yeah I agree. I see what you're saying about a language/meaning distinction not being the cause of this issue, but I do think a language/meaning distinction could be a solution!
29.01.2026 19:23 β π 1 π 0 π¬ 1 π 0@annaclemencia.bsky.social
have you seen something invisible today?
yeah I agree. I see what you're saying about a language/meaning distinction not being the cause of this issue, but I do think a language/meaning distinction could be a solution!
29.01.2026 19:23 β π 1 π 0 π¬ 1 π 0agree at the LLM level. disagree at the human decision-maker level.
29.01.2026 19:17 β π 0 π 0 π¬ 0 π 0the LLM would be pretty good at what, exactly? interpreting meaning like a person? making a decision/suggestion that someone wants?
29.01.2026 19:13 β π 1 π 0 π¬ 1 π 0I understand you better now. I agree with what you're saying about reliability and some of the issues with Bender's framing.
29.01.2026 19:04 β π 0 π 0 π¬ 0 π 0It's already proven true somewhat, like in Perceptrons by Minksy and Papert, they razz all the guys who thought that computers would just be able to interpret the meaning of images without any sort of "meaning" coded separately from the images
29.01.2026 17:55 β π 0 π 0 π¬ 0 π 0I do think treating language and meaning as collapsible gives LLMs unearned legitimacy/agency in some cases (maybe we agree on that?). I also think systems designed with language/ meaning as separate will end up being more useful in the long run (but that's mostly me betting on myself).
29.01.2026 17:44 β π 0 π 0 π¬ 2 π 0We agree that we can get systems that more or less do what we want with arbitrarily large arrangements of words. I don't think it matters where anyone falls on where "meaning" lies in a "what is meaning really" sort of way.
29.01.2026 17:42 β π 0 π 0 π¬ 1 π 0True, the article conflates "AI" with an LLM and not a bad search function. But the belief that language and meaning are interchangeable is what gives both those systems legitimacy (and in this case, it wasn't good to do). If we don't use them interchangeably, we could better avoid bad outcomes
29.01.2026 17:36 β π 0 π 0 π¬ 1 π 0That's not to say we couldn't have systems that differentiate language in ways that we want, because we obviously can. But starting from a place where we treat language and meaning as the same will hinder that goal in many cases.
29.01.2026 16:59 β π 0 π 0 π¬ 1 π 0At its most basic, I see her work as trying to get people to realize or admit "officer" doesn't always mean "police training." When we collapse language and meaning, it makes it a lot harder for people to anticipate this sort of outcome.
29.01.2026 16:53 β π 0 π 0 π¬ 1 π 0Sure. I separate language and meaning in my work (for different reasons than Bender; I don't really care a lot about cognition). I understand Bender as taking her stance so things like this don't happen: www.nbcnews.com/politics/imm...
29.01.2026 16:50 β π 0 π 0 π¬ 1 π 0Do you not like the idea of language and meaning as separate, or you don't like their reasoning for why? Or both?
29.01.2026 05:59 β π 0 π 0 π¬ 1 π 0We are hiring!
Interested? Apply here: go.mbl.edu/AS1887
Nature suggests you use their "Manuscript Adviser" bot to get advice before submitting
I uploaded the classic Watson & Crick paper about DNA structure, and the Adviser had this to say about one of the greatest paper endings of the century:
You might especially like "The Status of Style" (1975)
23.10.2025 18:58 β π 1 π 0 π¬ 1 π 0I study this for image data! I think all the best writing on the topic comes from art history. Their work is not quantitative, but analytical in other ways. Whenever I start to feel like I've lost the plot, I go back to Nelson Goodman's work on style: plato.stanford.edu/entries/good...
23.10.2025 18:53 β π 1 π 0 π¬ 1 π 0do itttttt
28.08.2025 16:59 β π 0 π 0 π¬ 0 π 0process being magic of course
28.08.2025 16:58 β π 0 π 0 π¬ 0 π 0also just say you think writing is the same as thought and go home?
28.08.2025 16:55 β π 0 π 0 π¬ 0 π 0hahahaha I hate it here
31.07.2025 18:27 β π 1 π 0 π¬ 0 π 03-year Postdoc in philosophy of biology w/ @lucielaplane.bsky.social in Paris on cell lineages; interdisciplinary project w/ mathematicians & biologists. Conceptually-minded historians & social scientists also welcome to apply. Project start in early 2026. Application opens soon #HPS #hpbio #STS
16.07.2025 17:32 β π 44 π 36 π¬ 1 π 1On the flip side, I think we're going to find out what kind and how many things individuals really need to "know." And my guess rn is those things will be far fewer and different than we assume.
16.07.2025 05:04 β π 3 π 0 π¬ 1 π 0so what you're saying is I can avoid penalty by never publishing got it
31.05.2025 21:50 β π 0 π 0 π¬ 0 π 0you're a natural
30.05.2025 19:18 β π 1 π 0 π¬ 0 π 0this is why I invite you to "bird gazing" club, where we don't worry about names only vibes
30.05.2025 18:59 β π 0 π 0 π¬ 0 π 0if a scrub jay got you you're toast
30.05.2025 18:23 β π 1 π 0 π¬ 0 π 0say it with me words are not the world
23.05.2025 15:52 β π 1 π 0 π¬ 0 π 0Always doubt.
23.05.2025 15:03 β π 504 π 81 π¬ 0 π 8