tc4v

actually recycling steel is not 100% lossless in that everytime you decrease purity which leads to a point where the still can only be used in low grade products. building materials is usually one of the low tiers though.

You use case is exactly what hedy tackles, your experience is really similar to what the author of Hedy tells in conference and interviews.

good article, but I am very bother by the "standard basis"... it's called canonical in math. I don't think standard is the right name in any context.

This dev seems really inexperienced and has weird uninformed takes. The "functional vibe" but it's just using boolean logic, the "reference in a table" bugs that would happen in any language but C, the ignorance of type annotations and so much time spend on problems caused by said ignorance...

How to define "floating freely" is not easy to define. The potential energy surface of gravitation extends in 1/r wgich makes it a long range interaction. So if the earth attraction is small enough that you want to neglect it, you may still be trapped in the sun potential well. And if you can escape the sun attraction you are still prisoner of the galaxy. There is no 0 gravity, but there is various levels of neglectable gravity of course. But neglectable is always defined relative to whatever you want to measure.

It works better when there is some viable alternative. Research job market is terrible in Europe right now because our governments are trying to make a US like system (project driven and without stability) but without putting the money. There isn't a lot of space in the world to accomodate US brain drain. A bit in Japan, a bit in Europe, most of it in China maybe.

> Seems like a good opportunity for other countries to recruit scientists.

would be if any other country actually put money in research... Well there is China, but in Europe we already have more PhD than research position.

cache misses are slow because of latency, not because of throughput.

You have to distinguish the actual research from what is relayed by the article. These people are looking for proofs existence of a theoretical objects, the primordial black holes. Finding them would: 1. add proofs to the validity of the current model of the big bang and 2. help quantify the PBHs which we can then add with more certainty to the model. That's obviously very fundamental research, it's not going to lead to application any time soon, but general relativity was too in the 30' and yet we wouldn't have GPS without it.

sounds like https://gittup.org/tup/ to me.

I assume the driving of choosing low er level architectures over this sort of "smart" CPU was the overall complexity of the chip, but since the 80' with been piling so much abstraction and complexity that there is nithing "vanilla" in modern CPU. I wonder if such "higher level" CPUs could be interesting implented on top of the usual microcode architecture. They would for example give more information to the branch predictor, which could have a positive impact.

Boomer facebook is filled to the brim with bad AI generated stuff and people actually buying it, so... maybe you're sample is biased.

I know you don't have a lifetime access to institutional email adress, but using a fake address is so counterproductive. You're only going to claim the paper once, and yuh ou should do it while you have access to your email. Then you update your account eith a new address.

But it's a very difficult problem. Am open forum offers a platform to troll and misinformation. You could pretend that the community will be able to filter this out but I seriously doubt this project is equipt to fight against bots/fake accounts better than huge companies like twitter and facebook.

This seems unlikely because LLMs don't produce high quality code, they produce average code. So they don't contribute to a better dataset, they contribute to a narrower dataset around the average. LLM tend to self poison, not to self improve. There is a good chance it already started because of the huge amount of chatgpt code that was put on github since 2021. Maybe if the LLM authors use some quality filter to discard 80%of the dataset it can be avoided.

Everyone need to stop comparing Nim with Python. Nim is closer to pascal, and the only thing it shares with Python is indentation base syntax. It doesn't have the expressivity and flexibility of Python. It's a nice language, but there is no good reason to compare it to Python.

Obviously it's not about the 0.1% that are athletes, but about the 0.01% that are making big money out of it. And the one caring for those are the one governing us.

so a 5x speedup from using 8x threads? I am not convinced this couldn't be achieved with numpy+multiprocessing ( or even maybe threading sinc numpy releases the GIL I think)