David Savitt
@dsavitt
Followers
812
Following
960
Media
164
Statuses
4K
Math department chair at Johns Hopkins University and a director of Canada/USA Mathcamp.
Baltimore, MD
Joined September 2008
And did so repeatedly, even after I called it on it. When I started a fresh chat and deleted the previous one, it was fine.
0
0
0
However, I did have to start a new chat for each department (deleting the previous one just to be safe). The first time I tried doing a second department in the same chat, it just hallucinated the faculty listings rather than parsing the uploaded data.
1
0
0
I eventually got satisfactory results by downloading the HTML from each of the websites, and uploading with a prompt to parse it into a spreadsheet with first name, last name, and title. (Gemini . GPT-5 had trouble even with this). Then I categorized each title by hand.
1
0
0
I think part of what was going on was that some of the websites weren't scrapeable and it wasn't being forthright about this. Another part is that in a longer "conversation" it would get lazy or sloppy.
1
0
0
AI is definitely not as good at humans at reading math department websites and counting up how many tenure-track faculty, teaching faculty, postdocs, and PhD students there are.
1
0
7
summary:.1) it gives incorrect argument but in the right spirit.2) I ask it to look for problems, it spots the two issues I had in mind and attempts to fix them.3) this is much closer, but still has some issues, which I direct it towards.4) it gives more or less correct proof.
1
0
1
Asked GPT-5 my recent-favorite question about congruence subgroups. Far from perfect but much better than previous models, I think.
1
0
3
RT @JHUmath: We are deeply saddened by the passing of legendary algebraic topologist and @JHUmath professor emeritus Jack Morava. https://….
0
10
0
RT @nadienadianadie: An annoying post here reminded me of one of my favorite things, the opening of this writeup from a talk by Jack Morava….
0
3
0
Finally, something of critical importance that cutting edge LLMs just can't do:.
gemini.google.com
Created with Gemini
1
0
1
Of to Colorado for a few days to pretend to be an algebraic geometer. Hope to see some of you there!.
0
0
6
Recently discovered that I can do a decent Scottish accent. New plan: insist on speaking to my university administration in Scottish accent until they stop asking me to do service.
0
0
2
Continues here:.
Now, the two conditions in Newman's theorem (the two divisibilities by 24) are precisely the necessary condition that an eta-quotient transform correctly under the two matrices T and L_N:.[ 1 1 ] .[ 0 1 ] .and.[ 1 0 ].[ N 1 ] .
0
0
0
I should add here, Dedekind sums have been making a comeback, e.g. in the work of @samit_dasgupta and others on Stark-Heegner points. (Or maybe I'm wrong that they'd slipped out of fashion?).
Dedekend sums are sums of products of sawtooth functions, and I think used to be more in fashion than they are now. Newman's argument is simply this. We want to see whether an eta-quotient is invariant under the action of the subgroup Gamma_1(N) of SL(2,Z).
1
1
2
The notes for the Mathcamp course are here, for anyone interested:.In 2 weeks / 10 lectures I made it to the end of section 3, i.e., the level one stuff.
dropbox.com
Shared with Dropbox
0
0
1
Alright, let me stop here. Thanks for reading to the end. The moral, if there is one, is: thinking about how to explain interesting things to smart high school students is great, and if you have the opportunity, you should do it.
1
0
0
Less surprisingly, the LLMs were quite helpful with proofreading. Gemini especially caught some typos that I'm very glad it caught.
1
0
0