The fear that open source AI is dangerous is flawed. History shows open platforms like Linux were far safer than closed ones like Windows. A broad community can identify and fix safety issues, like reward hacking, faster than a single proprietary company focused on benchmarks and profits.
The focus on China exploiting US open source AI is a distraction. The more significant national security threat is a single, monopolistic US AI company becoming powerful enough to defy its own government's interests, as Anthropic has demonstrated by refusing to work with the military.
Application developers building on proprietary models face existential risk. As soon as an app category proves successful, the platform owner is incentivized to enter that market, subsidize their own version, and use pricing or API access to put the original developer out of business, making open source a safer bet.
The narrative of a zero-sum battle between AI giants is misleading because the market is in its infancy. With less than 3% penetration, there is enormous room for growth for all players. New model releases currently lift the entire ecosystem rather than stealing market share from competitors.
The most viable business model for open source AI isn't selling high-premium access to a general model. Instead, it involves creating specialized, post-trained smaller models for specific B2B tasks. These can be cheaper, faster, and more effective than large models, resembling Palantir's tailored, high-value service approach.
Proprietary labs argue against 'distillation' (using their model outputs for training) while they have built their own models on vast amounts of copyrighted data. This opposition is an anti-competitive tactic, as model outputs are not copyrightable and distillation helps smaller, open players to compete.