提问 HN：什么才能让你认真看待人工智能？
5 分•作者: atleastoptimal•5 个月前
最近，OpenAI 宣布他们最近开发的一个 AI 模型/系统在国际数学奥林匹克（IMO）中获得了金牌。IMO 是一项非常难的考试，只有世界上最优秀的高中生才有资格参加，更不用说赢得金牌了。那些获奖者往往会从事前沿的数学研究，比如陶哲轩，他在 2006 年获得了菲尔兹奖。也有传言称，DeepMind 也用一个尚未发布的模型取得了同样的结果。
现在，在艰难的数学考试中取得成功并不能“自动化所有人类劳动”，但这确实是一个许多人认为 AI 难以轻易达到的基准。即便如此，许多人声称这并不是什么大不了的事，并且在可预见的未来，人类仍然会比 AI 聪明得多。
我的问题是，如果你属于上述阵营，你需要什么才能转变思维模式，大致认同“AI 系统变得比人类更聪明，并且可以在个位数年内自动化所有人类劳动和认知输出”的观点？
这是否需要看到一个类人机器人执行一些困难的任务？（Metaculus 对通用人工智能（AGI）的定义要求机器人能够令人满意地组装一个（或相当于一个）大约 2021 年的法拉利 312 T4 1:8 比例的汽车模型。）这是否涉及足够严格的图灵测试？我很好奇人们对“好吧，这真的发生了”的个人定义是什么。
查看原文
Recently OpenAI announced an AI model/system they had recently developed won a gold medal at the IMO. The IMO is a very difficult exam, and only the best high schoolers in the world even qualify, let alone win gold. Those who do often go on to cutting edge mathematical research, like Terence Tao, who won the Fields medal in 2006. It has also been rumored that DeepMind achieved the same result with a yet to be released model.<p>Now, success in a tough math exam isn't "automating all human labor" but it is certainly a benchmark many thought AI would not achieve easily. Even so, many are claiming it isn't really a big deal, and that humans will still be far smarter than AI's for the foreseeable future.<p>My question is, if you are in the aforementioned camp, what would it take you to adopt a frame of mind roughly analogous to "It is realistic that AI systems will become smarter than humans, and could automate all human labor and cognitive outputs within a single-digit number of years".<p>Would it require seeing a humanoid robot perform some difficult task? (the Metaculus definition of AGI requires that a robot be able to satisfactorily assemble a (or the equivalent of a) circa-2021 Ferrari 312 T4 1:8 scale automobile model.). Would it involve a Turing test of sufficient rigor? I'm curious what people's personal definition of "ok this is really real" is.