Box CEO Aaron Levie argues open-weight AI is not a zero-sum threat to closed models. Instead, it expands the total number of AI use cases and creates competitive pressure that pushes frontier labs like OpenAI and Anthropic to innovate more rapidly, ultimately benefiting the entire ecosystem.
It's a tenuous and difficult argument to suggest that training an AI on the public internet is acceptable, but training an AI on the output of another AI is not. Aaron Levie notes the logical inconsistency, suggesting the ethical line is blurry at best, especially when the original model provider is paid for the API usage during distillation.
An attempt to completely block China from US AI advancements is unlikely to succeed. Instead, it will catalyze China to invest more heavily in building its own independent hardware and software stack. This could leave the US with expensive, isolated AI while the rest of the world adopts cheaper, Chinese-developed alternatives.
Contrary to the fear of job displacement, AI enables engineering teams to tackle projects that were previously too complex (multi-year) or too small (not worth the effort). This expands a company's ambition and overall product roadmap, increasing the demand for skilled engineers to pursue these newly unlocked opportunities.
Aaron Levie suggests labs like OpenAI could become more competitive by consistently releasing open-source versions of their prior-generation models. This would keep more use cases within their ecosystem, cater to sovereign and fine-tuning needs, and ultimately drive more revenue back to them as they power the inference for these open models.
As frontier models from different labs constantly leapfrog each other, enterprises face 'analysis paralysis.' The most value will be created by an 'applied AI layer' that acts as a model router. This layer will abstract the complexity, select the best model for a given task, and prevent lock-in to a single provider like OpenAI or Google.
Frontier models like Fable can be too conservative, frequently 'falling back' to less capable versions when faced with sensitive or complex queries, such as in biosciences or security. This unreliability makes the most advanced models untenable for critical enterprise use cases, highlighting a fundamental tension between capability and lockdown.