Which is why my hope is this:

1. `Bigger = better` is building a shitton of datacenters
2. `Bigger = better` will turn out to be a fata morgana
3. There will be a lot of unused compute
4. Compute will be cheap
5. We can start training honest, truly open models, for cheap

many thumbs must already be on the scales.

Scoresby

I just worry about all the poisoning. If you wanna run a real psyop in 2025, you give OpenAI some cash to burn in exchange for some improved weights in gpt-n