Your virtual AI laboratory

Understand AI research papers by interacting with them

Stop reading complex PDFs. PaperLab turns papers into live simulations you can experiment with — drag a slider, watch attention flow, break gradient descent — and an AI tutor explains what changed.

Try:Attention Is All You Need YOLO GPT-3 ViT

See the Transformer paper explained

Interactive learning series

A guided path: basics → advanced

Ten papers that built modern AI, in order. Each 'Ready' stop is a full interactive deep-dive — live simulators, clickable architecture, math with analogies, a quiz, and a mini course.

Deep Learning Foundations

01 · 2012Ready

AlexNet: ImageNet Classification with Deep CNNs

The Big Bang of modern AI — GPU-trained convolutions crush ImageNet by 10 points.

02 · 2014Ready

Generative Adversarial Networks

Forger vs detective — two live networks duel until fakes match reality.

03 · 2015Ready

Deep Residual Learning (ResNet)

Skip connections fix the depth mystery — watch two real networks race to prove it.

04 · 2015Ready

You Only Look Once (YOLO)

Real-time detection in a single glance — tune confidence and NMS live.

The Transformer Era

05 · 2017Ready

Attention Is All You Need

The paper that started modern AI — attention simulators, math, and a mini course.

06 · 2018Ready

BERT: Bidirectional Transformers

Fill-in-the-blank with X-ray vision — play the masking game that broke 11 records.

07 · 2020Ready

GPT-3: Few-Shot Learners

Scale the decoder far enough and it learns tasks from your prompt alone.

08 · 2020Ready

An Image is Worth 16x16 Words (ViT)

Transformers leave language and beat CNNs at their own game — cut a photo into 'words' and watch attention read it.

III

Alignment & Efficiency

09 · 2020Ready

Learning to Summarize from Human Feedback (RLHF)

The recipe behind ChatGPT — feel the reward-vs-leash trade-off yourself.

10 · 2021Ready

LoRA: Low-Rank Adaptation

Customization 10,000× cheaper — watch a real SVD rebuild a full update from rank 3.

Try it right now

This is what learning feels like here

A live gradient descent lab — crank the learning rate past 0.9 and watch training explode. That intuition is worth ten textbook pages.

Gradient Descent Lab

Live

How neural networks actually learn

step 0/26w = -4.200loss = 7.152

Press play to drop the ball! It always rolls downhill toward lower loss — that is how a neural network learns. Can you get it to the flag?

Analogy: A ball rolling down a foggy mountain: it can only feel the slope right under its feet, and the learning rate is the size of each hop.

Start here

Landmark papers, ready to explore

The papers that defined modern AI — each one an interactive learning experience.

Interactive180k citations

Attention Is All You Need

Ashish Vaswani et al. · NeurIPS 2017 · 2017

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. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles, by over 2 BLEU.

Start learning

Interactive110k citations

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Jacob Devlin et al. · NAACL 2019 · 2018

We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks. BERT obtains new state-of-the-art results on eleven natural language processing tasks.

Start learning

Interactive170k citations

ImageNet Classification with Deep Convolutional Neural Networks (AlexNet)

Alex Krizhevsky et al. · NeurIPS 2012 · 2012

We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

Start learning

Why PaperLab

Built for understanding, not skimming

A summarizer compresses a paper. PaperLab teaches it — the way a great mentor would sit next to you and walk through it.

Three explanation levels

Every paper explained for a 15-year-old, a developer, and a researcher. Switch depth with one tab.

Interactive diagrams

Architectures you can click. Select any block to see what it does, with concrete examples — not static figures.

Math that makes sense

Every equation comes with its plain meaning and a real-world analogy. Formula → intuition → analogy, always.

Chat with the paper

Ask anything — “what's the main innovation?”, “why √d_k?” — and get answers grounded in the paper itself.

Papers become courses

One click turns a paper into a mini course with lessons, visuals, and quizzes that check real understanding.

Runnable code

See the core idea as clean, minimal PyTorch you can actually run — the fastest path from theory to intuition.