Al Merose (he/him)
banner
al.merose.com
Al Merose (he/him)
@al.merose.com
Technical staff @OpenAthena.ai, community member @pangeo.io & @m2lines.bsky.social. Weather, Climate, Oceans Research. Programming, Politics, Postmodernism (oPinions are my own).
Yes, and this works, but I have a few bones to pick: 1. The synthetic voice quality is poor and not natural sounding to me 2. It lacks mobile support IIRC.
November 15, 2025 at 4:57 AM
(full disclosure: I used to work at aira).
November 15, 2025 at 2:22 AM
spaCy, like a lot of traditional NLP, was outmoded pretty immediately post-GPT. Likely, not worth learning too much about (alas).

I think TTS and screen readers have been a long time huge win for accessibility. Maybe a frontier with AI involved would look like this: aira.io/projectastra/
Aira | Aira’s AI Visual Interpreter powered by Project Astra
Join the waitlist to participate in a trusted testers program of Aira’s AI Visual Interpreter powered by Project Astra.
aira.io
November 15, 2025 at 2:22 AM
I haven’t tried that yet! My plan was to use the ar5iv.org html pages and maybe pandoc to get an initial string for further processing. I think some good ol’ spacy and nltk utilities would make for good TTS post processing scripts.
ar5iv – Articles from arXiv.org as responsive HTML5 web documents
ar5iv offers a modern web view for arXiv's preprints. An open community resource, on a quest to a full collection of high-quality documents.
ar5iv.org
November 15, 2025 at 2:11 AM
My idea for a scrappy prototype would be to create a little webpage that just did the text processing to convert HTML versions of the paper to a Google doc, and then let ppl use their premium workspace subscriptions to listen to the document. A pure copy doesn’t make for good TTS/audio.
November 15, 2025 at 2:04 AM
I bought the domain aarxiv.org with the intention of doing the same thing, but it would generate a dictation of the paper. Still haven’t got around to building this, but it’s something I know I would use (dreaming of taking a walk while consuming papers).
aarxiv.org
November 15, 2025 at 2:00 AM
Have you tried replacing the “x” in an arxiv URL with a 5?
November 15, 2025 at 1:35 AM
I think this is how gmail implemented undo send!
November 11, 2025 at 4:11 AM
Goddamn you weren’t kidding.

Grotesque poetry if there is such a thing.
November 10, 2025 at 4:33 AM
I don’t know if I can do body horror
November 10, 2025 at 3:44 AM
My mistake, Doll, does it also go by “you”?

I like AethelOS so far.
November 9, 2025 at 11:06 PM
Oh, and I would recommend using Arrow as a IPC format, especially if you have a DB based FS

arrow.apache.org
Apache Arrow
The universal columnar format and multi-language toolbox for fast data interchange and in-memory analytics
arrow.apache.org
November 9, 2025 at 10:05 PM
I’m catching up on the sources. At a high level, if you wanted to implement the best ideas from arcs:
- I would take the IFC as a type system idea from Raksha… github.com/google-resea...
- …and implement them as rewrite rules with e-graphs: egraphs-good.github.io
github.com
November 9, 2025 at 10:01 PM
Atom54 Overview
YouTube video by Atom54
youtu.be
November 9, 2025 at 9:40 PM
A few of the core developers of Arcs went on to reinvent the project with XenonJS.

xenonjs.com
Atom54
xenonjs.com
November 9, 2025 at 9:29 PM
I’m glad I saw it in theaters, you’ll have a great time :)
November 8, 2025 at 7:03 AM