要判断 AI 进步接下来会带来什么，一个方法是去观察那些“不是由研究人员制作的、有趣的 Benchmark（基准测试）” One way you can sense what’s coming next as a result of AI progress is looking at interesting benchmarks that aren’t made by researchers.
Suhail DoshiFounder & former CEO, Mixpanel
观点原文
要判断 AI 进步接下来会带来什么，一个方法是去观察那些“不是由研究人员制作的、有趣的 Benchmark（基准测试）”
One way you can sense what’s coming next as a result of AI progress is looking at interesting benchmarks that aren’t made by researchers.
AI
相关观点
- 我认为举办一场“机器人奥林匹克”会非常有意思，也能用来规范、校准当下的炒作热度。尤其是如果参赛者事先完全不知道环境会是什么样子，并且有一部分任务是未知的。 在和这些机器人相处了大约六个月之后，我发现了一个不太光彩的秘密：它们几乎全部在泛化能力上遭遇灾难性失败。 因此，“微调（finetune）”这个词，是我会特别加粗、划重点强调的。 I think a robot Olympics would be very exciting to regularize the hype. Especially if the participants had no idea what the environment would look like and a portion of the tasks were unknown. One dark secret of these robots (after spending 6 mo), is almost all of them colossally fail at generalization. Hence "finetune" is the word I'd underline.Suhail Doshi · 2025/12/23
- 每次回到 Tokyo，我都会再次意识到：一座城市的标准可以有多高——人、人文、文化、安全、住房。San Francisco 其实有着巨大的潜力。 东京就像是一座到处都是“subreddit”的城市。 Each time I go back to Tokyo, I am reminded how high the bar can be for a city: people, culture, safety, housing. SF has so much potential. Tokyo is like a city full of sub-reddits everywhere.Suhail Doshi · 2025/12/20
- 我投资组合里那些表现烂的股票，只差一个印度籍 CEO，就能变成十倍股了。 All the bad stocks in my portfolio are just one Indian CEO away from being ten baggersSuhail Doshi · 2025/12/09
- 每当你宣布一笔你尚未真正“靠成果赚来”的巨额融资时，你往往会用公众对你“作为弱者的加油支持”，去换取那些“有才华但更偏向风险厌恶的人”的加入。我很好奇，这种交换到底值不值得，价值有多大。 Whenever you announce a giant round of funding that you haven't yet earned, you tend to trade the public rooting for you as an underdog in favor of enticing talented but risk averse people. I wonder how much value there is in the latter.Suhail Doshi · 2025/12/09