Every generative model is dreaming.
A video model invents motion that has never existed. A world model rolls reality forward so a robot can rehearse what happens next. We see what these models make.
Reverie is about what those dreams are made on.
This fall, LanceDB is bringing together 200 AI researchers, engineers, and technical leaders for Reverie, an exclusive summit on the work that decides what the next generation of models learns, shown by the people who do it firsthand.
The frontier has moved below the model
The next breakthroughs are sitting in the data, waiting for the teams that know how to look.
The videos selected from billions of frames. The strange edge cases recovered from a sensor archive. The captions, embeddings, trajectories, quality scores, and physical properties derived from raw data. The evaluation sets designed to reveal whether a model understands the world or has simply learned to imitate its surface.
For many frontier teams, this is where model quality is made. The next major improvement may not come from adding another order of magnitude of compute. It may come from running an order of magnitude more useful experiments with the compute you have.
Reverie puts that work on stage.
What you will hear
Reverie goes inside the data and research systems behind frontier AI. Speakers will share how teams curate training sets from petabytes of video, images, audio, and sensor data, evaluate whether models understand motion and physics, and find the rare examples that lead to better model behavior.
The program will also explore how researchers run feature experiments across massive datasets without rebuilding everything from scratch, keep evaluation sets trustworthy, and prepare for agents to operate more of the research loop. Expect technical systems, hard tradeoffs, failed experiments, and real numbers.
Meet the first set of speakers
The initial Reverie lineup includes researchers and engineers from the teams advancing generative video, physical AI, multimodal search, and world models:
- Max Li of NVIDIA Cosmos
- Ethan Rosenthal of Runway
- Enwei Jiao of Luma AI
- Shiyan Xu of Applied Intuition
- Bo Tian of XPENG
- John Trenkle of Tubi
- Jing Chen He and Miao Wang of Adobe
- Jan van der Vegt of Exa
More speakers will be announced soon.
Together, they represent different parts of the emerging AI research stack, from models that simulate visual worlds to the retrieval systems agents use to investigate them.
What dreams are made on
The models get the attention. Reverie is about the work that makes them possible.
The raw artifacts. The derived features. The failed experiments. The evaluation sets. The data engines. The researchers and engineers who turn an idea into evidence.
Because the next order of magnitude in AI will come from larger models and from the systems that help us understand what to train them on next.
Reverie takes place November 5 with limited capacity. Request an invitation to join the researchers and engineers showing what the next generation of AI is made on.
