Should we use LLMs?

I was just reading Friends Don't Let Friends Prompt, by Emily M. Bender and I found these arguments against using LLMs quite interesting:

• the "theft argument (even though they weren't using it when they would have otherwise paid someone, it's still based on stolen art)"
• "the labor exploitation argument (data workers enduring terrible conditions to prevent end users from seeing horrific output)"
• "the environmental argument (these things have an environmental cost)"
• the expansion argument: "every use of these systems builds the case for training the next one and building the next data center".

What do you think about these arguments?

I will interview Emily and Alex Hanna, the authors of The AI Con, on September 10th, 18:00 (UTC-3).