@mikarv@someone.elses.computer
Post #3082251
2026-06-03 21:18 UTC
wow, save energy by not saying please or thank you to your AI. the individualistic ‘turn off the lights’ moment for LLMs. models use tools and hidden thinking that expend thousands of tokens in seconds. two or three tokens of nicety isn’t going to move the needle. https://www.newscientist.com/article/2529017-ditch-the-niceties-in-ai-prompts-to-save-energy-use-say-researchers/?utm_campaign=RSS%7CNSNS&utm_source=NSNS&utm_medium=RSS&utm_content=technology
Replies (2)
-
@Dragon@toast.dragon2611.net 2026-06-03 21:53
@mikarv@someone.elses.computer probably more than 2-3 tokens as a lot of interfaces send the conversation history as context but yeh still naff all compared to the tokens they burn on their own
-
@j2bryson@mastodon.social 2026-06-03 21:59
@mikarv@someone.elses.computer I barely use genAI but to keep my hand in I use it for bibtex. I now have an elaborate prompt I paste in when it reboots for some reason, that ends something like “don’t do anything in response to this prompt, don’t try to generate a n old bibtex, save energy, just say “acknowledged”. Not only does it do that, it says “acknowledged” in other appropriate contexts too, which I only dream saves as much energy as it is retro STtoS.