Understand any AI paper in plain English
Paste an arXiv link and ask anything. Answers quote the paper so you can check them, and any answer can be shared with a friend.
Or start with a landmark paper
- The Transformer: the "T" in ChatGPT. Attention Is All You Need Ask: What is attention, explained without math?
- GPT-3, and why a bigger model could suddenly learn from a few examples. Language Models are Few-Shot Learners Ask: What does "few-shot" mean, and why was it a surprise?
- How chatbots learned to be helpful instead of just predicting text. Training language models to follow instructions with human feedback Ask: How did human feedback change the model's behavior?
- Why asking an AI to "think step by step" works. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models Ask: Why does showing worked examples make the model reason better?
- Why AI labs keep building bigger models. Scaling Laws for Neural Language Models Ask: What are scaling laws, in everyday terms?
- The idea behind AI image generators: turning noise into pictures. Denoising Diffusion Probabilistic Models Ask: How can removing noise create a new image?
- RAG: letting an AI look things up before it answers. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks Ask: Why does looking things up reduce made-up answers?
- Teaching an AI to follow a written set of principles. Constitutional AI: Harmlessness from AI Feedback Ask: What is the "constitution" and how is it used in training?
Shortcut from arXiv
Add bot after arxiv in any arXiv paper URL.
Works with /abs/, /pdf/ and /html/ links.
No jargon required
Answers start in everyday words and explain each technical term. Switch to Technical when you want the details.
Check it yourself
Tap any quote in an answer to see that passage highlighted in the paper.
Share the a-ha
Send a friend a link to one answer. They land on it and can keep asking in their own copy.