17 karma · joined November 13, 2021
But in this case it’s pretty easy, other model providers have in fact limited harm better than grok. So you don’t even need something arbitrary, just do it as well as competitors.
You can’t go after a company that makes kitchen knives if those are used to harm bc there’s nothing reasonable they could have done to prevent that harm, and there’s a legitimate use case for knives for cooking.
In this case, my understanding is other companies (OpenAI and Anthropic) have done more to limit harm, whereas XAI hasn’t.
If you produce a product that causes harm, and there are steps that could be taken to prevent that harm, you should be held responsible for it. Before the trump admin dropped the Boeing case, Boeing was going to be held liable for design defects in its Max planes that caused crashes. The government wasn’t going after Boeing bc a plane crashed, but bc Boeing did not take adequate steps from preventing that from happening.
Is it that low and high are both floating point, so you're not constrained by int precision and so you don't get an overflow error. The article makes it sound like sign switching is the issue, but this is just a general overflow problem, right?