Small Fine-tuned Models as an Alternative to Giant Generic Ones

Giant models are costly and raise latency and privacy problems. YC wants smaller fine-tuned models, and tools to build them, that match results at a fraction of the cost.

YC grants that giant generic models are impressive, but notes they are very costly and often bring latency and privacy problems. Smaller open-source models like Llama2 and Mistral have shown that, when fine-tuned with the right data, they can give comparable results at a fraction of the cost.

Why now: as new hardware goes into phones and laptops, running these models on the device itself is becoming more feasible, which YC expects to unlock many new use cases.

The ask: companies developing or fine-tuning specialized models, or building tools that make such models easier to build.

Read the full request on YC →