How do I know my synthetic training dataset will not silently collapse my model?
Opportunity
Fine-tuning and instruction-tuning pipelines increasingly rely on LLM-generated synthetic data, and when synthetic content crosses roughly 60 percent of the training mix, or when the same base model both generates the data and receives the fine-tune, quality degrades within two to three training cycles. Standard held-out loss curves do not reveal the collapse until the model ships and fails on real inputs. Research from ICLR 2025 showed that one recursive sample per thousand is enough to trigger collapse under data-replacement conditions, and a 2026 paper modeled cross-contamination spreading through shared public corpora like an epidemic. No production tooling exists to measure the synthetic fraction of a dataset, bound its generational distance from a source model, or issue a pre-training warning before a run starts.
Why it matters
Fine-tuning pipelines that cannot measure their own synthetic exposure rate are flying blind on model quality.
How I score the opportunity
The Opportunity Score is my own read, not a measurement: how much it hurts, how often it bites, and how little exists to solve it today. Higher means I think it is more worth building.
How much pain it causes when it shows up.
How often people actually run into it.
How little good tooling exists for it today.
More problems worth solving
Why does every AI app forget me the moment I close the tab?
AIWhy is learning a new field still gated by knowing what to ask?
AIWhy can a non-expert not verify what an AI just told them?
AIWhy do we test models on benchmarks but ship them on vibes?
AIWhy do AI agents have no memory of their own mistakes?
AIWhy can't I audit what a model was actually trained on?