This newsletter made possible by MosaicML. Tune As You Scale: Hyperparameter Optimization For Compute Efficient Training I don’t usually care much about hparam optimization algorithms, but this one has some really interesting findings regarding model + data scaling.
“societal-scale” can mean either individual harm that’s widespread enough or harm to society as a system—think democracy, supply chains, or financial systems.
…
Why does nobody ever talk about the 500LB gorilla here?
That normies have to use chatGPT which is woke-crippled AI ( filtered AI )
That LEO, US-MIL, and CEO’s get to use non-Woke AI which in real AI ( non-filtered AI )
…
Why knee-cap the general public and ignore? How long can you expect this two tier system to last??
“societal-scale” can mean either individual harm that’s widespread enough or harm to society as a system—think democracy, supply chains, or financial systems.
…
Why does nobody ever talk about the 500LB gorilla here?
That normies have to use chatGPT which is woke-crippled AI ( filtered AI )
That LEO, US-MIL, and CEO’s get to use non-Woke AI which in real AI ( non-filtered AI )
…
Why knee-cap the general public and ignore? How long can you expect this two tier system to last??