zacmps

This shouldn't be a problem, and the answer is to refresh devices less frequently.

Does it read the ntfs journal? Because that's really what makes everything, everything.

It makes me wish I used a journaling filesystem on Linux too.

I think you'd need 2 80GB A100's for unquantised.

I think rustdoc, rust-analyzer, and crater all deserve shout-outs as well.

The rust compiler does something similar by testing on all (ish) of crates.io, the public package repository.

How long until someone enrolls an LLM in a university course to get journal access.

Writing them primarily, for example there isn't a super simple replacement for something like `derivative` (https://docs.influxdata.com/flux/v0.x/stdlib/universe/deriva...) in timescaledb.

I've been using it recently and I will say it is definitely harder to perform common time series queries than something like InfluxDB.

I suspect they're off by at least an order of magnitude in that estimate, if not two (or more if GPU acceleration is possible).

I would say default to min(affinity, system threads, 16) and let the user input a specific number to override it.

I had a problem recently where a library started a process per CPU thread I suspect the author didn't intend it to start start 256...

Affinity is only easy to get on some platforms but you should use it instead of the system threads if available.

Thanks! That's useful to know. Unfortunately it looks like a lot of them are non-commercial which rules them out for me but there's a couple that look like they might work.

I don't know, maybe I'm desensitized by GPT4 but I'm running Llama2 70B and I certainly wouldn't call it 'disturbingly good'.

It was offensively bad and incorrect, and while I've heard it has improved since I think a lot of people just remember how it used to be and don't use it.

In fairness, that later one I would quite like to know. I've accidentally bought a game I already had access to at least once!

You say that like you or anyone else on the planet understands and trusts all the software they have running on their (phone/laptop/desktop).

That ship sailed a long time ago.

Not really, it's a whole field (model stealing).

> Humans will not voluntarily give up one of the nice things they currently have access to and enjoy

Plenty of people have, somewhat nullifying your comment.

70B at f16 needs about 120-160GB (4 A100 40GB or 2 A100 80GB).

Quantization still seems to have some issues for the 70B model.

> To you.

Which was the whole point of this chain :)

To contrast, I've not once used it where I would have previously used SO (professional developer).

I tested on a Zen 3 Epyc and wasn't able to get the POC to work, so I think it probably is just Zen 2.

It seems like it might be possible to get around this by letting the model emit moves like ` discard last 5` which would also let it keep a history of it's previous branches.

I doubt it to be honest, desktop GPUs use too much power (and hence produce too much heat) to be integrated in that fashion, and any kind of shared memory will be too high latency.

Terrible example, a large proportion of gamers (including myself and almost everyone I know with a PC) builds their own.

Upgrades are piecemeal and old components are variously sold, given to friends, or re-used in home servers.

What a mind-bogglingly stupid decision if so, I can just train a network on the identity function and get royalty free mickey mouse now?

But the original content is frequently recoverable.

You can't just take copyrighted code, base 64 it, sent it to someone, have them decode it, and claim there was no copyright violation.

From my (admittedly vague) understanding copyright law cares about the lineage of data, and I don't see how any reasonable interpretation could consider that the lineage doesn't pass through models.

IANAL

> supercomputers ever produce something meaningful

Absolutely, they contribute to research all of the time.

Some of them have pages where they list research outputs that they enabled (though this is of course limited to those authors tell them about!).

IMO there's absolutely something wrong with it, while millions of people don't have access to food or clean water.

Supply & demand is an explanation, not an excuse, and it is a problem that can be fixed (if there's political will).

> This is just another way for media organizations to distort the news

No, it's not. This is done because stories with images perform better, and obtaining images (& licenses) for photos of every event is not always possible.

Does that extend to internationalisation or accessibility?

(Real questions, Lazarus was before my time)