Tade0

It's particularly load-bearing to a non-native English speaker such as me.

It looks a lot like English, but it doesn't have the same rhythm human English has. It's as if a human tried and failed to be concise.

The current Kim's dad didn't help here giving his son a name that's so similar to his own.

Better: how does he justify having a car at all, considering it's probably stationary 95% of the time?

MoE models can be largely offloaded to system RAM - recent versions of llama.cpp automatically distribute parts of the model, with all the dense parts going into VRAM and just the routed experts residing in system RAM.

I'm currently running Qwen3 Coder Next, which is a 80B A3B model, at 15tps on a 8GB 7700S and 64GB RAM.

It's an idiot (no Thinking), so it has to go over its work several times to get it right, but it's the largest idiot I can feasibly run at the moment.

Also by accident I'm undercutting the only provider of this model on OpenRouter by 40%, doing $0.48/mTok.

Of course at my token rate it would take an entire day to even reach a million tokens, but it's the principle that matters.

The voxel version of Złota 44 doesn't do it justice. I mean, just look at this thing:

https://en.wikipedia.org/wiki/Z%C5%82ota_44

Hi, outsourced labourer here.

We also got laid off en masse. Well, I wasn't (for now), but several of my friends were months ago already and I'm not seeing them coming back anytime soon.

Laying off people is such an easy thing to do that no matter the region of the world, it's being done.

> Cracktorio

I still can't determine if it was the long COVID or staying up until 5am playing this fucking thing that made my coworker act so erratically for months on end.

I woved not to touch that as I prefer to remain employed.

You will not find a business willing to take such risks, so the net effect will be no products at all.

The company I work for has an automatic review system as a first pass.

It became actually useful only with the advent of Opus 5 and its competitors. Nowadays it usually makes a few good points, but in instances where it doesn't, it's quite the stubborn asshole.

This one time I had to paste a function from a library because it couldn't get it to its digital head that its assuming the function has a certain implementation based on its name only.

> Nutella because they think brown = chocolate.

I have pulled such a prank on a person at least once, yes.

Ok, so this appears to be a surströmming type of situation - people (including me) trying to eat it straight from the jar/can and being surprised by the foul taste and smell, whereas to those who grew up with it it's like eating mayonnaise straight from the jar - not illegal, just weird.

I've reopened mine twice already, but each such instance increases the rate at which, for lack of a better expression, coagulates.

Was going to say the same. I acquired a taste for it, but you really have to spread it thinly.

Also the jar lid can and will get stuck on dried Vegemite, so at the same time you have to open it frequently, so as to prevent that.

I've heard that this year 70% of all production capacity went to HBM.

Just an hour ago I found a stick that isn't officially supported, but has all the right parameters, so 5600MHz, CL46, JEDEC compliance (of course), no XMP. We'll see if it's going to be an expensive mistake.

On that topic: the other day a friend of mine was complaining that his parents' laptop has locked-down UEFI, aside from other anti-consumer practices.

That laptop, with 16GB RAM, costed 16.6% less than the stick I just bought.

The RAM situation is beyond helpless. I was meaning to buy another 32GB stick for my Framework, but they're out of stock and so are any validated alternatives.

I've found just one store that carries them, but seeing as they reported "50+ units in stock" I called the store and learned that "the item is available through a third party vendor within four business days", but looking at that the ratings it would be more like three weeks.

Occasionally it takes me at most 10 minutes to figure out that a solution the LLM took an hour to churn out is not up to par and I need to throw it away entirely.

There are diminishing returns, but half of the rate you mentioned would already help me keep focus.

I'm sorry, but my Eastern European senses are tingling when I hear this.

Why return only after being poisoned and not earlier?

The same, yes.

It's strange to me, as it appears that in the US it apparently doesn't matter if any charges have already been filed. In my country a pardon can only be given to someone, who was already sentenced - there was even some controversy when a couple of years ago the president gave out a pardon before the legal process concluded. As a PhD in law he should have known this made no sense.

> Isn't that kinda bad for their health? Polonium intake is unhealthy.

I believe the current modus operandi is to offer a window.

On a serious note: plenty of people were surprised that Navalny returned to Russia. I suspect that after his poisoning it became clear Putin and his cronies wanted him dead, so he chose to give them the judicial spectacle they wanted so that they would at least spare his family.

I think you're in a position comparable to me trying to get people to stop throwing their tyres, furniture etc. in the community dumpster (you're supposed to bring them to a municipal recycling center instead).

The people doing it don't care that the HOA (so us) pays a fine for each item and that we were already warned that there will be a doubling or quadrupling of the waste disposal fees. I can ask politely and sometimes it works, but ultimately I can't tell them what to do.

I don't really like tofu and I don't share your values, so I think I'll continue eating chicken breast - just sourced from a place that meets certain standards, as usual.

I don't believe in individual action and regarding eating animals I draw the line at those which don't readily breed in captivity - octopuses for instance.

That is my feeling as well.

I expect a major revision after India's census concludes next year.

We really don't know how many people there are in the world - some countries (mainly in Africa) hadn't had a census in years.

The world population graph is misleading. Looking at it one could be lead to believe that in 2100 there will be 13bln people.

In reality by 2100 the world's population will be around 10.2bln and declining from 10.3bln in 2084 or so.

And that also assumes the UN won't revise its projections once again, as every time they did, they had to lower their estimates.

Key figure missing in this article is "by how much"?

https://www.bbc.com/future/article/20221214-what-is-the-lowe...

Compared to chicken breast it's about 2.87x less impactful, but the real offender is beef, which has 25x the footprint of tofu.

Overall we really just need to get rid of beef. On an annual basis the carbon footprint of the respective protein sources is:

-tofu - less than 500kg

-chicken - less than 1250kg

-beef - over 10 tonnes

Mix one part peas with two parts chicken and you're around a tonne - a reasonable amount without having to even touch tofu.

I am yet to see such a person rawdog it without any performance enhancing substance or "winding down" in unhealthy ways without becoming severely depressed in the process.

Have you tried increasing the repetition penalty? Needing to do this is already a red flag, but I'm curious if it helps at all.

Running LLMs locally on outdated hardware, I've come to the realisation that sub-10b models are broadly speaking kind of stupid. They can answer simple questions, but can't really solve problems.

The effect is so strong that with an 8GB GPU a 27b model can sometimes finish the task earlier than a 9b model, despite residing largely in RAM. If it's a MoE model, then it isn't even a competition.

How many tokens was that in the end?

I've been meaning to get one of these, as I was looking to make a liquid cooling dock, so I can't have an interface that adheres to any of the sides.

Does it really simply not stick and is held in place by pressure alone?

Release the transcripts from the prompting sessions you had and I think most people familiar enough with LLMs will be able to replicate your results.

> The moat of hardware effects processors has just vanished.

There was never one to begin with. With a sufficiently good quality generator and a digital oscilloscope one could reverse engineer any kind of DSP - including harmonizers.

The real moat is the same as in every business - customer service, durability, warranty - things like that.

If someone vibecoded a clone of Valve's Steam nothing would change, because the system itself is not the thing that makes their business so valuable.

I told it to create a 3D model in Three.js and it failed.

Same thing with Qwen3.6 35B A3B produced mangled garbage, but still, it didn't have errors in the code.

Exact same prompt in ChatGPT produces a half-decent model (ok, but exactly not the thing I asked) so long as animations are not involved.

I can attest to the fact that Bonsai indeed tends to loop and it can be something really simple, like my usual test:

> What are signals in Angular?

> How does that relate to RxJS?

This was enough to loop Bonsai 2.