I have my own concerns about LLM and they have to do with the flattening of our lexicon and the possibility of manipulating via changing of attention and focus in weights within these black boxes.

It’s less possible now than it will be in the future, which is why I’m saying it needs to be addressed now.

Also that this activity is unspooling the linguistic encoder in humans and that speech therapist and teachers are already screaming about it.

No one is listening.

Also, all the software being "created" by vibecoding legitimately sucks, and as skilled SWEs concentrate in companies that don't use LLMs or at least require use, the shakiness of vibecoded shit is going to accelerate and eventually turn into critical bugs that nobody knows how to find, let alone fix.

All while the companies spend through the nose to pay models to piss out bad advice and worse code that won't help them out of the ditch they're digging.

So eventually, I wrote that item & pinned it. It is possible to ethically use LLM and that is the future, which is why I’m focused on it. Even if the oligarchs won’t.

Nah. Every SWE I know who works for any company that isn't FAANG has been sent explicit instructions to dramatically reduce/limit their use of tokens. Basically, they deskilled their entire workforce (except the smart ones that didn't listen or comply) and now they're trapped between the rentier techlords they wanted to become (oh, irony) and staff that have to be remediated from the damage LLM reliance wrought on their skillsets.

a) no, we are not getting close to that,
b) if we were, the energy story would likely be even worse, because although the power use is distributed it lacks economies of scale, and,
c) people only *say* this to justify their use of the big, harmful, hosted models, imagining that they will be able to replace this use with something less ethically compromised, any minute now.

Of course they will tell you that it’s not what I say. It’s the way I say it. But that’s just holistic, Neuronormative bias…. So yeah, another reason.

I still talk. 🪔

Some realities have this super inconvenient way of showing up at your house, burning it down, beating you to a pulp and dancing a round the fire.

This is one of those realities.

So even if they don't know it right now, in a year or two they will.

I think that's a big part of what's on display here.

I mean

The facts are in and the case is closed:

"AI" is a planet destroying pile of garbage that removes accountability from the powerful and deskills people while harming them cognitively. The models are built on lies and theft.

There *are* no "benefits" that can escape the damning gravity of its externalities.

Hank and Taylor and Cory Doctorow and all the other "open to AI" people are giving their credibility to some of the worst tech and people around today.

I don’t get why people like substack so much.

Did you see today's 404media story about a backlash to AI generated posters?

https://www.404media.co/we-are-living-in-a-chatgpt-flyer-pandemic/

From what I've observed--which isn't comprehensive--the problem he's fallen into, the trap deliberate set by AI marketing, is that "AI" isn't a thing, and so a ton of different things (LLMs, Stable Diffusion, a billion different machine learning systems) all get lumped into "AI", and tons of people can't tell the difference between them.

This lets AI boosters point at not-LLMs and go, "See? Look how useful it is, and it's not eating a Sun's worth of electricity!" anytime someone criticizes "AI" in general, without actually addressing the problems caused by LLMs and other resource-heavy, exponentially costly generative "AI" systems.

So, yeah, "AI" will survive the bubble because different "AI" technologies have been chugging along for decades under different (often better) names, and all that stuff will keep going even when the LLMs mostly crash and burn just like Sora.

Unfortunately, I think AI coding is here to stay (even though it's about to get super expensive), so I'll still need to find a new career because UUUUUGGGGGGHHHHHH I fuckin hate it.

I’m not trying to win a logical argument that it’s not actually artificial intelligence, I’m trying to claw back space marketing has tricked people into thinking how functional and miraculous this is.

no need to look this isn’t about some big horrible thing either said— it’s a lot of little things, assumptions, gullibility, boomer stuff really.

but then it starts to sound like a nice software update for spellcheck. i’m all for that.

@futurebird