Who could be in command of a powerful AGI? Couldn't anyone with temporary access to the system have it generate a game plan for how that person can steal the agi?
Powerful LLM, you mean. AGI isn't there, and with this approach won't be - MS may have all the models, but if it still misses people who are needed to make better models, it's a delusion in my book. Let Satya think he got it covered.
Can I apply with a product? What is research in your terms? What happens after the grant is over?
I was building an open-source alternative to Perplexity: https://omniplex.ai
Now I am adding custom agents actions to performs tasks on behalf of the user requests.
You can send a regular email, I'm happy to chat about it.
I consider it outside the intended scope of the grant because your project (as I understand it) is about applying existing techniques to build a product. What I'm looking for is work that tries to understand or improve on them, or create new ones.