LLMs, cats, and fig trees

June 8, 2026 @ 11:24 am · Filed by under Artificial languages, Linguistics in the comics

« previous post | next post »

The latest Questionable Content follows Yann LeCun:

See also this clip, among others.

June 8, 2026 @ 11:24 am · Filed by under Artificial languages, Linguistics in the comics


12 Comments »

  1. Jonathan Smith said,

    June 8, 2026 @ 1:47 pm

    Incidentally I asked Chat what these tools should *really* be called given "Large Language Model" is nerd (specifically: engineer/statistician) view, and it suggested (among crappier ideas) Language Engine, which duh, so this should happen.

  2. JPL said,

    June 8, 2026 @ 7:29 pm

    @Jonathan Smith:

    I think "LLM" seems to refer to extremely large corpuses of texts. A collection of texts is not a model. The model enters with the program's statistical analysis of the collocation relations exemplified in the text corpus. A more apt name for the endeavour could probably be found; I don't think "language engine" would be a good one, but maybe we should leave it to the wags.

  3. KeithB said,

    June 9, 2026 @ 7:40 am

    Internet Trained Guided Neural Network?

  4. KeithB said,

    June 9, 2026 @ 7:43 am

    Ah Wit of the staircase!, I forgot something:
    Internet Trained Guided Neural Network with a special carve-out to count the number of "r's" in strawberry.

  5. Ken said,

    June 9, 2026 @ 8:27 am

    @KeithB: The "strawberry" error illustrates one route to LLM collapse. At this point, there may be more internet posts saying "strawberry has four Rs" than the correct answer, because of humans circulating "look how stupid the AI is" memes, and LLMs trained on that corpus become more likely to repeat the mistake.

  6. Jonathan Smith said,

    June 9, 2026 @ 7:07 pm

    I just mean "(Large language) Model" = the network trained on the corpus; this is not a sensible name from the standpoint of the end user of these tools as (most commonly) employed. The kids say "ask Chat". So perhaps call these tools ChattyKathies or ChatterBoxes.

  7. Michael said,

    June 11, 2026 @ 5:54 pm

    I work in IT. I was recently "corrected" by a colleague who told me LLM stands for "Language Learning Model."

  8. Robot Therapist said,

    June 12, 2026 @ 5:49 am

    I take issue with Yann LeCun. The difference between LLMs and cats is not so much the number of synapses. It's that cats are trained on the real world, whereas LLMs are trained on things people have written on the internet. That's a HUGE difference.

  9. David Marjanović said,

    June 12, 2026 @ 7:26 am

    "(Large language) Model"

    Oh, so they're large-language models, not large language models?

    I've been fooled before: a "dank meme stash" is really a dank-meme stash, a stash of dank memes…

  10. Chris Button said,

    June 12, 2026 @ 2:42 pm

    I find the "language" part of the name confusing since the training inputs don't have to be text-based at all.

  11. JPL said,

    June 12, 2026 @ 4:35 pm

    @Jonathan Smith:

    How about "robot interlocutor"? (I thought of this before seeing Robot Therapist's handle above.)

  12. Ross Presser said,

    June 15, 2026 @ 1:08 pm

    I think the three-letter nature of "LLM" is going to keep it in use. Shortness wins vocabulary. "Robot interlocutor" is five syllables — ripe for being shortened to "RI".

    Then there's always "Agent". Very short and effective. Today one thinks of "agent" as "when I tell the computer to figure out how to do something I don't myself know how to do, and then do it.", but eventually it may take over for "when I talk to the computer about something I'm confused about."

RSS feed for comments on this post · TrackBack URI

Leave a Comment