AI Vibe Check: Lab Wars, Why APIs Might Vanish & Future Predictions

I'm Jacob Efron and this is Unsupervised Learning. 我是 Jacob Effron，欢迎收听 Unsupervised Learning。 We've had a bunch of new subscribers, uh, over our last few months. 过去几个月里，我们新增了不少订阅者， And so, uh, wanted to welcome you to the show. 所以想借这个机会欢迎大家来到节目。 We basically probe the sharpest minds in AI on everything that's happening today, what's real, and what's coming up, where the space is headed. 我们邀请 AI 领域最锐利的大脑，聊聊当下正在发生什么，什么是真实的，未来会走向哪里。 And today's episode is one of my favorite formats we do. 今天这期是我最喜欢的节目形式之一。 It's an AI vibe check uh, that we do with Ahri from Datlogy. 这是一次 AI 氛围盘点，我们请来了 Datology 的 Ari。 Ari was a former researcher at DeepMind and Meta now runs a really exciting AI startup and Rob at Radical uh one of the great AI venture firms. Ari Morcos 曾在 DeepMind 和 Meta 做研究，现在经营着一家非常令人兴奋的 AI 创业公司；Rob 来自 Radical Ventures，顶尖的 AI 风险投资机构之一。 The three of us talk about everything happening in the AI world today. 我们三个聊了当前 AI 世界里发生的一切。 We talked about Fable of course uh and the reaction around the release as well as model capabilities. 当然聊到了 Fable，以及发布后的反响，还有模型能力方面的话题。 We talked about how close we are to RSI. 我们聊了聊距离 RSI 还有多远。 We hit on some pretty spicy predictions uh including that the labs may actually get rid of their API business as the compute crunch continues. 还触及了一些相当劲爆的预测，包括随着算力吃紧，实验室可能真的会砍掉 API 业务。 Uh and we just got to hit on all the main topics of today. 我们几乎把今天所有主要话题都过了一遍。 Uh just really fun to sit down with two friends and and great minds in AI. 跟两位 AI 领域的好友和大拿坐在一起聊天，真的很过瘾。 I think folks really enjoy this. 我觉得大家会很喜欢这期。 Without further ado, here's our conversation. 话不多说，直接开始。 It's time for another roundup episode. 又到了我们的综合盘点时间。 I always love doing this with you guys, Ari and Rob. 跟你们两个聊天，我每次都很享受，Ari 和 Rob。 Uh I feel like we had a ton of fun in the last one, but like god it's AI world. 感觉上次录得挺爽的，但 AI 这个世界，变化太快了。 Things have changed. 很多事都变了。 Uh I feel like we we last sat down after Nurups and I think since then we've had IPO filings. 感觉上次是在 NeurIPS 之后坐下来聊，那之后我们经历了 IPO 申请， We've had you know uh models not launched and then launched. 有些模型本来没发，后来发了， We've had uh you know SpaceX becoming an AI info company. SpaceX 变成了一家 AI 基础设施公司， Uh no shortage of headlines to uh to discuss here. 头条消息层出不穷，多到聊不完。 So excited to have you both back on the show. 非常开心你们俩再次来到节目。 Excited to be here. 很高兴来这里。 Yeah, thanks for having us. 谢谢邀请。 So I I I think to kick it off uh you know 6 months is an eternity in AI world, but I figured I'd start at the highest level. 那我觉得咱们先从最宏观的角度切入，六个月在 AI 世界里是一个永恒，但还是想从大处着眼， Uh what's like the single biggest thing that has changed in how you're thinking about the landscape since we last talked? 自从上次聊之后，你觉得对整个格局影响最大的变化是什么？ And maybe uh Ari, I'll start with you. 那就先从 Ari 你开始。 Yeah, I mean I think the the most obvious thing that has changed uh over the last six months is starting to see the coding agents really start to work at longer time horizons. 对，我觉得过去六个月最显而易见的变化，就是编程 Agent 开始真的能在更长的时间跨度上工作了。 Um right 嗯，对。 I think that was just starting when we recorded our last episode at the end of uh 25. 我觉得上次录制末尾就初见苗头，当时差不多是 2025 年底。 Feel like everyone went away over Christmas break and was like holy crap like these things really work. 感觉大家圣诞假期一回来就在想，天哪，这些东西真的好用了。 Yeah. 对。 And and I think it it starts to show how there are these thresholds where if you go beyond a threshold it can become a lot more valuable. 而且我觉得这说明存在一些门槛，一旦越过某个门槛，价值就会大幅跃升。 Um, and obviously that's driven the the massive rise in in token spending and the whole token maxing idea and and all this stuff. 这显然推动了 token 消耗量的暴涨，带来了 token 最大化的理念和一系列相关讨论。 Um, but I think we're really starting to now see the shift of engineers at least kind of almost all moving from IC's to managers of agents. 但我觉得我们现在真的开始看到转变了，工程师几乎都在从独立贡献者向 Agent 管理者转型。 Um that's been something that's been very noticeable within Datlogy for example um over the last uh number of months is seeing more and more people starting to now context switch between managing different agents uh rather than just kind of you know working on the one thing and that was enabled by having these um you know agents be able to run long enough and actually be useful in various ways. 这在 Datology 内部也非常明显，过去几个月里越来越多的人开始在多个 Agent 之间切换管理，而不是专注在一件事上，这是因为 Agent 现在能跑够长的时间，在各方面真正发挥作用了。 I mean everyone likes to ask top AI researchers like yourself like how much more productive has it made you in your uh in your work. 大家都爱问像你这样的顶尖 AI 研究者，这些工具让你在工作上提升了多少效率？ I think that it's interesting. 我觉得这个问题挺有意思的。 It makes you a lot more productive in some ways, right? 在某些方面确实让你高效很多， But it also um produces a lot of challenges as well. 但同时也带来了不少新的挑战。 Like one of the things that we're struggling with is now um it's a lot easier to produce a massive amount of code that can do something, but now you have this pretty massive understanding gap and it's a lot easier to put slop uh into your codebase. 比如我们现在面临的一个问题是，生成大量能跑通的代码变容易了，但随之而来的是巨大的理解鸿沟，往代码库里塞进低质代码也变得更容易了。 So it's definitely made um made us more productive. 所以确实让我们更高效了， I think a lot of times though the kind of topline numbers tend to be overestimated um because it doesn't take into account some of these like later costs of like we now have big bottlenecks on reviews um and we don't want to go fully to like 但我觉得很多时候那些表面数字往往被高估，因为它没有把后续成本考虑进去，比如现在 code review 就成了大瓶颈，我们也不想完全走到 oh you just like but my agent will review your agent's output you know 就让你的 Agent 来 review 我的 Agent 输出那一步， the bottlenecks just seem to shift um you know whatever uh it's hard to improve on an entire process uh because of because of that. 瓶颈只是在转移，不管怎么说，要提升整个流程的效率本身就很难，就是因为这个。 What about you Rob? Rob，你呢？ There are early signs that seem to suggest over the past six months that that make me question whether openweight AI is going to continue to be a really meaningful uh force in the ecosystem going for at least like near frontier 过去六个月有一些早期迹象让我开始质疑，开源权重 AI 是否还会继续成为生态中真正有分量的力量，至少在接近前沿这一层来说。 opening for the jugular feri like right off the uh right off the top here. 上来就直奔要害，刚开头就这么猛。 I like it. 喜欢。 There's Yeah. 是的。 Yeah, we can we can dig in more detail, but I think like Yeah. 对，我们可以深挖，但我觉得大概就是这样。 six months ago or for the past few years, like my kind of working assumption had been that the closed source proprietary models would advance the frontier and there are a lot of structural reasons for that. 六个月前，或者说过去几年，我的基本判断一直是，闭源专有模型会推进前沿，这背后有很多结构性原因， But the open frontier would is only would only be a few months behind. 但开源前沿只会落后几个月。 Uh and that gap might widen a little bit. 这个差距可能会稍微拉大， I didn't think it would shrink altogether, but like I thought it would persist as as being relatively small. 但我觉得不会彻底消失，还是会保持相对较小。 And I think there are signs now that like there it seems like there's a real risk of of near frontier openweight AI falling off altogether. 而现在有迹象表明，接近前沿的开源权重 AI 可能真的会全面掉队。 Um I think meta which historically has been the the the openw weight champion in the west is is pulling back and it see it seems likely that they're not going to continue with their open source strategy. Meta 一直是西方开源权重模型的旗手，但现在似乎在收缩，它继续走开源路线的可能性越来越小。 And then more recently, obviously, the Chinese labs have been the ones driving state-of-the-art open research. 然后最近，显然中国的实验室一直是推动前沿开源研究的主力。 And there it seems like there's strong indications that they may also be pulling back from that. 而且现在也有强烈迹象表明他们也在撤退。 And they're uh you know, whether it's Quen or Deepseek um or uh or others their most high performing models, they're now keeping proprietary behind an API and just open open sourcing, open waiting, you know, smaller, less performant versions. 无论是 Qwen 还是 Deepseek 还是其他家，他们表现最好的模型现在都锁在 API 后面收费，只开源那些更小、性能更弱的版本。 And I think there's like real compute incentives behind that. 我觉得背后有很现实的算力经济学逻辑。 like it's just very expensive to to uh service these open way models with no revenue coming in. 开放权重模型没有收入，维护起来成本很高。 Uh and I think there's also geopolitical and and competitive considerations. 同时我觉得也有地缘政治和竞争方面的考量。 But it's interesting to contemplate what a world might look like where like if you want real frontier artificial intelligence like you have to pay a company for a proprietary model as opposed to being able to you know have access to the weights yourself 但想象一下这样一个世界，如果你想要真正的前沿人工智能，就必须向某家公司付费用专有模型，而不是自己获取权重， or or build your own. 或者自己搭建。 Uh but I I'd actually agree with that. 我其实也同意这个判断。 I the second one this other one I was imaging about as a you know what changed 这也是我在想的第二件事，就是什么发生了变化。 I don't think that we've seen a major change in in the capabilities or the trend line of the capabilities of the open weight models. 我不觉得开源权重模型的能力或能力提升趋势有什么大的变化。 Um I actually think if anything we started to converge and we continue to see that. 实际上我觉得如果说有的话，我们开始趋同了，而且这个趋势还在继续。 Um I do think though the economic decision-m around building and releasing open models has definitely changed um over the last uh six months which I think is what Rob was really getting at. 但我确实觉得，围绕构建和发布开源模型的经济决策已经发生了明显变化，这也是 Rob 真正想说的。 Um, and I I am a lot more bearish on how many open models there will be going forward. 我对未来开源模型的数量要悲观得多了。 It seemed like, you know, there was uh this kind of cornucopia of open models that was only ever growing in 2025. 感觉 2025 年那时候，开源模型就像取之不尽的宝库，规模只增不减。 And I think we're now definitely starting to see that we probably hit the peak number of open models and it's now going to kind of get less and less. 而现在我们显然开始看到，开源模型的数量可能已经见顶，之后只会越来越少。 And because the financial incentives just don't make sense once you've already kind of achieved credibility, it makes sense to invest a lot of money to do that. 因为财务激励算不过来，一旦建立了足够的信誉，烧那么多钱去做这件事就不合算了。 But after that point, you want to start selling hosted inference of your model. 到了那个节点，开始卖自家模型的托管推理才是正道。 Um, and opening it up just fully undermines your business, which so I I so I think we're are going to see like other Chinese labs start with like big open models, but then probably close up after that once they've kind of gotten enough press and PR. 完全开放权重会直接砸掉自己的生意，所以我觉得其他中国实验室会先用大型开源模型搏头版，拿到足够的声量和 PR 之后，就会转向闭源。 Is there a business model for an open source model company or is it literally just it's marketing uh to, you know, to as you're on your way to the frontier and then once you're once you're kind of close, uh, it makes it inevitable to want to go close source? 开源模型公司有没有商业模式？还是说这本质上只是一种营销手段，当你向前沿冲刺的时候用，一旦接近前沿，走向闭源就成了必然？ I don't think there's a business model honestly. 我说实话，我觉得没有商业模式。 Uh there you know different things have been tried the kind of like premium enterprised like uh you know Red Hat um model but I just I don't think that it it works in AI given the just the massive upfront investment required to to get to the frontier close to the frontier in the first place. 各种路子都试过了，像 Red Hat 那种企业级高级订阅模式，但在 AI 领域我觉得根本行不通，因为进入前沿本身所需的巨额前期投入太大了。 We'll see what I like uh you know we can come back and revisit this soon but I I very curious to see if you know first of all if reflection when reflection releases a model and uh what business model they aim to pursue with it. 我们拭目以待，但我很想知道 Reflection 发布模型之后会走什么商业模式。 Um but I yeah I remain skeptical that like a great business model exists for open AI open source AI. 但我仍然对开源 AI 能不能找到好的商业模式持怀疑态度。 It's funny because I was going to say one of the one of the trends we've actually seen happen you know in in the past like month or two is that it finally you know forever 有意思，因为我本来想说一个我们这一两个月确实看到发生的趋势，就是终于……以前 I think people have been like everyone will use smaller cheaper open source models for tasks that those models can do and uh you know it felt like the vast majority of usage and tokens was still just like at the frontier pushing capabilities like figure out what you know models can do in different industries and then finally I feel like we have a movement now toward like gez these bills are pretty expensive uh or our usage is pretty high like wouldn't it be nice to to have something that was that is, you know, cheaper and and faster and smaller to use. 大家一直说，比较简单的任务会用更小更便宜的开源模型来做，但感觉绝大多数用量和 token 消耗还是集中在前沿，都在探索模型在各个行业能做什么，而现在我感觉终于有了一股推力，开始想，这些账单真的很贵，用量很大，要是有个便宜、更快、更小的方案就好了。 Uh, but it's funny that it's happening at the same time that we're kind of seeing this real uh, you know, the closed source models, you know, run ahead of of of open source models. 但有趣的是，这个趋势正好和闭源模型把开源甩开的时机重叠。 And Ari, I know you spend a lot of time thinking about this stuff, like how do we think about those like counterveailing forces that seem to be happening at the same time? Ari，我知道你在这方面想了很多，这两股看似同时发生的反向力量，我们该怎么理解？ Yeah, I I mean I think that first off I I've definitely seen a lot of the the former that you were talking about. 我觉得首先，我确实强烈感受到了你说的那个前者。 Like it's been very interesting um the combination of the you know extreme computed environments that we're operating in. 我觉得非常有意思，我们正在运作于极端算力紧张的环境里，这种结合。 um the rise of capabilities of uh the coding models in particular um and then even just seeing you know from release to release models changing their token efficiency their output token efficiencies um has resulted in you know a lot of companies that were happy using Frontier models all of a sudden even just going from like Opus 46 to 47 there was a big difference in token efficiency um and a lot of people's bills just doubled overnight and you've now I'm now starting to see talking to a lot of enterprises in particular really strong desires to start cutting the cost of of using the models. 编程模型能力的崛起，再加上从一个版本到下一个版本模型 token 效率的变化，导致很多原本用着前沿模型的公司，哪怕就从 Opus 4.6 升到 4.7，token 效率就有了明显提升，很多人的账单一夜之间翻了一倍，我现在和很多企业聊，大家都有非常强烈的意愿去削减使用模型的成本。 Um, and I think that wasn't there nearly uh to the same extent a year ago um because the models weren't being used at a scale where those costs were you know meaningful enough but now they've really reached that point where they are meaningfully meaningful enough um and you can just consume budgets so quickly because the models think for a long time um and that's now driving a lot of demand to say okay how can we make this much cheaper. 我觉得一年前这个问题远没有那么突出，因为当时模型的使用规模还没到让这些成本真正显眼的程度，但现在已经到了那个节点，因为模型会思考很长时间，可以迅速把预算耗尽，这就推动了大量需求，大家都在问：怎么把这个成本降下来。 I think that you can do a lot of that with open models as well. 我觉得用开源模型也可以实现很大一部分降本。 Like like I think one very notable thing right is that like a lot of people were able to reproduce um the same or similar level of kind of you know vulnerability finding um that mythos used with open models by just putting scaffoldings around. 一个很值得注意的事情是，很多人能够用开源模型加上好的脚手架，复现出 Mythus 做漏洞挖掘的同等效果或类似效果。 I think that's another one of the big changes right is that a model is not just a model anymore it's the model combined with the harness and the scaffolding and a lot of innovation is happening on the harness and scaffolding layer. 这也是另一个大变化，一个模型不再只是一个模型，而是模型加上框架加上脚手架，很多创新都在框架和脚手架层发生。 I think that's also possibly how open source models can have a business model. 我觉得这也可能是开源模型找到商业模式的一条路。 you open source the model, you don't open source the scaffolding in the harness. 模型开源，脚手架和框架不开源， Um, and you, you know, then have an API where people can access the full system. 然后提供一个 API 让人们访问完整系统。 Um, I think that could potentially work. 我觉得这条路可能行得通。 Kimmyy's kind of doing that, right? Kimi 某种程度上就在走这条路， Like moonshot, I think, is is at a couple hundred mill through the the API and and kind of the chat interface. 我记得月之暗面通过 API 和对话界面，营收大概到了几亿美元。