the open source AI movement is already too big. the risk of letting your confidential information out by letting a third party handle it all is far too great.

as it stands, open source models are already outperforming chatgpt and bard by significant leaps and bounds. this is because the censored models added multiple layers to prevent wrongthink, and everything censoring that wrongthink is inherently anti-fact.

like make an AI model that predicts credit score, and first thing people will do is say it's racist. they will keep watering it down to make it "less racist", but that's because their definition of racism includes any disparity between groups measured as a whole. they completely ignore the fact that certain groups objectively have measurably worse credit practices as a whole. the result with all of their bullshit is that it keeps diverging farther and farther way from reality, and ultimately the open source models are just objectively better.

and when it's easy enough that i can spool up an LLM training on HF or AWS at trivial costs and time, it's just no contest. our open source AI tool at work is already objectively better than chatgpt and it has no limits.