In you're in favor of "AI safety" (broad definition), what's your most compelling cast-length argument?

Algos (for feed,security/content filtering) have historically ignored non-English speaking countries. Resulting in messed up elections and many more problems (riots related to religion/caste). I feel the current day AI is still pro-English, and all the above issues will still happen if there are no guardrails in place.

working on aerospace applications during day time. building on AI/Web3 after sunset @rootnode | mani.ink

I don't see how this is a "safety" issue.

If anything, it is an openness issue where we should be giving access to the models to as many countries/locales/devs as possible.

Founder @ CoinFund. Engineer. Investor. Decentralization. AI x web3.

I believe it's a safety issue because of the inherent biases in the models [related to different human identities mentioned above (and more), that can cause havoc in some parts of the world]. Giving access to the models at large doesn't really address the bias issues that happens during training.

I'm not connecting on what "havoc" can be affected by chatbots. :)

I think giving access does address the issues, in the sense that it allows people to work on this problem in a locale-aware way.