@emilymbender the ending referring to how embedded plastic is and how hard it is to avoid being compared to LLM embeddedness made me squirm uncomfortably

Thanks for writing - great takes and imma remember these for the next convo I come across on it

@emilymbender oh, the "please absolve me" thing is SO REAL OMG.

I wrote about it in a different context for Library Journal a long time ago: https://dsalo.info/ego-non-te-absolvo/

@emilymbender I personally find the energy use argument a bit unconvincing. Sure, AI tools use energy. So does everything else: your laptop, your TV, the infrastructure for your internet connection, heating, transport, food... The additional power used by the LLM is very likely tiny in comparison.

Google just released a report claiming that a median sized Gemini query takes 0.24 Wh. It's not a lot. See https://www.technologyreview.com/2025/08/21/1122288/google-gemini-ai-energy/

Andy Masley also wrote about it well: https://andymasley.substack.com/p/individual-ai-use-is-not-bad-for?utm_source=bensbites&utm_medium=newsletter&utm_campaign=bb-digest-secrets-of-o1-revealed

@emilymbender I'm surprised that personal privacy didn't figure among your arguments, especially in the e-mail case. I would hesitate to use ChatGPT or other cloud-based AI tools (in particular ones that are free of charge, without any data security guarantees) to write e-mails and other private documents on my behalf. Who knows what happens with the prompts I send to the cloud? Maybe they'll collect them, use them to train their next models, publish them in a data set...

@emilymbender

I would be interested to know how the folks in the dealers’ room with an AI audiobook–reading platform were greeted, compared to the folks with the human narrators. The two booths were initially next to each other, but the AI people were moved later on.