Some reminders for everyone confidently shouting from the rooftops that the US gov is about to ban open source models:
- The computing industry in America has accounted for roughly 1/3 of ALL real economic growth in the country since 2000. Instead of ~2.1% real growth, America's real GDP growth would be 1.5% annually, roughly matching the UK.
- Computers were not originally invented in the Sates. The first computers were built in... drum roll please... the UK!
- The UK gov mandated that after WW2, the work being produced at Bletchley Park would either be destroyed or classified under the Official Secrets Act. Engineers were not allowed to discuss their work, and industry's 3+ year head start over other countries quickly evaporated.
- Because the US gov took a more open-source friendly approach to early computing, the industry instead developed on America's west coast. Civilian teams like Bell Labs and IBM turned this esoteric scientific instrument into a commercially valuable tool, and sold it to the masses.
- In the 60s when DARPA invented the internet, instead of keeping it private (despite enormous military comms benefits), they open sourced the tech and protocols that made it work. They did this to ensure that the exploration and development of this frontier happened in America, and that a broad range of American businesses would benefit from this technology shift; not just one or two companies getting all of the spoils from these discoveries.
- The right to use models whose weights they own and control cannot and will not be denied to the American corporation. It would cut them off at the knees and limit their ability to compete in global AI markets. It would force them to hand precious IP to a small number of companies which have grown 1,000,000x in a matter of years, at the expense of the rest of the economy - made up of businesses whose knowhow must suddenly be vacuumed up by these frontier labs if they want to participate in this new frontier.
- Thanks to the modern state of training and distillation, the cat is out of the bag on data provenance, and it will be impossible for anyone to prove a model whose weights they possess was trained purely from "good data" (or distilling "good models"). There is no world in which companies will be forced to buy weights from specialized firms vs being able to rely on open source weights that are freely available.
- In a moment of market headwinds, as capital markets grow colder and token margin compression begins, a small number of companies will attempt to leverage fear-mongering and reg capture to close the American AI market. They will argue that this time is different, that THIS paradigm shift is more important than the last one. These arguments are based on fiction, and are a fairytale concocted to help you forget one simple fact:
- The bedrock of the entire American computing industry rests on the belief that the market wins when foundational technologies are allowed to be free.