It could be a much bigger MoE model

Then it would be slower.

On modern hardware a cow page copy should only take 1-5ms. Redis forks to save the db to disk and it's been a solid design choice.

I guess it depends on how sensitive your application is to main thread pauses.

So like 1000-5000s if you have 4GB of data? Over an hour?

Redis absolutely suffers from long-executing fork() in practice, its developers even griped about it a couple of times on their blog.

I have found that design choice to be annoying

These LLM replies are really getting annoying.

Mine? I literally wrote what I wrote because “context window” as a term of art refers to the LLM’s context window.

I guess get better at detecting LLMs instead of accusing everything of being an LLM reply?

Maybe it has to be fine-tuned per tool just like functiongemma

what are you talking about

They are probably talking about claims like this: https://www.doge-impact.org/

Good artists get paid plenty. The hard part of art was always about either representing emotion, or a story. AI can't do that as it has no emotion, nor story.

> Good artists get paid plenty.

Absolutely false. You don’t even have to try hard to find scores of examples of great artists who died penniless. Artists who lived in squalor for decades until they broke through, and even then made only a modest living.

0.001% of which get “paid plenty” and that’s more due to business acumen.

Creative fields are cutthroat, brutal and soul crushing.

LLMs have proven that most people don’t care or can’t tell the difference between human expression and a probabilistic approximation of it.

On the other hand, this article is about a sports bar logo. Only in Santa Cruz would you see people lose their shit over something so trivial. Are we supposed to be moved to tears by the human expression of a cartoon otter on a tacky surfboard?

> how you do one thing is how you do everything.

That simply isn't true though. It's not even possible to be true. Will a neurosurgeon put as much time in their cooking/cleaning/etc as they do their surgeries? There's not enough time/energy.

that’s a pretty big oversimplification. it means that the way you do one thing is indicative of the type of person you are. if someone cheats on their wife, don’t trust them as a business partner. if someone puts in a lot of effort into a group project, you can probably trust them to take on responsibility outside of school as well. if someone always cuts corners on the “small stuff” like not tucking in their bedsheets all the way, not vaccuming under furniture, etc, they’re probably going to take shortcuts on other things as well. and if someone takes lazy shortcuts by generating mediocre ai slop art, they probably have a similar mentality to the food they make as well.

I wouldn’t trust anyone claiming they don’t take shortcuts, because it’s simply a lie.

They are profitable to opex costs, but not capex costs with the current depreciation schedules, though those are now edging higher than expected.

Amazingly, the current depreciation overestimates the retained value of GPUs.

In 2023, the depreciation schedule for H100s was 2 years, but they are still oversubscribed and generating signficant income.

Coreweve has upped their depreciation for GPUs to 6 years(!) now, which seems more realistic.

https://www.silicondata.com/blog/h100-rental-price-over-time

That's GPT4 thinking. New models use tools to look at current events or latest versions, and rely very little on weight knowledge.

You can pull new information into the context via RAG, but that is expensive and only gives very shallow understanding compared to retraining.

Mythos is basically this