[seqfan] Re: Sequence from ChatGPT

[Alex Meiburg]

Without getting too deep into the epistemology of what it means to
"understand" something
vs to merely "know" it, I believe that is an unfair characterization of
ChatGPT. Humans never
directly learn a fact about the nature of the universe: we are merely a
neural network that
is training on a multimodal input stream of
{vision,sound,touch,taste,smell,equilibrioception,
proprioception}. Mostly vision.

And look, you show someone to an optical illusion printed on a 2D page? And
they say it looks
3D! They'll even happily point which parts of the page are "higher" than
others!
If you put a VR headset on them and play a scene of a wrecking ball coming
at their head? They
get deep fear and stress responses, and probably instinctively try to evade
it! Even though it's
just a screen! And people with missing limbs? Still experience touch and
proprioception!

The fact that humans make these "mistakes" is because we're just noisely
trained noisy machines,
just like GPT. We're trained (and evolved) for a very different set of data
than GPT, so of course
our biases and mistakes are very different. Do I trust GPT to write good
proofs? No. But saying
that its current failure to do so indicates a lack of understanding is
pretty silly: you're judging
a fish by its ability to climb a tree.

I would like to remind you that neither you nor I have gills.

-- Alexander Meiburg


On Mon, Apr 3, 2023 at 5:28 AM M. F. Hasler <oeis at hasler.fr> wrote:

> On Sun, Apr 2, 2023 at 4:04 PM <israel at math.ubc.ca> wrote:
>
> > As for looking up something, ChatGPT appears to know about
> > SuperSeeker.
> > Response: ******************* To check if the Online Encyclopedia of
> > Integer Sequences (OEIS) contains a sequence containing both 3633 and
> > 3598745, we can use the "superseeker" feature on the OEIS website.
> >
>
>
> > Using the superseeker, we can search for sequences that contain both 3633
> > and 3598745 in their list of terms.
>
> ...
> > As we can see, there is only one sequence in the OEIS that contains both
> > 3633 and 3598745 in its list of terms, and that sequence is A262132.
> > ****************************
> > Which sounds great until you check that A262132 is not that at all, and
> > does
> > not contain either 3633 or 3598745.
> >
>
> Which shows that "ChatGPT knows" means that it just reproduces what it has
> "read"
> without understanding anything, or even being able to "use" this
> "knowledge" otherwise than by making new phrases from it.
> (Although it might be within reach of the technology to make an
> auto-curated knowledge base (with semantics) of what can be found on
> internet (attaching probabilities or "confidence levels" on the
> "data"(relations) depending among others on the source and an possible
> "proofs" from other, related "data")
> -- which is not what ChatGPT does, however).
>
> - Maximilian
>
>
> > On Apr 2 2023, Alonso Del Arte wrote:
> >
> > >Proofreading might be a more productive use of ChatGPT here, e.g., "how
> > >would you explain that numbers with property *P* also have property
> *Q*?"
> > >Such a prompt might generate more useful content than asking it to come
> up
> > >with an entire new entry from scratch.
> > >
> > >By the way, has anyone tried looking up something on the OEIS using a
> > >mobile device rather than a desktop computer, e.g., a Samsung Galaxy
> S8+?
> > >If you don't know how to fix that problem, ChatGPT might be able to help
> > >you.
> > >
> > >Al
> > >
> > >
> > >
> > >
> > >
> > >On Sun, Apr 2, 2023 at 12:37PM Arthur O'Dwyer <
> arthur.j.odwyer at gmail.com>
> > >wrote:
> > >
> > >> On Sun, Apr 2, 2023 at 10:43AM <israel at math.ubc.ca> wrote:
> > >>
> > >> > Out of curiosity, I asked ChatGPT:
> > >> >
> > >> > Write a new sequence for the Online Encyclopedia of Integer
> Sequences
> > >> >
> > >> > The result was
> > >> >
> > >> [nonsense]
> > >>
> > >> Please don't submit machine-generated spam to OEIS. As I understand
> it,
> > >> OEIS is supposed to be a mathematical resource. If it ends up full of
> > >> machine-generated spam, it becomes useless.
> > >> See also
> > >>
> > >>
> > >>
> > >>
> >
> https://meta.stackoverflow.com/questions/421831/temporary-policy-chatgpt-is-banned
> > >> https://news.ycombinator.com/item?id=33891538
> > >>
> > >>
> > >> The sequence of concatenations of Fibonacci
> > >> > numbers and squares in both orders (not allowing leading 0's) is
> > >> >
> > >> > 10, 11, 12, 13, 14, 15, 18, 19, 20, 21, 24, 29, 30, 31, 34, 39, 40,
> > >> > 41, 42, 43, 45, 48, 50, 51, 54, 59, 80, 81, 84, 89, 90, 91, 92, 93,
> > >> > 95, 98, 113, 116, 121, 125, 130, 131, 134, 136, 139, 149, 155, 160,
> > >> > 161, 162, 163,
> > >> 164,
> > >> > 165, 168, 181, 189, 210, 211, 214, 216, 219, 225, 236, 249, 250,
> 251,
> > >> 252,
> > >> > 253, 255, 258, ...
> > >> >
> > >> > which doesn't seem to be in OEIS, and I think I'll contribute it.
> > >>
> > >>
> > >> Why? Is there anything noteworthy about this sequence?
> > >> "Computer <https://en.wikipedia.org/wiki/Computer_says_no>said
> > >> <https://en.wikipedia.org/wiki/Computer_says_no> submit it" is not
> > >> noteworthy. Computer says lots of things.
> > >>
> > >> I think it's particularly un-mathematically-interesting because
> > >> "concatenation" is nothing but an artifact of base-10 notation. If you
> > >> think there's something special about concatenating Xs and Ys in base
> > 10,
> > >> why not also in base 8, or base 3, or base 37?  And why squares but
> not
> > >> cubes, Fibonacci numbers but not strings of 1s,...
> > >>
> > >> -Arthur
> > >>
> > >> --
> > >> Seqfan Mailing list - http://list.seqfan.eu/
> > >>
> > >
> > >
> > >
> >
> >
> > --
> > Seqfan Mailing list - http://list.seqfan.eu/
> >
>
> --
> Seqfan Mailing list - http://list.seqfan.eu/
>