Post

Heroin is useful painkiller. Psilocybin shows promise for treating a range of mental health issues. Perhaps every profession should use them just so we can find the best use cases! Food and drug companies should put them in everything, we could make sure to have them in schools! -SATIRE & SARCASM

There’s a lot of people who haven’t revised their view from gpt2 when it was a cool tech demo. The latest models are shockingly useful. I’ve completed very complicated personal projects that would have been impossible to fit in alongside work commitments using them.

Just an obvious thing, if you had an all in on renewables, by your definition that’s more efficient than an AI strategy since the environmental costs are lower and unlocks growth in other industries It’s what Norway did after WW2 with state investment in hydro which seeded modern Norway’s wealth

that's precisely the criticism I admit

There's also competition. In another context I see a green fertiliser project being screwed over by data centres willing to pay more for previously surplus clean energy. I expect the project is now dead, so fertiliser made with steam-reformed H2 will continue to be imported, thanks data centres. 2/3

We absolutely *can* serve this level of growth at this speed through renewables, as the typical datacenter here in Nevada demonstrates. The motivations for picking gas over renewables are political, not economic.

A very slight reduction in your beef intake would do vastly more to help the environment than eliminating even all datacenter-hosted software (let alone AI specifically) from your life. That slight reduction will not kill you (hell, it might even do the opposite!).

You're lying.

If the looms had been owned and operated by unions and cooperatives it would have been far less destructive.

That particular view has been supported by multiple studies involving all of those aspects. We keep having to relitigate them because you AI ding-dongs literally ignore all that because you're literally addicted to the (unfounded! Incorrect! Wrong!) feeling of productivity you get from them.

What discussion do you want to have about AI?

Like how it's impossible to have a serious debate about John Wayne Gacey's clowning skills without people bringing up irrelevant things I don't care about.

The overhype is so pervasive that it's the industry itself that has made rational discussions almost impossible. We could start by agreeing that LLMs are not AI and have zero chance of becoming intelligence.

Yes, I can see it’s certainly harder for you because people will expect answers from you whereas I’m able just to post questions.

Why do you think running statistical models on large (stolen) datasets has 'transformative potential'? Can you describe what an LLM is and how you think it works? If not, that might be why you're getting pushback.

Because you just end up embroiled in "no it's shit, it's a plagiarism machine, etc". And the conversations I want to have include "how do we properly regulate this technology," and "how to we make sure we handle this technological change better than previous eras"

Thing is there’s no ROI for most use cases at actual cost and we’re at the end of Moore’s Law and there are inherent limitations in LLMs so we’re kinda stuck except for burning needless energy and capital. The best thing for the actual good use cases isn’t better discourse, it’s a bubble pop.

There's such a vibe of 'we have to meet MAGA halfway' about this. By the way, are they still generating CSAM on X? Because the only constructive conversation about that is 'let's burn it with fire' and anyone who thinks otherwise does not deserve respect.

Driving off “AI” fans is good, actually.

The spittle-flecked frothing-at-the-mouth cohort has definitely brought about a tremendous amount of self-censorship...

Same experience here. I don't want to talk that much about #LLM s here because of the knee-jerk reactions, even from people whose opinions I value and would like to follow in other topics (which is why I haven't started blocking and muting them). bsky.app/profile/jmko...

Will you be doing a more nuanced thread to get the ball rolling? Or maybe @mcuban.bsky.social

Is it fair to call them 'AI critics' rather than sceptics, because I think it was the sceptics who met the critics/obstructionists here who arent really interested (as with lots of SM) in a good faith discussion?

We had a pretty decent first 18 hours or so on this thread, which I'll count as a good run.

not my problem

How much of this is outsourcing blocklists to the most polarized of participants?

OK James, you put that data center in your backyard and then come talk to us OK seriously. Where is the nearest data center to where you live? How many more do you want in your area because if you’re not willing to support a data center in your backyard, you’re not supporting AI

You ought to go live near a data centre permanently.

Yup bsky.app/profile/mack...

As an ND person I find Copilot incredibly helpful as a starter for 10. I throw in my brain work shit & ask it to help sort it out. This might involve multiple (10’s) of iterations. So please don’t tell me I’m crap for finding this v helpful. I’ve heard I’m crap too much in my life already.

Sounds like you lost in the marketplace of ideas! Sucks to suck!

Hey man do you think we are also not bored and frustrated by having to constantly repeat ourselves all the time? Do you think that anyone who is harmed by this stuff actually wants to spend tons of their own time trying to convince people in your position? Because I assure you: it actually sucks!

People need to make a habit of openly disagreeing with people who share their conclusions but have arrived at them for the wrong reasons. I'd call myself an AI sceptic (I think it's not as good or useful as people believe) but also despair a lot of the misinformed critique.

There isn't just one thing, one category. None of it qualifies as AI. There are a number of Machine Learning Models and implementations, some of them, dealing in quantifiable outcomes, are very efficient. Those dealing with Language and Transform are great leaps but they don't have intelligence.

That shouldn’t be enough to force you to go to the Nazi bar though, right?

Name one.

It IS pretty much useless for the everyday person. AI is only good for a set number of things. Frankly I would LOVE to see every LLM shut down and an unequivocal ban on new ones being developed unless and until some hard legal and enforceable limitations are set to keep them behaving.