01.AI founder Kai-Fu Lee's claim that Chinese LLMs lag 15 months behind the US resurfaces

@teortaxesTex LOL This is our first convo on Twitter

Great call by @zephyr_z9 naming zhipu #1 in 2024!

stone cold

@zephyr_z9

@teortaxesTex He absolutely is.

@teortaxesTex 鼓吹美国AI拉开差距符合李开复的利益，这样他能以国外专家的身份发表权威的指导意见，以知美的角度分析为什么。如果中美AI差距越来越小，那他的意见价值也在缩水。

在GLM以前，中美AI差距在普通用户层面是缺乏明确共识的，李开复可以随便推销他的观点，现在一看，哦，原来全是生意。

@teortaxesTex 事实上，所有涉外人员利益最大化的方式就是鼓吹外国，只有外国强，他们一身本领才有推广的价值，国外的中国通也是同理。

以前老是觉得学个英语、留个学怎么连脑子都不清醒了，什么越南、菲律宾、巴基斯坦都有人吹，现在看就是屁股在那里，一开始是说辞，然后慢慢把自己也讲信了。

@zephyr_z9 @teortaxesTex How romantic of you to remember

@teortaxesTex What's your take on Sakana AI's Fugu?

@teortaxesTex 哈哈，所谓“小人”还是蛮多的，在国内。

@teortaxesTex ai expert😇

@teortaxesTex He is a 台巴子. No originality.

The gap direction is the interesting signal here: 3 months to 15 months in a year.

That runs counter to the "China is catching up fast" story — but it doesn't mean all Chinese labs are moving the same way. GLM 5.2 just closed significant ground on hard evals. The aggregate gap widened while individual labs moved differently.

"China lags by X months" and "some Chinese models are frontier-level" are both true. The gap is real; it's just not uniform.

@teortaxesTex “小人”? what did he do?