nyantangle I’ve argued that the vocabulary that young people are exposed to these days is limited by IMEs, but no one listens. All books since the word processor have been bias by the dictionary in the IME, right?

nyantangle There were restrictions even before that, such as restrictions on character types and the quality/quantity of books an individual could access, but the emergence of things like conversion selection order and recommendations is qualitatively different since it was after word processing. I think it is qualitatively different because it was not until word processing that things like conversion selection order and recommendations appeared. “There is also the point that “vase used to hold flower offerings (often made of gilded copper)”, and I wonder if the fact that this used to be science fiction indicates a change in the quality of the restrictions?

nishio 20 years from now, “people born after ChatGPT” will shift to “language biased by LLM”

daijapan I very much agree with this point of view. So I am interested in oral tradition.

teramotodaiki “alignment”

  • nishio “Let’s say “alignment of homo sapiens with AI” because homo sapiens will emotionally rebel if you say “alignment of homo sapiens with AI”. Let’s call it “alignment of Homo sapiens and AI. Self-centered Homo sapiens will mistakenly believe that the AI side will step up to the plate.” --- Alignment of Homo sapiens and AI

diizuka I keep saying that we will start teaching unambiguous language use in elementary school that even machines can understand. I thought that would be natural, since we also teach them how to use the arithmetic board and PCs.

tokoroten Then it is very important to make LMM selectable, right? “Tiananmen Square Incident” will never come out of a Chinese LLM, and I think this is where domestic LLM is needed.

He who controls the past controls the future. He who controls the present controls the past. 1984


This page is auto-translated from /nishio/LLMă«ă‚ˆăŁăŠăƒă‚€ă‚ąă‚čăźæŽ›ă‹ăŁăŸèš€èȘžăžăźă‚·ăƒ•ăƒˆ using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.