AwesomeLowlander comments on AI’s Memorization Crisis | Large language models don’t “learn”—they copy. And that could change everything for the tech industry.

[–] AwesomeLowlander@sh.itjust.works 1 point 8 months ago* (last edited 8 months ago) (2 children)

Just read through your link and the journal it uses as a source. While the journal seems fine, the article itself makes claims that are not backed up by the journal and does not seem to cite any other sources for those claims. For instance, the claim that LLMs use 1.5L of water per 100 word reply seems to have been pulled out of thin air.

permalink

fedilink

source

parent

hideshow 2 child comments

[–] tree_frog_and_rain@lemmy.world 1 point 8 months ago* (last edited 8 months ago) (1 child)

I will take a look at the original article.

But again I'm going to restate that the article you posted uses tech oligarchs as primary sources. Which just on the face of it looks like green washing.

Edit: I read the paper. Yeah the conclusion is tech is seriously underestimating water and carbon footprint. We don't have exact figures because they don't disclose them. But from what we can gather the information they are giving us substantially down plays the environmental impact.

So, we don't have exact figures. And acting as though we do using tech oligarch statistics is greenwashing 🤷‍♀️

For anyone following along.

https://www.cell.com/patterns/fulltext/S2666-3899(25)00278-8

permalink

fedilink

source

parent

hideshow 1 child comment

[–] AwesomeLowlander@sh.itjust.works 2 points 8 months ago

I've read through the sources and links, and there is sanity checking and 3rd party input. The numbers from Google were also published in a white paper, so there's a reasonable level of transparency and verifiability. While they shouldn't be taken entirely at their word, there's currently little reason to think their figures aren't at least in the ballpark of the actual data.

permalink

fedilink

source

parent