The viability of open-weight models shouldn't be a concern. If a model provides genuine business value, the entire economic ecosystem—from chip providers to cloud infrastructure—will naturally orient itself to create a supportive and profitable supply chain around it, ensuring its sustainability and growth.
An unintended consequence of stringent safety measures on American frontier models is that they often refuse security-related queries. This perversely pushes cybersecurity professionals to use less-restricted Chinese open models for essential tasks like vulnerability analysis, creating a strange competitive and security dynamic.
As open-weight models close the performance gap, the defensibility for labs like OpenAI is no longer just their model's intelligence. Their lasting moat lies in user-facing products or 'harnesses' like ChatGPT or Claude, which offer a sticky, integrated experience that is harder to commoditize than the underlying API.
A critical imbalance exists in AI development: Chinese models can distill capabilities from top American models with few repercussions. Meanwhile, American open-weight startups face significant legal uncertainty for doing the same, creating an uneven playing field that favors foreign competitors in the global AI race.
The debate over distilling from other AI models is becoming moot. The internet is now so saturated with AI-generated content ('AI slop') that any new model trained on web data is already, by default, being trained on the outputs of its predecessors. Pure 'human data' is a dwindling resource.