Linear Tech π Recruitment: AI Data Startups—Can They Achieve Big Results? | Linear Event

See you in Beijing on December 5.

"Linear Tech π" is a closed-door frontier tech salon launched by Linear Capital, creating a high-quality, substance-rich exchange space for founders and practitioners through its most hardcore discussion format.

Each session focuses on one industry pain point, bringing together top technical founders to cut through the noise and discuss possible solutions to the industry's real problems.

For the 4th "Linear Tech π" closed-door event, we'll be in Beijing on December 5, hoping to bring together the people who produce data in AI, the people who use it, and the people who study models — to talk about the business of AI data startups. We welcome you to sign up and join Linear: start with technology, become your own school.

In the past, we mostly understood AI's progress through models: parameters, architectures, training methods, and the ever-expanding capability frontier. But behind the models, other questions are becoming concrete: Which experiences haven't been digitized yet? Which judgments are hard for models to learn? Which real-world feedback has yet to enter the training process?

These questions all point to AI data.

Over the past few months, in frequent conversations with AI founders and researchers, we've repeatedly discussed a shift: judgments about data's value are deepening into how it participates in the formation of AI capabilities. An expert demonstration, a complete task trajectory, a correction after failure, or an environment that can sustain interaction and verify results — each carries different information. Their value needs to be understood in the context of specific training goals and application scenarios.

This has also created new problems and opportunities in the AI data industry. Some are rushing in wildly to strike gold, while others are deeply skeptical of this kind of business. But for this frantically advancing AI world, everything that exists is reasonable.

So through this new session of "Linear Tech π," we want to invite friends working on data-related work in this AI wave to a small-scale closed-door exchange.

To ensure the quality of attendees, we've already invited co-founders from some top industry companies and researcher friends from major AI model labs — we can't claim the lineup is absolutely heavyweight, but it's absolutely frontline. We want to start from these questions:

▍What data, exactly, are models missing right now?

From pre-training to post-training, from single-turn answers to long-horizon tasks — how is the shifting capability bottleneck changing data demand?

Of expert knowledge, operational processes, environmental feedback, and evaluation standards — which have become real procurement needs, and which are still being explored? How do you judge the gain a piece of data brings, and how long that gain can last?

▍When data can be generated, where is the value of real experience?

Synthetic data and interactive environments have expanded the sources of training material — and made the value of "real" more worth scrutinizing.

Which experiences can be effectively synthesized? Which still depend on the constraints, failures, and feedback of real business? How do you confirm that generated data brings new learning signals rather than repeating existing capabilities?

▍What kind of data capability can support growing into the next big company?

From project delivery to continuous supply, from collection and labeling to deep involvement in training and evaluation — the boundaries of data companies are being re-discussed.

Where will long-term value accumulate: a scarce expert network, a stable scenario entry point, an understanding of model needs, or the ability to continuously validate data effectiveness? As models and client needs keep changing, what can be accumulated?

We hope to meet people like you:

  • AI data founders and core team members: exploring expert data, synthetic data, multimodal data, embodied data, or RL environments
  • Model researchers and training engineers: directly involved in pre-training, post-training, evaluation, or building data and environment pipelines
  • Big tech data teams and buyers: actually involved in defining data needs, selecting vendors, quality acceptance, and delivery

To ensure the quality of on-site exchange, this closed-door event is open only to AI data-related founders and researchers, and we will strictly control the number of participants. We hope every friend comes with concrete questions and practice, and leaves with new inspiration.

We believe that when it comes to AI data, many important questions remain unsettled. On the evening of Saturday, December 5, we look forward to making these questions more concrete together with those of you exploring on the front lines.