temp2525

Demis is toothless, it will not work because you cannot trust people to self censor when their future work at the committee will be at stake. you cannot setup reasonable bonds in unlimited problem space because there is no example or metric to judge against. possible solution would be to limit higher priority access to agi to only people that demonstrated integrity in their decisions. all other metrics can be easily forged, while integrity can be publicly verified by agi itself. it would probably separate agi from politics while allowing it to do governance - agi will do its lowly ai tasks while human meatbag will do politics.

amd ai apus are incapable of running models fast enough for productive work, you need over 100t/s for text mode and above 180t/s for image gen/ image recognition to not wait minutes. you cannot run claw and leave it - you would need to wait multile hours for small project like chatbot+landing page. price wise hw was not worth it even when it cost under 2k$, nowadays it cost even more.

just a perspective(you can re check results in youtube reviews): ryzen ai runs most moe models under 60t/s while nvidia gpu can runt them at 100t/s. and you preferably need deep models for code, they would run at 40t/s.