Tal

[Cheng Lou @_chenglou]

My dear front-end developers (and anyone who’s interested in the future of interfaces):

I have crawled through depths of hell to bring you, for the foreseeable years, one of the more important foundational pieces of UI engineering (if not in implementation then certainly at least in concept):
Fast, accurate and comprehensive userland text measurement algorithm in pure TypeScript, usable for laying out entire web pages without CSS, bypassing DOM measurements and reflow

[Cheng Lou @_chenglou]

I’m very happy to present my toy research project: Sotaku!

It's a neural net that automatically discovered the rules of sudoku and learned to solve them, achieving a new state-of-the-art score of 98.9% on one of the hardest sudoku datasets, while being agnostic to the game, and beating all other sudoku-optimized neural net architectures*

Read more for fun motivations, plus some extremely unconventional discoveries, e.g. reverse curriculum consistently beating curriculum (!), emergent reasoning-like capabilities, and the future of traditional programming

[Tal @eiopa]

@shlomiatar @yacineMTB Authors were easy (thanks OpenReview!) but it was tricky to figure out the institutes affiliated with the paper

[Tal @eiopa]

Many thanks to @shlomiatar for being an awesome collaborator on this project

[Tal @eiopa]

I created this app for NeurIPS, mostly driven by my own desires to not feel lost 🙂
https://t.co/029Pbmq3lh

It’s small but mighty! 🧵

[Tal @eiopa]

Lastly, there’s a party hub! Send me yours if I missed it :)

See you all next week!

[Tal @eiopa]

There’s per-paper discussions, powered by GitHub/Giscus, as well as easy navigation to authors homepages

[Tal @eiopa]

@comma_ai landing page isn't a 30mb pdf of figures and dense math notation. how is this even research.

[Tal @eiopa]

@gabriel1 That is unfortunately not the case generally (unless your opinion happens to align with the interviewer’s).

Depending on specifics and sheer luck, this can get you flagged.

[Tal @eiopa]

@rakyll I’d really love a simple bulk product for using it as a document extractor.
PDFs in -> jsons out.
Latency requirements are much more relaxed than the regular completion API, so can be made more price competitive.

Also rate limits 😛

[Tal @eiopa]

@phirephoenix Just started playing it last week and omg so good

[Tal @eiopa]

@Drc0nfused @_joelsimon @viccpoes I did work on it, but it was a three way thing with @_joelsimon and @viccpoes

[Tal @eiopa]

@soniajoseph_ It reminds me of how people were trying to do video on the internet back in the 90s. Some good ideas, but tech isn’t there yet.

[Tal @eiopa]

@djbaskin I just got tested there today!

[Tal @eiopa]

@kovasb https://t.co/2a6lwcXxoV

[Rivers Have Wings @RiversHaveWings]

Similarities of various CLIP text embeddings to the text embedding of "a human being". It looks like out of the races I tried, "a white man" is considered the most similar to "a human being" and for all races I tried, women are considered less 'typical' human beings than men.

[Tal @eiopa]

@kovasb it works in CLIP (use the text encoder for text-text similarity) so there's a proof of existence :)

[Tal @eiopa]

@jm_alexia Would also add to this list the incredibly stylish Ape Out

https://t.co/ZIICiRS1gP