Like other commentators, I have worked with LaTeX for a long time (> 10 years) and have to say that it very much shows its age.
For once, the compilation time is terrible - it can take several minutes to compile a typical document.
I recently tried to create a poster for a conference - a single A0 page - and even this took more than a minute to compile.
Because positioning in images and figures in LaTeX is not obvious, this made iteration extremely slow.
Additionally, by forgetting a single bracket, one can easily encounter cryptic error messages relating to boxes.
In the past, I have also worked on LaTeXML (the LaTeX-to-HTML "converter" amongst other things being used to power ar5iv [1]).
Even though this exposed me to some of (La)TeX's internals, it is still extremely difficult trying to decipher error messages.
Typst, on the other hand, is much faster when compiling and provides much clearer error messages.
For this reason, I also decided to write my 130-something PhD dissertation in Typst [2].
My university only provided a LaTeX and a Word template, but I ended up copying that in Typst extremely quickly.
My thesis compiles in about 7 seconds on my laptop - I cannot imagine how long it would take had I chosen LaTeX instead.
Not only that, when I showed an initial draft of my dissertation to my supervisor - also a decades-long LaTeX user - he could not tell that I wasn't using LaTeX.
I only really encountered one problem with Typst so far - citation styles.
In particular, the alphanumeric style preferred by my field produced duplicate citation keys, making the output unreadable in printed form.
But given all the other benefits Typst offered, I ended up switching to IEEE's citation style instead.
I am also hoping that Typst improves its HTML export - it is very barebones at the moment - but that seems to be in the works [3].
I'd second this - linking directly to `https://arxiv.org/abs/{{id}}` maybe even include the formats in the `Access Paper` box on the RHS of that interface would be a neat feature.
I believe the questionable legality also extends to any kind of video calls taking place on systems with this turned on.
If Recall indeed "does not perform content moderation", then having it on would include having screenshots of all video calls.
Especially in many-party consent jurisdictions recording video calls by taking screenshots at a sufficiently high frequency without the explicit consent, or at least option to opt out, of remote parties is probably illegal.
At one point a project in a research group I'm part of did automatic parsing of the entire OEIS to find relations between different existing sequences. Using a very simple approach, they found ~300 000 000 relations (e.g. meaning one sequence can be expressed as some combination of other sequences); see section 4 of [1].
However they submitted only three (!) of those back to OEIS.
Even with 130 reviewers on the OEIS side, submitting all of those relations would have basically been a Denial-of-service attack on the review process.
Wouldn't the grant given through section 5 of GitHub TOS [1] apply here?
That says "you grant each User of GitHub a nonexclusive, worldwide license to use, display, and perform Your Content through the GitHub Service and to reproduce Your Content solely on GitHub as permitted through GitHub's functionality".
In other words, without an explicit license you can do whatever you like with publically viewable content as long as it runs via GitHub.
Even if the official tool does shut down, a quick search shows several projects on GitHub to replace the official flashing utility. This seems to include a backup of the original site [1], an extended version [2] and a dump of the firmware [3].
However D4 also explicitly says "This license does not grant GitHub the right to sell Your Content". One could argue that because Copilot is a commerical product it is in fact selling (a derivative of) user code, and thus the grant in D4 does not apply.
For once, the compilation time is terrible - it can take several minutes to compile a typical document. I recently tried to create a poster for a conference - a single A0 page - and even this took more than a minute to compile. Because positioning in images and figures in LaTeX is not obvious, this made iteration extremely slow.
Additionally, by forgetting a single bracket, one can easily encounter cryptic error messages relating to boxes. In the past, I have also worked on LaTeXML (the LaTeX-to-HTML "converter" amongst other things being used to power ar5iv [1]). Even though this exposed me to some of (La)TeX's internals, it is still extremely difficult trying to decipher error messages.
Typst, on the other hand, is much faster when compiling and provides much clearer error messages.
For this reason, I also decided to write my 130-something PhD dissertation in Typst [2]. My university only provided a LaTeX and a Word template, but I ended up copying that in Typst extremely quickly. My thesis compiles in about 7 seconds on my laptop - I cannot imagine how long it would take had I chosen LaTeX instead. Not only that, when I showed an initial draft of my dissertation to my supervisor - also a decades-long LaTeX user - he could not tell that I wasn't using LaTeX.
I only really encountered one problem with Typst so far - citation styles. In particular, the alphanumeric style preferred by my field produced duplicate citation keys, making the output unreadable in printed form. But given all the other benefits Typst offered, I ended up switching to IEEE's citation style instead.
I am also hoping that Typst improves its HTML export - it is very barebones at the moment - but that seems to be in the works [3].
[1] https://news.ycombinator.com/item?id=36947004
[2] https://doi.org/10.25593/open-fau-1825
[3] https://github.com/typst/typst/issues/5512