I just finished watching the latest issue of the All In Podcast, summarizing their thoughts on AI sovereignty, open-source models, and future employment 1. Palantir collaborates with NVIDIA to directly challenge OpenAI and Anthropic Palantir and Nvidia have developed a "sovereign AI operating system" that customizes defense grade AI for the US government using open-source models, with hardware, data, and model weights all owned by the government itself. Palantir CEO Alex Karp directly criticized in an interview, saying that if companies and governments hand over their core data and IP to closed source modeling companies, it is equivalent to handing over their future sovereignty to others. David Sacks also added a point of view, saying that the current tactics used by Frontier Labs are very similar to those of Microsoft and Google in the past, where they first monopolize the underlying models and then drill into upstream product lines, directly competing with their ecological customers for business (such as Anthropic's Claude Code and Claude Design, which is equivalent to competing with customers such as cursor and Figma for jobs). Enterprise customers are now starting to guard against this move. 2. Open source models really save money, Chamath directly shares data Chamath shared his own team's actual test data on enterprise code migration: using closed source Claude Opus directly has the highest cost; By using our own framework to package open source models, the cost has been reduced by more than 16 times, although it runs about 3 times slower. The conclusion is very direct. Currently, feeding data to closed source model companies is equivalent to helping competitors train their products, which is not very rational. Friedberg also mentioned that Anthropic had previously sought out life science giants to exchange experimental data for priority usage rights, but was widely rejected. Everyone realized that data is the real moat. He predicts that in the future, AI architecture will move towards the direction of "universal big models+enterprise private cloud training+local inference deployment", and enterprises will eventually run their own proprietary models forked locally to maintain "intelligent sovereignty". Will AI cause massive unemployment? Jason is a staunch technology disruptor who believes that positions such as customer service, data entry, autonomous driving, and factory logistics will be heavily replaced within 5 to 10 years. But Sacks, Friedberg, and Chamath are basically on opposite sides. Sacks cited a joint study by Ramp and Relic Labs on 21000 companies, which showed that companies heavily using AI saw an average increase of 10% in total employee numbers and 12% in junior positions within two years, while companies without AI remained stagnant. Friedberg is more direct, stating that "AI leads to mass unemployment" is basically a false proposition created by the media. Currently, AI is more like a somewhat clumsy efficiency tool that will bring about job restructuring, but in the long run, it will create more high-value jobs that require "human-machine collaboration".