On 10/07/2025 09:38, Tristan Miller wrote:
Greetings.
On 2025-07-10 07:26, Tristan Wibberley wrote:
[Just now, ChatGPT] had no problem recognizing equations written in
LaTeX markup.
... "recognizing" ... implies mind via "cognize". Is "recognizing"
an accepted technical term for LLMs or should we say "predicting from"
?
["recognize"] has been used
academically in the context of AI systems since long before the advent
of LLMs, and has even become entrenched in popular usage through terms
such as "speech recognition" and "facial recognition". There are even plenty of computational but non-AI uses going back to the 1950s -- books
and articles write of computers "recognizing" magnetically encoded bits
on a storage medium, or symbols in a computer program, or numbers within
a certain range.
Thanks for the detailed and helpful response.
It seems those uses of "recognize" and "recognition" are specifically
for classification and, typically, classification of an input as a representation of a specific symbol.
Do you find that normally a face is said to be recognized when the
system provides a name, classified when it provides a species, detected
when it provides an assertion of presence?
I wonder if "recognize" is properly applied to LLM prediction of a
statement that asserts a specific member of a class rather than for just
any useful prediction or for prediction of any classifying statement in general?
That continues to raise some interesting questions. If you ask "is
[code] latex?" and the system responds "yes", did it even get as far as 'classifying' the code as latex? Did the system merely admit it to a "similar-to-latex" set rather than discriminating the code from other
things from which you would expect it to be discriminated when it is classified?
--
Tristan
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)