[ExI] Bender's Octopus (re: LLMs like ChatGPT)

[Gordon Swobe]

On Thu, Mar 23, 2023 at 7:16 PM Giovanni Santostasi <gsantostasi at gmail.com>
wrote:

> Gordon,
> Basically what Bender is saying is "if the training of a NLM is limited
> then the NLM would not know what certain words mean".
>

No, that is not what she is saying, though seeing as how people are
misunderstanding her thought experiment, I must agree the experiment is not
as clear as it could be. She is saying, or rather reminding us, that there
is a clear distinction to be made between form and meaning and that these
language models are trained only on form. Here is the abstract of her
academic paper in which she and her colleague mention the thought
experiment.

--
Abstract: The success of the large neural language mod-els on many NLP
tasks is exciting. However,we find that these successes sometimes lead to
hype in which these models are being described as “understanding” language
or capturing “meaning”. In this position paper, we argue that a system
trained only on form has a priori no way to learn meaning. In keeping with
the ACL 2020 theme of “Taking Stock ofWhere We’ve Been and Where We’re
Going”,we argue that a clear understanding of the distinction between form
and meaning will help guide the field towards better science around natural
language understanding.
--
Bender is a computational linguist at the University of Washington. I think
I read that she is actually the head of the department.

the paper:
https://docslib.org/doc/6282568/climbing-towards-nlu-on-meaning-form-and-understanding-in-the-age-of-data-gts
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230323/71d942c7/attachment.htm>