JULY 9, AI Startups Decagon CEO Jesse Zhang recently published an article entitled "Eoone is wrong about open source AI in the environment" which presents an anti-intuitive assertion:
IN THE CORPORATE AI MARKETOpen Source ModelThere is no competitive relationship with closed source models, but rather two different stages in the life cycle of the same product。
Zhang, for example, is a company: Decagon currently has approximately 90% work loads running on open source models, but this is not because of cost considerations, but because client service scenarios are extremely sensitive to delays and require small fast models that are heavily fine-tuned, and the front-line closed source model does not meet this demand in terms of fine-tuning flexibility。
He summed up the path as "Close Source Leading Discovery, Open Source Taking over Production" — an enterprise, while exploring an entirely new example, tends to opt for the most capable front-line model; until the example is fully mature, the input distribution is stable, the failure pattern is clear, the task is moved to a light open source model that has been fine-tuned。