Using LLMs is like driving a big-ass gas truck.
-
Using LLMs is like driving a big-ass gas truck.
No one *should* do it, but it can be useful, and you—as an individual—shouldn't *really* feel that bad for doing it under limited circumstances.
But please remember, using it feeds some of the worst companies in history, and *they* (and especially their decision-makers) need to be held responsible for their destructive practices. Every dollar in their pockets helps promote more late-stage capitalism and fascism.
Lastly, If you *do* use it, don't drive like an asshole.
---
Credentials: years spent working both for and against various AI companies.
*braces for backlash*
-
@alice 1/2 By far the biggest problem with LLMs is how they were presented as "knowledge bases" rather than what they actually are: data churners. A "knowledge base" ought to have a way to determine the quality of the data from which produces answers. It should be able to distinguish fact from fiction, truth from falsehood. LLMs can't do anything like that. They churn all the data in their possession, good, bad or bat shit insane, and regurgitate it in response to a prompt. .
-
@rozeboosje I thought the biggest problem with LLMs was that they're built on stolen work, harvested by mega-corps who gleefully gutted DEI and backed a fascist dictator, with the end goal of replacing skilled workers and selling expertise back to wage-slaves at a premium, while diverting community water sources to quench their planet-burning data centers.
But like, they make mistakes too.
-
@alice @rozeboosje to be fair, I think that most of those issues can be mitigated by using smaller models that have been trained on ethically obtained data using renewable energy, while the mistakes are intrinsic in the technology and will always be there
but there is also a lot of openwashing in the industry, and I wouldn't trust any claim of the training data from the providers of the models unless they can be independently proved, and honestly my interest in using the thing isn't enough to justify the work of verifying those claims, so I don't actually *know* whether those models actually exist.
