OpenAI teams up with Scale for enterprise fine-tuning

Fine-tuning a frontier model sounds great until you realize your enterprise data is a chaotic mess.

OpenAI just teamed up with Scale, designating the data infrastructure company as its preferred partner to help enterprises customize its top models. The move follows the recent rollout of fine-tuning for GPT-3.5 Turbo, with GPT-4 fine-tuning scheduled to land this fall.

Outsourcing the heavy lifting

Deploying custom models into production requires rigorous data enrichment and evaluation, which most internal developer teams simply are not equipped to handle.

Scale customers can now fine-tune OpenAI models just as they would directly, but with Scale's Data Engine and enterprise tooling layered on top. OpenAI also reiterated its privacy commitment: all data sent through the fine-tuning API belongs to the customer and will never be used to train future base models. FinTech startup Brex has already used the pipeline to fine-tune GPT-3.5.

Why it matters

OpenAI wants enterprise revenue, but it does not want to become a bespoke IT services consultancy. Partnering with Scale solves this problem neatly, outsourcing the messy work of dataset cleaning and model evaluation to a team that specializes in it.

For enterprise teams, this significantly lowers the barrier to getting tailored model weights into production without hiring an army of data engineers. That said, it introduces another vendor into your core AI stack, meaning higher overall implementation costs and deeper platform dependency.