synthetic data
Artificially generated data used to train AI models, supplementing or replacing real-world data.
As defined in et_982
Other definitions
Artificially generated training data (as opposed to real-world data), such as procedurally generated examples or novel compositions, which can be used to address data scarcity in AI training
Bret Taylor: Facebook, Salesforce, OpenAI, and SierraTraining data generated by AI models rather than collected from the real world.
Gavin Baker - AI, Semiconductors, and the Robotic Frontier - [Invest Like the Best, EP.385]
Where it comes up · 34
- Sir Demis Hassabis on The Future of Knowledge | Institute for Advanced Study7 claims
- Capital, Compute & AI Scaling3 claims
- Dwarkesh Patel Podcast - The avalanche that hit math is coming for AI development3 claims
- Gavin Baker - AI, Semiconductors, and the Robotic Frontier - [Invest Like the Best, EP.385]3 claims
- Vlad Tenev - Navigating Robinhood's Evolution - [Invest Like the Best, EP.384]3 claims
- AI's Trillion-Dollar Opportunity: Sequoia AI Ascent 2025 Keynote3 claims
- Demis Hassabis — Scaling, superhuman AIs, AlphaZero atop LLMs, AlphaFold2 claims
- Fully autonomous robots are much closer than you think – Sergey Levine2 claims
and 26 more
Most important claims · 10 of 59Search all claims for “synthetic data” →
- Self-generated synthetic experience lets a system rehearse and consider counterfactuals but cannot teach it more about the world; world information must be injected from somewhere, and the most powerful way to generate synthetic experience is from a really good model that itself learned by experiencing the world.Fully autonomous robots are much closer than you think – Sergey Levine
- Simulation has moved beyond games to first-principles physics and synthetic data, enabling systems to train on generated edge cases; this flips training value because prior video-game-based training failed precisely on the rare fat-tail events.Josh Wolfe and Tony Thomas - The Past, Present, and Future of Defense – [Invest Like the Best, EP. 216]
- Synthetic data must be validated against real data to ensure the distribution of synthetic outputs matches the real distribution, otherwise the model risks compounding bias or error.Sir Demis Hassabis on The Future of Knowledge | Institute for Advanced Study
- Synthetic data and simulation are important mechanisms to address data scarcity, with simulation being higher quality than arbitrary synthetic data because it encodes real-world constraints (like physics), limiting the permutations of invalid data that could be generated.Bret Taylor: Facebook, Salesforce, OpenAI, and Sierra
- Pre-training scaling is slowing down because 9-10 orders of magnitude of progress since AlexNet means most low-hanging fruit has been picked, requiring research breakthroughs like reasoning, synthetic data, tool use, and AI scaffolding to continue intelligence scaling.AI's Trillion-Dollar Opportunity: Sequoia AI Ascent 2025 Keynote
- AlphaFold required approximately 150,000 experimentally determined protein structures from the Protein Data Bank as initial training data, supplemented with synthetic data generated by an earlier version of AlphaFold that predicted ~1 million structures and was then filtered to ~300,000 of the most accurate predictions to create a refined training set.Sir Demis Hassabis on The Future of Knowledge | Institute for Advanced Study
- You can generate as much synthetic data as you want from games because the system can play against itself, and games also have very clear metrics—win conditions and score maximization—which are useful from an AI perspective for optimization.Sir Demis Hassabis on The Future of Knowledge | Institute for Advanced Study
- Competition between model companies will stay fierce, informed by the observation that previous supposed flywheels (chat user engagement, synthetic data) have done little to diminish competition; the big three rotate around the podium roughly monthly, suggesting some force—talent poaching, rumor mills, or reverse engineering—has neutralized any runaway single-lab advantage.An audio version of my blog post, Thoughts on AI progress (Dec 2025)
- Video and audio data remain far from exhausted, so new modalities could produce capability leaps in human interaction that are as disorienting as image and video generation advances have been.Capital, Compute & AI Scaling
- Privacy technologies such as homomorphic encryption, differential privacy, and synthetic data now make it possible to join datasets, including medical and IRS records, without any party seeing the underlying data, which was not feasible when SafeGraph started about six years earlier.Auren Hoffman - A Deep Dive on Data - [Invest Like the Best, EP.320]