Apple already thought of an idea like that to charge their devices, but it turns out to generate a huge amount of heat so it doesn’t work. I would expect this to be even worse in cars. At most you could implement wireless charging at stoplights, but the goal of stoplights is not really to stop there for a long time. In any case, a train or tram with electrified tracks or overhead wires is superior in throughput when compared to EVs.

I just use systemd because it’s the default.

The police should indeed be called, since the price is at least partially due to cartel/monopoly behaviour.

Don’t worry too much, if Chinese models surpass Claude you can just distil it right back.

It’s more likely that these biases arose organically, both from biases present in the original training data and possibly also from the selection process of RLHF workers, as well as the biases of the workers themselves. If it’s not proactively targeted in RLHF you would expect that bias would be present. AI companies are more focused on getting the models to work at all than on eliminating biases.

Rlvr is done because you have to pay people for rlhf.

Imagine if they spent 100 billion making software that was actually good and that people actually wanted to use

I want one

Funniest shit I ever seen

They’re having a pretty hard time keeping Claude up even without releasing an even more inference intensive model. It would make sense that this model, if they allowed a lot of people to use it, would outstrip available capacity. They’re obviously trying to make more datacenters but these things take time to get going (and hopefully they’ll take even longer given the negative impacts on communities.)

Also one of the advantages of the Internet is that new software can be deployed very very fast. It wasn’t the case in USSR times.

Oh no not my wholesome hecking self help industry!

Logseq looks like a good option for Linux. I still prefer concepts since you get the infinite canvas and more drawing features, but I could probably barely get by with Logseq whiteboards for some topics.

Edit to clarify: Logseq has an infinite canvas implementation that you can draw on but things like the stroke nudge tool are very handy and unique to concepts. I also prefer tap with two fingers to undo which seems to be absent from Logseq. If I didn’t have Windows and Concepts I think Logseq would be good enough.

Edit edit: I tried to launch it again and my pen wouldn’t write even though I had the writing tool selected. Absolute garbage dogshit, and I want my money back.

obsidian and concepts are the ones i use, obsidian for reference notes and concepts for mindmapping and understanding on an infinite canvas. neither one is free and open source but obsidian can be used for no cost.

Oh no! Anyway,

That assumes the breakthrough can’t be copied by the others, which hasn’t been the case yet.

Of course not. Why would I pay for something that doesn’t work? The AI companies should be paying me to use it.

Yeah, it looks like this is a specialized card for inference only, which makes sense as the course can be much simpler that way. It looks like there’s another different model that’s specialized for training. It’s a smart approach that makes sense.

It says it’s an inference card. Does this mean that it’s only going to work for inference? Let’s see if there’s any more information in the listing.

literally yes-man from fallout new vegas