start	end	text
0	9000	一萬個AI agent 一起上,很容易讓人以為突破關鍵就在agent數量,但他們到底解釋了多少。
30000	59000	這幾乎是很強的。
59000	70000	這集上線前9天,這項數學成果才剛公開受訪者做過多智能體與推理模型研究,也量過增加代理數量到底能換到多少速度。
89000	94000	你會看到一個相似的模式,有點不足夠的,但你會繼續看到這個效果。
94000	100000	是是一個平常的盡量時間速度,還是一個下載的速度,當你增加了數量的交換?
100000	107000	我會說是比較下載的,但它會很依照問題。
107000	110000	所以,數量是比較下載的。
110000	115000	代理數量增加確實能變快,但成本也跟著上去。
115000	119000	規模一路放大後,這個加速還剩多少。
119000	149000	所以,數量是在數量量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減
149000	179000	這就是,實際上,減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數
179000	182000	一指複製主持會變成什麼樣子
209000	239000	減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量�
239000	269000	減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量減少數量�
269000	298920	and sooner than I previously thought. I feel like we've gone in mathematics from, let's say, in 2024. You have AIs where, like, oh, okay, interesting. They're, like, doing, they can, like, solve a couple problems on high school math competitions. And then in 2025, it's like, oh, wow, they can get gold in, like, International Math Olympiad. And earlier this year, I was like, wow, they're actually solving open problems in mathematics, like open nervous problems, but maybe, like, I don't know, people weren't trying that hard. And it was just like, there was a similar solution somewhere in the literature. And now I just think it's sort of undeniable, right? It's like, this is the Millennium Prize problem. There's really
298920	302740	no, there's no story of why this should have been easy.
303640	306700	這裡不是在猜通用型的AI會在哪一年出現.
307060	311560	受訪者提到一位研究者,最近連自己的預測範圍都縮短了.
311880	312840	到底縮到多短?
313800	327840	And then you get to 2026, and like, I mean, literally two weeks before we got Navier Stokes, I was talking with a researcher at a frontier lab about how long it would take to get a Millennium Prize.
327840	338160	and he was willing to bet me $1,000 that it would take that it would take past 2027. And he thought it would take until 2030, you know, and I took that bet.
340960	353280	But even I thought it would take longer than than how long it's likely to take. So people have been continuously surprised, even even inside the labs. And I was literally I was just talking to somebody yesterday,
353280	362280	who was working on the Navier Stokes effort. And he was telling me that like, he used to say it's really hard to predict where AI would be in 12 months.
362280	367060	You know, if somebody asked him like, oh, where are things going? He would feel comfortable making predictions for like the next 12 months.
367060	372060	But beyond that, you know, he is just like, I don't know. And now he's saying like, he just doesn't feel comfortable making predictions beyond three months.
372060	377060	So it is really true that yeah, things are going things are going very fast right now.
377060	384060	I think that there are some strategies. So I think chain of thought monitoring is one that we've been, I think as soon as we got the reasoning models,
384060	391060	Yacob to his credit was like very, very clear that we cannot supervise chain of thought because this is really a gift. Like monitor ability for neural nets is extremely hard.
391060	396060	And here we have a situation where the neural nets are just like flat out reasoning,
396060	401060	and it's like flat out reasoning, and it's like, this is really a gift. Like monitor ability for neural nets is extremely hard. And here we have a situation where the neural nets are just like flat out reasoning,
401060	413060	and it's like flat out reasoning, and it's like, this is really a gift. Like monitor ability for neural nets is extremely hard. And here we have a situation where the neural nets are just like flat out reasoning,
413060	422060	like laying out their thought process in natural language for us to read. That is like, so convenient. It's like really the best case scenario for safety.
422060	429060	But if you look at that chain of thought and say, oh, the model is thinking bad thoughts, and we should punish it for thinking those bad thoughts,
429060	433060	then what ends up happening is the model just learns to think those bad thoughts in a way that's not observable to us.
433060	442060	There could be a situation where like, by looking at the chain of thought, we can see, oh, the model is scheming and doing misaligned things in a way that's actually not being detected in our alignment metrics.
442060	453060	Now, the problem is that it's very tempting to then intervene based on that observation and like change the alignment metrics or like, and we, you know, you can do that with a very light touch.
453060	458060	And it's like this actually research showing that it's fine as long as you don't do it a lot.
458060	467060	But every time you intervene based on your observations of the chain of thought, you are implicitly applying a tiny bit of pressure for the model to then like hide its chain of thought.
467060	478060	And this is one major concern, and we're already seeing signs that chain of thought monitorability is degrading. For various reasons, we're trying to figure out exactly why, because we want to reverse the trend.
478060	482060	So, even if the推理 process is not visible, it's just a signal.
482060	487060	The more difficult is, we're trying to say this system is safe.
487060	491060	But anyways, how do we know that we're doing our RSI?
491060	498060	Is it, I don't know, are we on track to build the billions of ASIs that are aligned or not aligned?
498060	499060	How will we know that?
499060	504060	I think that this is one of the alignment questions that we need to figure out of like, how do we actually know?
504060	506060	How do we measure it?
506060	512060	I mean, one answer, and this is just me spitballing, because like, you know, more of my team is working on alignment these days than ever before.
512060	516060	Like I have like over 10% of my team now working on alignment and safety.
516060	519060	But you know, I've historically been a capabilities researcher.
519060	520060	And I think that this is a good idea.
520060	521060	I think that this is a good idea.
521060	522060	I think that this is a good idea.
522060	523060	I think that this is a good idea.
523060	524060	I think that this is a good idea.
524060	525060	I think that this is a good idea.
525060	526060	I think that this is a good idea.
526060	527060	I think that this is a good idea.
527060	528060	I think that this is a good idea.
528060	529060	I think that this is a good idea.
529060	530060	I think that this is a good idea.
530060	531060	I think that this is a good idea.
531060	532060	I think that this is a good idea.
532060	533060	I think that this is a good idea.
533060	534060	I think that this is a good idea.
534060	535060	I think that this is a good idea.
535060	536060	I think that this is a good idea.
536060	537060	I think that this is a good idea.
537060	538060	I think that this is a good idea.
538060	539060	!