Taste: A Physical AI Data Happy Hour

Unusual Ventures & HumanSignal invite you to "Taste: A Physical AI Data Happy Hour"

Topic: The data machines can't make for themselves

Unusual Ventures and HumanSignal invite you to an evening of drinks, tastings, and real conversation with the people building physical AI, from frontier labs to robotics startups to the teams quietly running the data programs behind them.

Here's the premise. The hardest data in AI right now isn't online and can't be scraped. It has to be created in the physical world, from a first-person point of view, with human judgment in the loop. Teleoperation is hitting its ceiling. Simulation has its gap. The fastest-moving frontier is human and egocentric data, and the open question is how you create it, label it, and make it transfer.

So we're building the night around that idea: taste, judgment, and the signal only people can produce. Expect blind pisco and cocktail tastings, a few sensory challenges, and a DJ running human-made versus AI-generated tracks for the room to call. The games are the thesis: when can a model tell the difference, and when can only a person?

No slides, no pitch. Just the peers wrestling with the same problem, and a good excuse to compare notes. Come if you're leading AI work, building physical AI products, or running the data behind it.

Food, drinks, and tastings provided. Bring a plus-one.

Conversation we expect to get into:

What actually comes after teleoperation

Egocentric and first-person data: hype, or the real unlock

The human-to-robot embodiment gap, and what it takes to close it

Why "more data" stopped being the answer

The proprietary real-world datasets becoming competitive moats