this post was submitted on 15 Apr 2025
25 points (100.0% liked)
Technology
38547 readers
256 users here now
A nice place to discuss rumors, happenings, innovations, and challenges in the technology sphere. We also welcome discussions on the intersections of technology and society. If it’s technological news or discussion of technology, it probably belongs here.
Remember the overriding ethos on Beehaw: Be(e) Nice. Each user you encounter here is a person, and should be treated with kindness (even if they’re wrong, or use a Linux distro you don’t like). Personal attacks will not be tolerated.
Subcommunities on Beehaw:
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
So imagine the language model can produce grammatically correct and semantically meaningful dolphin language, how does it translate that to a human language?
The reason LLMs can do this for human languages is that we have an enormous corpus of Rosetta stones for every language that allow the model to correlate concepts in each language. The training data for human to dolphin is going to be just these “behavioural notes.”
So the outcome is that the bullshitting machine will bullshit the scientists that it knows what they’re saying when it’s actually just making stuff up.
It’s a big problem with LLMs that they very rarely answer, “I don’t know.”
LLMs use a tokenizer stage to convert input data into NN inputs, then a de-tokenizer at the output.
Those tokens are not limited to "human language", they can as well be positions, orientations, directions, movements, etc. "Body language", or the flight pattern of a bee, are as tokenizable as any other input data.
The concepts a dolphin language may have, no matter what they are, could then be described in a human language, and/or matched to human words for the same description.