latently:2026.00001v1  [cs.READ]  10 Jun 2026

Papers, meet pocket.

On re-typesetting the scientific PDF for the hand

correspondence: alpha@latently.app

*  building the reader you deserve.

every paper, reflowed to fit your phone. see it in full ↓

Abstract

The scientific PDF assumes a sheet of A4 paper. Most reading now happens on a digital screen. Latently takes any paper and turns it into a readable format that actually fits your phone, or whatever device you're using. Equations and figures come along too. I'm starting with a small alpha on iOS. 2 It's free while we're in alpha, and some papers still come out wrong.

2  a few hundred TestFlight seats. no card, you'll just need an iPhone or iPad.

1The problem

You know this feeling. You open a paper on your phone and zoom in so you can read it. Now every page is a little maze: down one column, back up, across to the next.

exhibit A · the PDF, as your phone sees it

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder and a decoder. The best performing models also connect the encoder and decoder through an attention mechanism.

We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train.

see? hostile.now keep reading ↓

By page 3 you save it for later — but later never comes.

2This is how it should be

The whole thing reflows to whatever width your screen is. That's it, that's the product. Exhibit B, the same abstract as set by Latently:

arXiv:1706.03762 · cs.CL · as set by Latently

Attention Is All You Need

A. Vaswani · N. Shazeer · N. Parmar · +5 — NeurIPS 2017

We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train.

Attention(Q, K, V) = softmax(QKᵀ / √dₖ) V

ahh. better.

3Same paper, before and after

Drop in a PDF or link to a paper and watch it become something you can actually read on your phone.

The Attention Is All You Need PDF as it looks on a phone (a) the PDF, on a phone
arXiv:1706.03762 · cs.CL
Attention Is All You Need
A. Vaswani · N. Shazeer · N. Parmar · +5 — NeurIPS 2017

We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely.

Attention(Q, K, V) = softmax(QKᵀ / √dₖ) V

Experiments on two machine translation tasks show these models to be superior in quality…

67% read
(b) the same paper, in Latently
Figure 1: the same paper, before and after.

4How it works

You give Latently a PDF. It takes the layout apart and rebuilds the paper as one column that fits whatever you're reading on. Figures stay near the text that mentions them. Equations are rendered properly.

arXiv:1706.03762 · cs.CL
Attention Is All You Need
A. Vaswani · N. Shazeer · N. Parmar · +5 — NeurIPS 2017

We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely.

Fig. 1 · the Transformer architecture
67% read
(a) in the browser
arXiv:1706.03762 · cs.CL
Attention Is All You Need
A. Vaswani · N. Shazeer · N. Parmar · +5 — NeurIPS 2017

We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely.

Attention(Q, K, V) = softmax(QKᵀ / √dₖ) V

Experiments on two machine translation tasks show these models to be superior in quality…

67% read
(b) in your pocket
Figure 2: probably the most-read paper in machine learning. Now it fits wherever you are — (a) in the browser, (b) on iOS.

5It remembers where you stop

Close the app on the train, open it in bed and you're back at the same paragraph.

6What the first release brings

The first release is about three things: the reflow, the equations, and your reading position. Everything else is on the note.

[ request a feature → ]

Reading papers shouldn't
be this hard.

Latently is on a mission to build the reader you deserve. It's early and some papers still come out wrong. If you read papers on your phone anyway, come break it and tell me what you find.

private alpha · one email when the doors open

Two ways in.

The alpha runs on iOS through TestFlight, so you'll need an iPhone or iPad. On any other device, a web app is on the way — join the waitlist and you'll get one email when it's ready.

Latently · Alpha Pass first to read
From Pinch & zoom down one column, back up
To Just scroll it reflows to fit
Alpha Pass Alpha tester
seat№ 0001
cohort01
gateTestFlight
No. 0042 · cohort 01
limited seats · replies within 7 days · no card