Raw LLM Responses

The difference between the two in the real world will be one of pedantry and tautology. The first AGI systems will be running in a datacenter on the GB200, or its successor. This card runs at 2 Ghz. With massive inefficiency, we could expect it to run as low as 100,000 up to a ceiling of around 50 **million** times faster than a human brain. I'm not sure what anyone considers 'human level' about a virtual person that lives a million years to our one would be. That can build out any arbitrary mind that fits within RAM. AGI will be an aimed-for suite of capabilities running on NPU's, closer to animal speeds and a brain's power consumption. The things they'd like to build with their datacenters are ASI's. And sure yeah things exist on a spectrum. The janky early proto-AGI stuff has a wider suite of capabilities than a thermostat, absolutely. Kids and simps look at an 'LLM Plays Pokemon' stream and think to themselves 'ah, AGI is far.' Me, a guy who's been in this rodeo nigh on 30 years now, get the same feelings I did for [StackGAN](https://procedural-generation.tumblr.com/post/154474148263/stackgan-text-to-photo-realistic-image-synthesis) and what that meant for the future of image generation. There's an enormous chasm between 'impossible' and 'needs to be improved.' Having a problem become tractable is effectively a miracle.