The sort of AI that reads radiology scans has almost nothing in common with the sort of AI that develops drugs has almost nothing in common with large language models. A key difference is that the first two require orders of magnitude less processing power. If that was all AI was used for, we wouldn’t be claiming people’s houses under eminent domain in order to build beefier power lines for AI datacenters, and the EPA wouldn’t be loosening environmental regulations on power plants devoted exclusively to AI datacenters. We would also be able to afford computer parts.
The sort of AI that reads radiology scans has almost nothing in common with the sort of AI that develops drugs has almost nothing in common with large language models. A key difference is that the first two require orders of magnitude less processing power. If that was all AI was used for, we wouldn’t be claiming people’s houses under eminent domain in order to build beefier power lines for AI datacenters, and the EPA wouldn’t be loosening environmental regulations on power plants devoted exclusively to AI datacenters. We would also be able to afford computer parts.