AGI Upon Us?

In conclusion AI Agents do not yet know do to work on long running tasks because they have not been trained on long running tasks. The same logic can be applied to transfer learning in robotics. 

Astra is pending release and ive read a lot of things but it gets 99% on the agi benchmark and its long running. I wonder if you can make long running ai with harness and the harness helps it get out of reasoning loops. That being said I believe that intelligence is not only based on long running tasks or AI lifecycle orchestration but also things like context and tokens per second all matter. In order to advance further we need to build machines that can start building themselves. I need to view Terrance tell speaking, but I saw a blog post that he was advising mathematicians not use AI to solve math and I think that's fundamentally wrong I think the next new generation of mathematicians doctors engineers are all going to build machines that can build other machines in their field and those are going to be the next great companies.