“Critics keep talking as if it’s useles, but it isn’t. It’s cheap and even though it has some flaws, is extraordinarily effective at some of its use cases.”

“So, what? Are we supposed to accept all the downsides of LLMs just because it’s occasionally useful?”

“LLMs? No I’m talking about asbestos.”

Tech punditry keeps harping on the notion that nobody has ever successfully banned “scientific progress”, but LLMs and generative models are not “progress”. They’re products and we ban those all the time

E.g.: epa.gov/newsreleases/biden-har

And: npr.org/2021/08/30/1031429212/

If a product does vastly more harm than good, we actually ban them. Even the dysfunctional capitalist societies that we are. The “nobody ever bans this shit” defence is thin because we actually do ban this shit all the time. It takes a while for the evidence to come in, but we do get there in the end.

LLMs? The research is clear. It’s actually much too unreliable, which means that it’s not that useful at scale—the utility is much less than the promoters claim. There’s clear precedent for just outright regulating it out of existence

But, yeah, we will have to wait for the bubble to pop first.

@baldur

Asbestos wasn't banned because it didn't work well. LLMS are fine and "mostly harmless". We're misusing them a bit because they're new.

@atzanteol @baldur I propose the hypothesis that LLMs will kill people with their misinformation if they haven't done so already

Which makes the comparison to asbestos favorable

To asbestos

@RandomDamage @baldur
We should regulate all speech then, LLM generated or otherwise. Human generated speech has caused far more death.

@atzanteol @baldur I could write a treatise on all the ways your post is wrong and confused, but I will settle for challenging you to document the first time a computer is jailed for fraud or medical malpractice

@RandomDamage @atzanteol @baldur a single cloud data center can evaporate a million liters of fresh water PER DAY [1] for cooling and humidity control. Data centers currently use 2% of U.S. electricity consumption [2], and that number is only growing.

So, to sum up, we want to
- induce yet more water stress on local aquifers (many are already in dire straits [3])
- displace an entire class of workers from their livelihood with little recourse

@zrb @RandomDamage @atzanteol Those data centers are used for far more than chatbots.

Also, the "evaporated water" literally goes right back into the water cycle, free of pollutants, and some data centers are experimenting with using salt- or grey-water.

The water usage is a nonsense stat, especially from folks who never complain about golf courses being built in deserts.

@LouisIngenthron @RandomDamage @atzanteol Yes, data centers are used for all kinds of things. However since the process of training and retraining LLMs not only takes vasts amounts of computational power but also data, demand on data centers will only grow as more work is handed to LLMs to decrease labour costs.

@LouisIngenthron @RandomDamage @atzanteol As for water usage; the recharge rate of an aquifer is finite.

If water is drawn from the resevoir faster than it can be replenished, the water pressure within the aquifer itself decreases and the weight of surrounding sediment compacts the available volume, leading to a permanent reduction in groundwater storage capacity for the aquifer.

nature.com/articles/s41467-023

Sign in to participate in the conversation
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.