RAG
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Retrieve documents before generating, plugging external knowledge into a language model.
No deep dive on this stop yet; the arXiv paper is linked below.
Time Machine · AI Search
15 stops, ordered by the month each paper first appeared; pick a stop for the key points, and read our deep dive inline where there is one.
← → stops · Esc closes
WebGPT
WebGPT: Browser-assisted question-answering with human feedback
The model drives a browser to search and cite, trained with human feedback.
No deep dive on this stop yet; the arXiv paper is linked below.
ReAct
ReAct: Synergizing Reasoning and Acting in Language Models
Interleaves reasoning with search calls, setting up the basic agent loop.
No deep dive on this stop yet; the arXiv paper is linked below.
IRCoT
Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions
In multi-hop QA, retrieves again after every reasoning step.
No deep dive on this stop yet; the arXiv paper is linked below.
FLARE
Active Retrieval Augmented Generation
Retrieves only when the model is unsure.
No deep dive on this stop yet; the arXiv paper is linked below.
Self-RAG
Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection
Reflection tokens let the model decide whether to retrieve and judge the evidence.
No deep dive on this stop yet; the arXiv paper is linked below.
Search-o1
Search-o1: Agentic Search-Enhanced Large Reasoning Models
A reasoning model searches on its own in the middle of a long chain of thought.
No deep dive on this stop yet; the arXiv paper is linked below.
Search-R1
Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning
Trains multi-turn search with outcome-only reinforcement learning.
No deep dive on this stop yet; the arXiv paper is linked below.
DeepResearcher
DeepResearcher: Scaling Deep Research via Reinforcement Learning in Real-world Environments
Runs RL in the real web environment instead of a local corpus.
No deep dive on this stop yet; the arXiv paper is linked below.
WebThinker
WebThinker: Empowering Large Reasoning Models with Deep Research Capability
A reasoning model searches, browses and writes a report as it goes.
No deep dive on this stop yet; the arXiv paper is linked below.
ZeroSearch
ZeroSearch: Incentivize the Search Capability of LLMs without Searching
Simulates the search engine with an LLM, saving the cost of real search calls during RL.
No deep dive on this stop yet; the arXiv paper is linked below.
R3-RAG
R3-RAG: Learning Step-by-Step Reasoning and Retrieval for LLMs via Reinforcement Learning
Learns with RL when to retrieve, what to retrieve and when to stop.
LatentRAG
LatentRAG: Latent Reasoning and Retrieval for Efficient Agentic RAG
Moves multi-step reasoning into latent space to cut agentic RAG latency.
SIRA
Superintelligent Retrieval Agent: The Next Frontier of Agentic Retrieval
Argues retrieval agents need real control over retrieval rather than many rounds plus long context.
AGR
AGR: Agentic Generative Retrieval with Tool-Augmented Search
Generative retrieval that calls tools to look up features only when needed while generating codes.