I find this annoying for what I assume are obvious reasons. Recently I had to request that a teammate of mine redo 100+ storyboards bc they used MJ to generate them and they were unusable as comp guides bc they did not convey the info we needed. This wasted our time and put us behind schedule.

LLMs have plenty of actual use cases, but some people seem to be so excited about them that they assume they are equally well suited everything, which is just not in line with current reality and costs people time and money. I’m all for using it when the final product doesn’t suffer as a result.

A corollary to this is, at least in my work, I refuse to make quality concessions for machines. In order for the work to be present in a final product, it must be as good or better than the work that a talented human could produce. Many genai platforms do not measure up to this in pipeline work.

Just commenting so I can easily find this video later

But girl the shareholders will pay MONEY to be told we're going to live in the matrix.

But are any of these companies profitable? Also no, but....

Exactly. Which makes me extremely suspicious of what their actual endgame is.

Honestly surprised at this level of cringe from Hank, tbh.

Not normally a super credulous guy and yet here we are

“I agree with you for the most part, but there may in fact be some uses of a technology that we are all reflexively hateful of because the world’s worst people are pouring billions of dollars and our rainforests into a wood chipper” is not credulous, it’s the opposite. It’s willingness to see nuance

Note: nobody says that just because LLMs *have* uses that they are *worth* the cost Once we can agree there is a ledger on both sides it’ll be a lot easier to agree on how that measures up (SUPER BAD!)

this exactly this - there are uses but there are also other solutions that aren't so fucking bad for everyone.

I kinda disagree here, because almost any time I've allowed someone an inch on this, they try to twist it into an argument that LLMs are everything they claimed. We're mostly talking to highly credulous loyalists at this point, it's not a very rational conversation.

If you deny reality in order to have a “rational conversation” there is a bit of a disconnect there though, no? I think the “give a mouse a cookie” logic, while understandable, is no way to handle truth or reality

I'm not sure I follow.... I said it wasn't a very rational conversation, and I dont usually try to have it.

Ahhh I thought that was in the sense that implicitly a rational conversation would be a nice thing to have Never mind then

To everyone tone policing: there is no future where GenAI both improves in such a way to actually become useful while not also increasing global warming at an astonishing pace and making millions of ppl unemployable in the process. You're living in a fantasy world if you think otherwise.

That’s a total non sequitur

when it does work correctly, it can be downright dangerous: its basically a powerful database search function that can assemble retrieved information into a useful format for a specific goal. simply, you could ask it to give you the recipe for nerve gas and, provided it has the data sets...

...it could assemble that information into an instruction set and list of necessary ingredients faster than you could do it by yourself. it cant invent new kinds of nerve gas, it couldn't determine how to deploy it, but sure as hell it could put the power to make it in the hands of nearly anybody.

Can you elaborate on the environmental aspect? Like what percentage/quantity of emissions/energy/resources are we talking?

I dont have a complete picture, but just from first principles: mining for rare earth minerals and disposing of electronic waste can both cause environmental damage, water usage for production and cooling goes up and there's also the obvious energy demand.

here's an article from MIT which goes into more detail:

generally speaking, all high level computing and communication tech consumes large amounts of electricity in use and rare earth metals for production. All industry of any sort tends to consume water. we can circularise the tech economy if we try, but the VC's and tech firms have no incentive to.

This seems more like an argument against data centers in general. I'm not convinced that AI represents all that large a fraction of total data/electricity. And the idea that it will inevitably grow to consume, you know, half of all world, electricity or whatever is just more VC hype.

Which isn't to say data center growth or AI electricity consumption is good, but it does seem like small potatoes compared to the big emissions levers around transportation, land use, agriculture, etc.

And short of something like a carbon tax I don't see a viable path to convincing people to become frugal with data. We could all stop watching 4K video switch back to 720P, but we won't.

Professional expensive spoof machines

It's ironic that Microsoft are the ones saying it kills your critical thinking skills... gizmodo.com/microsoft-st...

People who aren't writers but have to write emails or documentation for work tell me that they find it useful. But there's a huge asterisk there, which is that (I daresay relatedly!) people who used to send me ungrammatical but perfectly intelligible emails now send me literal gibberish.

I do know a fair few developers who use LLM stuff in day-to-day, to automate tedious tasks. Here there are *multiple* enormous asterisks though: First, you have to know your stuff in order to correct the LLM's many mistakes (so it's only useful to senior developers).

Second, lots of bosses see the tedious tasks getting automated and think "Wow, we can replace our junior developers!" And if the tool is only useful to senior developers, and you're not hiring junior developers, then in short order, where would senior developers even come from?

Third, the whole premise of LLMs automating tedious coding tasks relies on being able to call libraries. And the more the libraries are themselves filled with LLM slop, the less viable this strategy becomes. The whole field decays as a direct result of using this stuff, even if/when it's useful.

So in summary, even those use cases that are not precisely bunk are both destructive *and* self-destructive. Meaning that in the medium-to-long term, "they suck and aren't useful" is a pretty decent summary, actually. The water is held, .

The thing I find myself saying to AI-pilled coworkers lately is: "It's probably about as useful right now as it's ever going to be." Which sounds to them like a wildly pessimistic take. Which, I think, is telling.

they are over-hyped and harmful.

"Theft" is not an inherent trait of AI technology. Both because there's *tons* of art and writing in the public domain, and because it's fundamentally a legal & not a tech issue. Plagiarism, theft, etc of copyrighted works is an extremely solvable problem

...And yet its a problem AI companies are doing their damnedest to avoid "solving"

So is the problem with this thread that people are reading the question in two completely different ways? - Is the technology of LLMs sometimes useful? Vs - Is the way capitalism and tech bros currently abuse the shit out of that technology a net benefit for society? Very different questions imo