一站式工作空间 让研究更清晰。

深入阅读,连接想法,让知识真正属于你。
文献、笔记与 AI,尽在一个专注的工作空间。

交互式产品预览 · 示例内容为英语

Example document
Highlight Section
Translate: Workdocs enables you to translate content from multiple languages into a single target language of your choice. Currently, we support translations in over 100 languages with fast, reliable performance.
Workdocs ermöglicht es Ihnen, Inhalte aus verschiedenen Sprachen in eine einzige Zielsprache Ihrer
Wahl zu übersetzen. Derzeit unterstützen wir Übersetzungen in über 100 Sprachen – schnell, stabil und zuverlässig.

Highlight

Copy

Translate

Ask chatbot ↘

I am a chatbot that helps you with your questions about The Work Docs. I can summarize a text for you, highlight important information, and even generate a table of contents for you. Just type your question or request in the box below and I will do my best to assist you.

Multiple Models Support
GPT-4.1 mini
Gemini 2.0 Flash
DeepSeek V3
Incoming Features
Search Browser
Browse and search the web
Ready
Deep Research
Advanced research tools
Incoming
WorkDocs Canvas
Visual workspace for docs
Incoming

Ask the PDF anything that you need

从第一个问题,到下一次发现。

Trusted by researchers from some of the best institutions in the world.

  • penn logo
  • cambridge university
  • MIT logo
  • Oxford university logo
  • stanford university logo
  • penn logo
  • cambridge university
  • MIT logo
  • Oxford university logo
  • stanford university logo
  • penn logo
  • cambridge university
  • MIT logo
  • Oxford university logo
  • stanford university logo
一个空间,连接一切

阅读、笔记与 AI,在同一个工作空间中紧密相连。

文档阅读与高亮

阅读 PDF,随时高亮、添加批注和颜色标记。

分屏视图、键盘快捷键与多色高亮,专为深度阅读设计。

可替代
Adobe Acrobat
Zotero
Mendeley
笔记系统

用简洁的笔记本记录想法,支持清单、内容块和便签。

将笔记链接到来源页面,按主题整理,并即时搜索全部内容。

可替代
Notion
Evernote
Apple Notes
AI 知识库

针对全部文档自由提问,让 AI 成为你的专属研究助手。

结合上下文作答,并引用你的文件、笔记和高亮内容。

可替代
ChatGPT
NotebookLM
Perplexity
12 highlights · 4 notes
Ch. 3 of 8
AT
Attention Is All You NeedVaswani et al. · NeurIPS 2017 · p. 3

The Transformer relies entirely on self-attention to compute representations of its input and output without using sequence-aligned RNNs or convolution, enabling significantly more parallelization.

Architecture
Key claim
ME
My margin note · Ch. 22h ago · synced
Insight

Multi-head attention lets the model attend to information from different representation subspaces. Each head learns a distinct projection — together they capture syntactic and semantic patterns in parallel.

Pinned quote
Exam-worthy
p. 12

“Attention is all you need — recurrence and convolutions are not required for state-of-the-art sequence transduction.”

Research — TransformersUpdated 5 min ago · 4 backlinks
Linked to PDF

Goal: understand why attention beats RNNs for long-range dependencies — prep for Friday seminar.

Read Section 3 — Model architecture
Highlight key equations (Q, K, V)
Sketch encoder-decoder diagram
Write 200-word summary in own words
Add 5 flashcards to Anki deck
STICKY NOTEPinned

Compare encoder-decoder attention vs self-attention — could be a great blog post. Reference Bahdanau (2014) as the historical anchor.

Lecture notes — Mar 12CS 224N · Stanford
Auto-synced

Prof. Tanaka emphasized that positional encoding is critical since attention has no inherent order. Sin/cos works because relative positions become linear functions of the encoding vector.

PE(pos, 2i) = sin(pos / 10000^(2i/d))

What does the paper say about scaling self-attention beyond 512 tokens, and how does it compare to Longformer's sliding window?

3 sources
My library
The Work Docs AI
Synthesis · 3 docs
Streaming

Vanilla self-attention has O(n²) memory and compute complexity, which makes contexts beyond 512 tokens impractical on standard GPUs. Three approaches dominate the literature:

Longformer — sliding window of 512 + global tokens → O(n)
Performer — kernel approximation with random features
FlashAttention — exact, but IO-aware tiling for speed

From your highlights: “the dot-product becomes the bottleneck above 1k tokens” — matches Vaswani §3.2.

Vaswani · p.6
Beltagy · p.3
Dao · p.2
My notes · Ch. 3
Ask anything…
为你研究与学习的 AI 助手

The Work Docs 了解你的学习流程,协助整理来源、总结论文,并提供附有引用的答案。

To Read
Attention Is All You Need
NeurIPS 2017 — 12 pages
V
Vaswani
BERT Pre-training
NAACL 2019 — 16 pages
D
Devlin
Reviewed
Drop here
Attention Is All You Need
NeurIPS 2017 — 12 pages
V
Vaswani
What did the author say about transformers?
Vaswani et al.Author · Google Brain
Attention Is All You NeedPDF · 12 pages
My highlights — Ch. 3Note · Updated 2h ago
Summarize Paper
Generate Notes
Find Citations
Translate Document
导出笔记
Ask anything…
Skills
整理阅读队列

按待读、阅读中、已阅等阶段整理论文。

搜索全部来源

通过一个搜索框查找作者、论文和自己的笔记。

一键调用 AI 技能

总结、测验、翻译,无需离开文档。

研究工具,汇聚于一个互联工作空间

了解每个来源的聊天助手、帮助你专注的阅读器,以及链接到原文的笔记,满足研究流程所需。

💬
AI聊天机器人

询问整个资料库的内容。实时获取答案及来自PDF、笔记和高亮的页级引用。

📓
智能笔记

链接到来源页的块式笔记。反向链接、清单、便签均可搜索。

🖍️
高亮与批注

多色高亮、边注和置顶引文,自动同步到笔记。

测验生成器
即将推出

将任意章节转为记忆卡或选择题,测试记忆并跟踪掌握程度。

🔖
引文查找

查找资料库中的相关段落,将引用拖入笔记。

Summarize paper

One-click TL;DR with key claims and methods.

Translate doc

Translate any PDF into 30+ languages.

Extract figures

Pull figures, tables, and equations to notes.

Compare papers

Side-by-side compare methods and results.

Generate flashcards即将推出

Auto-create Anki-ready cards from highlights.

Outline writer

Build paper outlines from your sources.

Define term

Inline definitions sourced from your library.

Find related

Surface related papers and notes by topic.

Export bibliography

BibTeX, RIS, or formatted APA / MLA.

Pull quotes

Extract highlighted quotes into a digest.

Knowledge recall

Resurface key facts marked weeks ago.

Smart highlight

AI suggests the lines worth highlighting.

由你的阅读、笔记和高亮驱动

Every paper, quote, and idea — connected into one knowledge graph. Find anything by following the links.

Attention Is All You Need.pdf
Shared · 1.8 MB
TypePDF Document
AuthorsVaswani et al.
ModifiedYesterday
Lit Review Session
Apr 22 · 45 min
Summary4 papers reviewed
Notes12 highlights
DateApr 22, 2026
Note · Transformer Arch
Updated 2h ago
Linked to3 sources
TagsNLP · ML
Backlinks5 notes
Highlight · "self-attention"
Page 4 · Yellow
来源Vaswani 2017
ColorYellow
Linked2 notes
AI Chat · "Explain heads"
Today · 4:32 PM
Sources2 papers
Citations5 pages
Saved toNote · TA
Citation · Vaswani et al.
NeurIPS 2017
Cited in3 notes
FormatBibTeX
Year2017
Quiz · Chapter 4
12 cards
Score9 / 12
Reviewed2 days ago
来源Note · TA
Flashcards · Encoder
18 cards
Mastered11 / 18
来源Note · TA
Due4 cards
BERT.pdf
Shared · 2.1 MB
TypePDF Document
AuthorsDevlin et al.
Year2018
Highlight · "masked LM"
Page 3 · Pink
来源Devlin 2018
ColorPink
Linked1 note
Note · Pre-training methods
Updated yesterday
Linked to4 sources
TagsNLP · pretraining
Backlinks3 notes
连接每个来源

PDFs, notes, highlights, and AI chats — all linked into one workspace.

结构化高亮

Page-level citations, colors, and tags organize every quote you save.

自定义笔记本

Group papers, ideas, and drafts your way — model any research workflow.

一张知识图谱

Follow the links between notes and sources. Search once, find everything.

伴你不断成长

Whatever stage of your work — from first read to final draft — The Work Docs grows with your workflow.

阅读

Multi-color highlights, margin notes, and live presence — built for deep reading sessions.

PDF
Attention is All You NeedVaswani et al. · p.42 / 156
ED
SK
+2
ED
Edward
Key insight — link to Section 5

AI 对话

Ask anything across your library. Streamed answers with page-level citations every time.

Compare results in Section 4?
THE WORK DOCS
Recall improved 18% over baselineVaswani p.42, but precision dropped slightlyDevlin p.7.
3 sources
Ask anything…

研究

Find related papers, cross-reference quotes, and build a citation web automatically.

SOURCE
cited by 248
92% match
78% match

写作

Block-style notes, AI suggestions, and drag-drop citations from every source you own.

AI
ED
SK
Saved
H1
Smith 2024
💡
H2
📊
fig-3.png
— Vaswani et al., p.42
Cite Devlin 2018Tab
342 words · 2 citationsSynced 2s ago

导出

One click to Markdown, Notion, BibTeX, Drive — formatting, links, and sources preserved.

NotesBib
Markdown.md
Ready
Notion互联工作空间
Synced
Drive.docx
Ready
BibTeX.bib
24 refs
Syncing 4 destinations…

学习

即将推出

Auto flashcards, quizzes, and spaced recall — track mastery and keep your streak.

7-day streak
3 / 12
QUESTION
What is multi-head attention?
Answered correctly · 92%
用户怎么说

与数千名满意的用户一起,改变文档工作方式

用户评价译自英语。

我一直在寻找这样的文档阅读器。它简单且非常有效,帮助我保持进度。使用之后,我的工作效率提高了很多

Megan Torres
Megan Torres
神经科学博士生 · UCLA

文档搜索真的很棒。正是这种流畅无阻的使用体验,让我每天都想用这个应用。它彻底改变了我的工作方式。

James Whitfield
James Whitfield
法学副教授 · NYU

面对堆积如山的研究论文,它简单得令人惊讶,却能大大帮助我集中注意力。真的很喜欢。

Camille Rousseau
Camille Rousseau
医学三年级学生 · McGill

使用 The Work Docs 后,我比以往完成了更多工作。所有研究论文都整理在一个地方,真是帮了大忙。

Arjun Mehta
Arjun Mehta
高级研究分析师 · McKinsey

面对大量案例文档和课程作业,这正是我需要的工具。PDF 高亮和批注功能为我节省了无数时间

Hannah Yoo
Hannah Yoo
计算机科学与法学预科 · Stanford University

这些 AI 功能真正带来了改变。与文档对话并即时获取摘要,已经改变了我的工作流程

Marcus Bell
Marcus Bell
战略顾问 · Deloitte

我每天都要审阅合同。The Work Docs 让我轻松浏览长篇 PDF。AI 摘要功能节省了大量时间

Nadia Osei
Nadia Osei
初级律师 · Clifford Chance

终于有一款不会碍手碍脚的应用了。界面简洁、速度快,搜索功能强大得令人惊喜。我推荐给了团队里的每位开发者。

Kenji Nakashima
Kenji Nakashima
软件工程师 · Stripe

我经常处理内容深奥的技术论文,这款应用真的帮了大忙。AI 对话帮助我更快理解复杂内容

Priya Nair
Priya Nair
机器学习工程师 · Google DeepMind

我们整个团队都用它进行尽职调查。协作从未如此轻松。这个工作空间让大家保持一致并专注工作

Lucas Ferreira
Lucas Ferreira
联合创始人兼首席执行官 · Fundo

我一直在寻找这样的文档阅读器。它简单且非常有效,帮助我保持进度。使用之后,我的工作效率提高了很多

Megan Torres
Megan Torres
神经科学博士生 · UCLA

文档搜索真的很棒。正是这种流畅无阻的使用体验,让我每天都想用这个应用。它彻底改变了我的工作方式。

James Whitfield
James Whitfield
法学副教授 · NYU

面对堆积如山的研究论文,它简单得令人惊讶,却能大大帮助我集中注意力。真的很喜欢。

Camille Rousseau
Camille Rousseau
医学三年级学生 · McGill

使用 The Work Docs 后,我比以往完成了更多工作。所有研究论文都整理在一个地方,真是帮了大忙。

Arjun Mehta
Arjun Mehta
高级研究分析师 · McKinsey

面对大量案例文档和课程作业,这正是我需要的工具。PDF 高亮和批注功能为我节省了无数时间

Hannah Yoo
Hannah Yoo
计算机科学与法学预科 · Stanford University

这些 AI 功能真正带来了改变。与文档对话并即时获取摘要,已经改变了我的工作流程

Marcus Bell
Marcus Bell
战略顾问 · Deloitte

我每天都要审阅合同。The Work Docs 让我轻松浏览长篇 PDF。AI 摘要功能节省了大量时间

Nadia Osei
Nadia Osei
初级律师 · Clifford Chance

终于有一款不会碍手碍脚的应用了。界面简洁、速度快,搜索功能强大得令人惊喜。我推荐给了团队里的每位开发者。

Kenji Nakashima
Kenji Nakashima
软件工程师 · Stripe

我经常处理内容深奥的技术论文,这款应用真的帮了大忙。AI 对话帮助我更快理解复杂内容

Priya Nair
Priya Nair
机器学习工程师 · Google DeepMind

我们整个团队都用它进行尽职调查。协作从未如此轻松。这个工作空间让大家保持一致并专注工作

Lucas Ferreira
Lucas Ferreira
联合创始人兼首席执行官 · Fundo
常见问题
为你解答疑问

免费试用
The Work Docs。

带上下一篇文献,还有你的好奇心。

开始研究