Re: Deyan's list of concept names

Hi Neil,

Certainly, not meant to be trusted - but useful to accelerate a quick
initial seed, which can then be vetted by an expert. Even in the case of
the original Google Spreadsheet I had made I used an old "concept index"
system we had, which had auto-crawled Wikipedia (and other encyclopedic
sites) a long time ago. So automated tools help a lot - but need extensive
vetting to get to "trust".

Indeed there may be less controversial prompts one could use, which avoid
asking for a specific number of concepts, or for "most common". If I
seriously get back to extending the Open list, I would try with a sequence
of targeted prompts, asking for specific grade levels and subject areas,
which should help us get closer to 10,000 concept names.

Btw, in case anyone is wondering, the permittivity-of-vacuum[1] and
permeability-of-vacuum[2] concepts are quickly detected in grade 12 physics
education in India, via a classic search that appends "khan academy" to
them. Which is just my expedient heuristic to check if something is
ubiquitous in educational materials. So while it is quite hard to
meaningfully compare if they are more or less common than "mean" and
"median" (and to whom? an electrical engineer may have a different answer),
at least they fit the K-12 designation. Another point is that since they
have natural self-voicing speech, it is possible they fit better as
":property" annotations. But I think that is still an open question - they
should likely get the same treatment that the imaginary-unit (i vs j) would
get, whichever direction we decide that.

[1]
https://www.khanacademy.org/science/in-in-class-12th-physics-india/in-in-electrostatic-potential-and-capacitance/x51bd77206da864f3:capacitance-parallel-plate-capacitors/a/capacitors-article
[2]
https://www.khanacademy.org/science/physics/magnetic-forces-and-magnetic-fields/magnetic-field-current-carrying-wire/a/what-are-magnetic-fields

Greetings,
Deyan

On Sat, Jul 1, 2023 at 12:31 AM Neil Soiffer <soiffer@alum.mit.edu> wrote:

> I really should get in the habit of using AI more often!
>
> The lists you have are impressive, but as is (currently) the case, not to
> be trusted. E.g, you asked for a list of 100 items and you got 78 back.
> More importantly, the list is clearly not "the most common mathematical
> symbols and notations that are universally taught in K-12 STEM education".
> For example, it contains permittivity-of-vacuum and permille but doesn't
> mention mean or median. Nonetheless, the lists can be used as lists from
> which to grab notations.
>
> Thanks,
>
>     Neil
>
>
> On Thu, Jun 29, 2023 at 9:42 PM Deyan Ginev <deyan.ginev@gmail.com> wrote:
>
>> Hello Neil, everyone,
>>
>> Thanks for fixing the spreadsheet.
>>
>> That 2020 list feels outdated in multiple ways nowadays (even though the
>> content should still be useful with some curation).
>> It's a whole new world out there. I recently jumped into the GPT-4 chat
>> interface, and - well - you can just ask the large language model to
>> bootstrap a list of this nature. That list will need secondary curation and
>> trimming down (e.g. what belongs in Core? what should be a property instead
>> of intent?), but so did my original spreadsheet.
>>
>> Here's a list of names with LaTeX notations (generated near instantly) in
>> a 10 minute chat session. It's nice that OpenAI now gives shareable links
>> to the sessions.
>>
>> It was much faster, and a lot less painful, to generate a few hundred
>> concept names by prompting the model, than it was to spend a few weeks
>> pouring over encyclopedia pages:
>> https://chat.openai.com/share/aa8d65c6-ac72-405d-a789-e1d2b784292e
>>
>> And it can easily provide more concepts, and be guided to the domains it
>> missed... probably thousands more.
>>
>> Greetings,
>> Deyan
>>
>> On Thu, Jun 29, 2023 at 2:13 PM Neil Soiffer <soiffer@alum.mit.edu>
>> wrote:
>>
>>> On the call, I tried to show the list of intent names that Deyan came up
>>> <https://docs.google.com/spreadsheets/d/1EsWou1K5nxBdLPvQapdoA9h-s8lg_qjn8fJH64g9izQ/edit#gid=1358098730>
>>> with but it was clearly not correct. I think I must have sorted the first
>>> column by mistake and it was messed up. I have reverted to the previous
>>> version so the list is back to correct.
>>>
>>>