Josh Gao says LLM pre-training data was driven by labs through web crawlers and never became a large market. In robotics, he says pre-training is already a large market because there is no free internet corpus; Mecka must create, process, and license physical-world data.
At founding, Josh Gao says robot data was mostly teleoperation, which was hard to scale. He bet that first-person human video could close the vision gap and that humans and humanoids look alike, letting ordinary people contribute like a Tesla-style data fleet.
Josh Gao says wrist cameras are a hacky substitute for measuring contact force. Mecka is building gloves that capture how much force daily tasks require because no off-the-shelf wearable reliably records force and turns it into ready robot-training datasets.
Josh Gao says Mecka's data engine is far more painful than people expect: it had to design custom hardware, distribute it to thousands of homes and businesses, install 100-gigabyte upload lines, and process hundreds of petabytes while enforcing quality gates.
Josh Gao says fraud is an underestimated problem in data collection. Creative people recorded POV YouTube videos of working at McDonald's and tried to submit them as their own contributions, forcing Mecka to build quality gates and design better incentives. He adds that you also have to be where collection happens.
Josh Gao says large-scale robot factories doing repetitive warehouse tasks revealed a threshold of diversity. Beyond some point, adding more robot data of the same kind oversaturates the need, so Mecka tracks job and industry heat maps and adapts quality heuristics as edge cases emerge.
Josh Gao says the industry is converging on roughly a million hours of human egocentric data, tens of thousands of hours of UMI data, and a few hours of last-mile teleoperation or in-domain demonstrations. He stresses that tactile and simulation data have not even entered the mix yet.
Josh Gao contrasts AlphaGo, where self-play could use the complete rules of Go, with robotics, where the rules are the entire physical world. Because no simulator can recreate that detail today, egocentric data is used to ground basic physics and help build the world simulator.
Josh Gao says deployment is not a binary switch. Initial robots will hill-climb constrained, goal-oriented jobs with existing SOPs, such as packaging lines and industrial cleaning. Homes come later because they require extreme diligence, like making sure a robot does not fall on a baby.
Josh Gao expects Mecka to look different as robots deploy: not just egocentric data, but a platform that collects domain data, post-trains models, and white-gloves enterprises. He compares the future role to a modern robotics systems integrator whose installed bot keeps improving with data.
Josh Gao says scaling the current data recipe by orders of magnitude will solve many constrained tasks. But a drop-in robot that handles any task amid real-world complexity, like weather, is still far off; he expects tactile, simulation, and even smell-like modalities to enter omni-model training.
Josh Gao says it is expected that models like GPT-6 Astra generalize to robot tasks because robotics data and action trajectories are being added to pre-training mixes of large foundation models. He still sees room for specialized robot models that reach four-nines reliability in narrow tasks.
Josh Gao says Mecka was an early contributor to Egoverse, a consortium with researchers who believed in scaling human data for robotics. It shared action trajectories, computer vision post-processing, quality gates, and platform tooling to lift the whole industry because robotics is far behind.
