I'm curious about the potential for the big AI companies. If I understand right, they have already admitted that the current models cannot be used safely, even when used by their employees (who presumably know them best), and even with extreme care to "sandbox" them.
If this is true, it seems unlikely to me that future, more powerful models, can be distributed to random end-users over the internet.
Given this, their business case could be limited to the distributing the weaker models over the internet, and potentially allowing bug businesses to periodically use an "air gapped" data center to solve specific problems. But, for the latter, I have to think most big business would go for the free open source models over the forking over $$$ for chat gpt / anthropic / .... models with maybe only a slight advantage.
Is this accounted for when they estimate future revenues? Is there any way these companies can be worth multi trillion USD valuations given these limitations (and extreme debt)? I'm no expert, but seems like these companies are almost worthless in such an environment.