@signulll They did both bro. Before codex actually became useful I saw people clearly being paid to promote it. The game is big accounts quote tweet launch posts or whatever and say something sensational like "this is actually insane" and the quoted post gets boosted engagement
@lafaiel People always cite this but consider that it actually probably was.
It's likely it had incredibly dangerous knowledge in its training corpus (drugs, b*mbs, cyber crime knowledge scraped off dw sites) and they probably didn't have the necessary safety guardrails in place yet
@wxrrjxr@skeptrune Nah the market share will always go to the company with the best model. The people solving hard problems will always pay for the best model. If there is identify verification and it’s restricted to US citizens they won’t be able to distill them.
@skeptrune Sure, I think this release friction is temporary though. They’ll align on safety guardrails for this release and future releases will have the same guardrails in place. Everyone in my government understands the entire US economy is in the hands of these labs.
@skeptrune Yeah but practically, how does it help them?
1. Anthropic/OpenAI still have the best models. Still have the biggest feedback loop of RL training data.
2. This makes it harder for Chinese labs to distill
3. Even if more people flock to their models they don’t benefit from it