jeetu patel

The US must maintain many open-source alternatives for frontier-class models and SLMs to avoid falling behind Chinese open-source models, which are gaining momentum despite chip disadvantages. Open-source models also provide cost and control benefits for enterprises running models in their own data centers.

If open-source models disrupt closed-source models, monetization will shift to the application layer and infrastructure layer, as model providers cut costs with custom chips and move up the stack. The $3-5B cost to train frontier models requires sustained economic incentives that open-source disruption could undermine.

The debate over whether open-source AI models help defenders or attackers more is overly polarized. Dario Amodei's concerns about guardrails are credible, but open-source models also enable security applications like vulnerability detection, and the US needs a nuanced policy approach rather than dogmatic positions.

If the US bans Chinese open-source models, the cost of tokens and AI tasks could rise relative to the rest of the world, potentially putting the US at a long-term disadvantage. However, keeping them open creates security risks if models are trained to cause harm on US soil.