Everything steam was just cool. Then computers took over my life. Today I find myself nerding out about wood working and back garden agriculture when I'm not doing AI stuff.

I still believe I qualify as a nerd.

This is actually at the core of our (EU) problem. Instead of thinking, "here is this new technology, how can we build new markets/products out of it", our industry leaders always think, "here is this new technology, how can we apply it to our existing industries".

You can see it a lot in AI, right now. AI4industry is not where the gains are right now. We should be competing on training foundational models. Once the foundational models from US/China are strong enough, they will do AI4industry and we wouldn't have any edge on them anyway because their tech is more vertically integrated and they have better insight in how AI works under the hood.

At the time it was a perfectly rational decision that followed the liberal foreign policy dogma of the time, which advocated that we tie up unstable countries with economic trade, so they become less incentivized to create havoc. Let's please not rewrite history.

That last graph is surely AI generated.

The abstraction is by necessity. Our puny brains have only a very small working memory. The only way we can reason about many problems is by creating multiple levels of hierarchy. That is actually the essence of what mathematics is.

It's an economic metric, not a political statement.

I find all this gender wars crap competely untrue in real life. I wouldn't be surprised if it was all due to sampling errors.

My parents bought their house cash. No they were not high earners and not even median income earners. Yes the house is now worth close to a million dollars.

Why should I not just port my kernel to Triton? What's the appeal of Scale?

Here's a thought: maybe liberal democracy was never very free.

Corporate taxes don't work and taxing the rich is super hard. Could this be a way for the state to suplement tax revenue?

For the kids, sure. Incidentally, if you post anything political you must now do it with the knowledge that your identity is tied to your comment or opinion.

I'm excited for the opportunities that an abundance of compute brings, even in the absense of AI applications. Big compute is the right direction, if we can afford it. There are always things to simulate or discover or automate in some way shape or form.

My grandparents were all born in Europe between the two world wars. Actually, despite humble origins they all had a fairly prosperous life. Even though I have much more education than they did, I don't think I can ever achieve the same level of prosperity as they had.

Like, I certainly cannot afford a family of 12 children. Nor can I afford to buy the amount of land that they acquired, and certainly not by working the same kind of jobs they did.

Maybe I'm weird, but does anyone else have worries about what future their prospective children would inherit? In particular things that worry me: 1. the growing geopolitcal turmoil which is likely to eventually descend into a great war of sorts, the footage coming out of Ukraine is horrifying, 2. climate change isn't going to be dealt with and again, lots of violence will ensure because of that, almost certainly, 3. not sure what to think about AGI, but I'm not entirely dismissive and at best it seems like a dual use technology, 4. a GATTACA-type future where the super rich figure out a way to birth super humans with perfect genetics and top 0.001% IQs. All of those make the future look so unappealing.

I'm super excited to give this one a spin. It seems like a neat idea, Triton, but simpler and with automatic autotuning. My head is spinning with options right now. I love how everyone was hyping up CUDA this and CUDA that a couple of years ago, and now CUDA is all but irrelevant. There's now so many different and opinionated takes on how you should write high performant accelerator cluster code. I love it.

It's also kinda of ironic that right now in 2025, we have all this diversity in tooling, but at the same time, the ML architecture space has collapsed entirely and everyone is just using transformers.

The bubble will burst when all this agentic AI slop fails to deliver. Might take a while yet, because generative models are very good at aping competence and it's therefore easy to produce compelling demos.

I love how they always style themselves experts on economics, as if there is one global economic policy that benefits everyone, equally. How cute.

Fair point, but here's my counter: consumers won't analyze the data but insurance companies will.