Eyal En Gad

[Eyal En Gad @eyal_eg]

New paper:
Polynomially larger deletion codes.
For two deletions, we improve the leading redundancy coefficient to 3.
The previous best coefficient was 4, unchanged since 1965.
The proof is elegant, elementary and formalized.

https://t.co/49cT8ssTjn

[Eyal En Gad @eyal_eg]

@littmath Not anymore

[Eyal En Gad @eyal_eg]

@DimitrisPapail @MountainOfMoon Internal models are probably also much better at AI research, the labs just don't publish that research

[Eyal En Gad @eyal_eg]

@letonyo Next month the next model will likely solve most of them.

[Eyal En Gad @eyal_eg]

@Thusatharan @southpolesteve Replace Google docs with GitHub

[Eyal En Gad @eyal_eg]

@lu__jasper Recomputing the KV cache?

[Eyal En Gad @eyal_eg]

@Aaroth Does anyone have time to read the deslopped paper with the amount of results dropping?

[Eyal En Gad @eyal_eg]

In my experience so far, if anthropic doesn't start quantizing opus 5.5 soon, math is dead

[Eyal En Gad @eyal_eg]

@MountainOfMoon Thanks! Forgot to mention that there's also a lean repo

https://t.co/YplZDXd1Mw

[Eyal En Gad @eyal_eg]

@MountainOfMoon Hi Arya, self promo, if you like improvement on old coding bounds, I got one last week for deletions that may interest you
https://t.co/49cT8ssTjn

[Eyal En Gad @eyal_eg]

@ipeirotis AI can already judge the importance of results pretty well and connect it to what we know. It is not yet good enough at explaining, but that could change quickly

[Eyal En Gad @eyal_eg]

@kfountou With astra ultra i formalized a 20 page coding theory paper in 6 hours

[Eyal En Gad @eyal_eg]

@letonyo I just leave it for the next version of the AI that will be smart enough to rewrite it in a clear way.

[Eyal En Gad @eyal_eg]

For t deletions, the coefficient improves to from 2t to 2t-1.
Developed and written with digital friends.
Formalization:
https://t.co/YplZDXd1Mw

[Eyal En Gad @eyal_eg]

New paper:
Polynomially larger deletion codes.
For two deletions, we improve the leading redundancy coefficient to 3.
The previous best coefficient was 4, unchanged since 1965.
The proof is elegant, elementary and formalized.

https://t.co/49cT8ssTjn

[Eyal En Gad @eyal_eg]

Submitted a nice result to arxiv, improving a 60 year old basic combinatorial gap.
Preview:
https://t.co/YplZDXd1Mw

[Eyal En Gad @eyal_eg]

@kfountou That's still the hard part, but probably not for long

[Eyal En Gad @eyal_eg]

@neatpromptsai @Hesamation Because their models finally because good enough for it

[Eyal En Gad @eyal_eg]

@gautamcgoel Yes, go a couple of steps back, fork and ask it to write down a summary of the conversation. Download and give it to a new chat

[Eyal En Gad @eyal_eg]

@DimitrisPapail also don't forget to run:
while 'proof can be simplified': simplify; end
(credit: zac tamo)