cyanydeez

I'm pretty sure we're dooming because capitalism; not because of any real or imagined capability of an llm.

for comparison, Qwen3.8-Flash-Next only requires 6B parameters for computation, but stores 125B, 51B of those can be comfortably offloaded as they're not actively used in decode but a single token look up.

The Quant iQ4 of this model loads, then, in ~60GB of vram, and on disk it's 85GB.

So if you could etch it, you'd need a ~25GB ssd chip and 60GB of vram.

The vram costs likely contributed to these things being out of reach of the current economic cycle.

negligence is the word you're looking for. US law has a lot of negligence crimes.

But they're not even talking about that, either.

it should also be said on every one of these but i bears repeating: owing a lot of people a lot of money or favors means you can be a criminal.

all they need is more lawyers than sense. To grow a business, thats typically what you need also!

I guess we're talking about p(doom) for humans; in that having stood by if not actively engaged in the currnet oil crisis and allowed several nation states to commit genocide and active war. These might not be direct causes but attention is finite, they're indirectly buying politicians to not solve these problems.

it never respected the ICC in the first place. When you examine a lot of America's behavior internationally, it's much the same way billionaires and oligarchs operate. This is mostly mask off.

America has decided they'd prefer to be a 2nd world country. Literally, as defined at the time.

I think some of it is they know they're so full of debt that they'll never survive intact.

Also, the assumption about AGI, singularity, etc is just a delusion of a cult leader. They might realize that more watts are not getting them more intelligence from LLMs.

It also is naive to think, if AI is shrinking head count, that companies will seek anything other than seniors.

2008 triggered a glut of cheap experienced workers that slowly reengaged juniors but this time that the entire ladder is going to be more valuable than any grad.

the skill issue is "having to use a cloud model to do work of any value"

might as well offer your life to a king to work in their fields.

like every economic bubble, once you've bought in, you either hype it up to make it worthwhile or you are delusional enough to think there's a next level.

With the current regime, there might be a next level, but not at all an enjoyable one.

it's not pornography, but art is a subjective thing, so you really can't argue that AI art is valid, because Art is subjective.

You can, however, argue that you like the art I guess.

yeah, but if whatever the guy was talking to is something other than a toy hobby, if it's the foundation of real business value, almost any moderately sized business would fight you about ownership.

having discussed this with people and saw similar discussion, it seems what they're doing is limiting long running context _and_ tweaking the models to run agents which basically strip mine your usage tokens because they want to avoid users taking up precious KV cache & VRAM space.

So whatever they advertise as the context window, assume the model has been tweaked to keep it a quarter. Other commentors think all models are generally useless past 250k, and that might just be a reality of thesemodels.

Eitherway: they're enshittifying precisely according to hardware vs users vs actual cash on hand, and smaller agents with less cache/vram on hand makes the overall hardware more performant.

This of course is ignoring whether they're trying to stealthly deploy quantized models to eek out even more space on the hardware.

This stuff is easy to learn when you play with local models.

AI, the new minority!

HR and the lawyers would tell you that it is all company property. Make any of it at work? Company property. Bring it into work to do company work, company property.

Whatever miracle maker you're talking about, if it exists, should be making you money so, it's either doing that, is a weird train hobby, or a delusion.

But none of that pays the bills. The context is entirely about how people get paid for this stuff and, that's turning it into a commodity, and saying that "my AI talks to their AI" is pretty close to saying "my job is redundant"

mmm, but software products _have been_ commoditized. Especially the bloated ones that are just a fluffy bunny attached to a CRUD.

We're watching in real time the conditions that created the Luddites.

Yes, sure, if you have the desire to be the AIcentipede iin WallE, sure.

you're assuming media gives a lick about doing more than soundbites. The actual affect will be at best, 2 weeks after this is implemented, but probably 2 months while they actually figure out what legal means they can to infer which packages they can safely claim are foreign, along with actual foreign pharmacies giving 2 shits about todays shipping.

I asked this in another thread: who benefits from all these bot integrations?

It's not the consumers of the AI, or the consumers of the real businesses? It just seems like everyone things "if they build it" it'll come.

But it's clear Amazon cares about how it's costomers have to view all the items as both upsells and advertising, and a bot is going to either be _really_ into it or act much like ublockorigin does.

But either way, no company making money on costumer direct sales sees a real benefit to allowing a bot to easily bypass all the necessary details.

Whether it's advertisement, liabilities, manufacturer's recommendations; a bot will ignore these things and generate a costumer expectation that suddenly the producer has to live up to.

Then you'll tell me: YEAH, that's why you need a MCP friendly website.

And I'll tell you: that's how the mafia works.

So of course if the FAA is going to put AI into the watchtowers, they're obviously going to use local AI and not rely on some random Amazon cloud infra in some middle eastern country.

right. of course, no ones going to cut corners on the safe use of AI and secure infrastructure in this administration.

Qwen3.8-Flash-Next seems pretty much auto pilot when I get it the right context.

Perhaps reverse the question: Are your build/construct requirements just really counter-productive to how LLMs need to understand things?

I've found constructing the code, writing the tests, adding the docs; then running through them gets most of the way there.

I've also found that making a simple obvious edit is a useless endevour when the LLM is primed for the long context tasks.

So, again, the question is reversed: are you over reliant on the LLM to do even stupid simple likes like editting a css variable?

you mean like a Shepards Tone (https://en.wikipedia.org/wiki/Shepard_tone); i wouldn't doubt they slowly tweak quants to try to eke out.

there's also probably load balancers that downgrade models during high use.

Qwen3.8-Flash-Next is pretty damn worth the extra ram you need.

effective communication is knowing who the audience is. Everyone naturally knows their audience to some extent, except for the "neuro-atypicals".

It is unsurprising that a LLM fails, without coaching, to effectively communicate.

Likely botched resellers claiming theyre RPI and the company getting angry customers.

Perfectly rational.

Guys, guys, tthe cook only turned the boil up a degree or two, ypu cant even feel it!

Hes of the mind that american fascism will hold together long enough to be competent decesion makers

I think you mean "pace the frontier"