特设人工智能许可制度

今天在《AI每周简报》中，我们来聊聊你临时的AI授权机制。

Today on the AI Weekly Brief, meet your ad hoc AI licensing regime.

《AI每日简报》是一档每天更新的播客和视频节目，聚焦AI领域最重要的新闻和讨论。

AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

好了，各位。

Alright, friends.

我们又回来了，带来又一期五分钟的AI每周简报。

We are back with another five minute AI weekly brief.

你可以用这一期来补上错过的资讯，或者转发给那些还没跟上每日更新、但想知道最新动态的人。

Use this episode to catch up on what you missed or to forward along to someone who needs to know what's going on but isn't quite at the daily episode stage yet.

在正式开始之前，先说一句提醒。

Quick note before we get into the show.

我们在录完并剪辑完这期节目后，周五下午和晚上又收到了一些重大新闻更新。

After recording and editing this episode, we got some big news updates on Friday afternoon and evening.

你接下来听到的内容，是这些新消息发布之前的版本；别走开，节目最后我会聊聊关于Mythos和GPT-5.6的最新动向。

What you're about to hear is the episode that happened before that news came out, And stick around at the end for a few thoughts on the latest on Mythos and GPT-5.6.

这一周的特点是，AI领域没有哪一件大事能称得上头条，但从政治、市场到模型，每个方面都发生了些小变动，一点一点把我们推向了未来——而这个未来，很多人并不喜欢。

This was a week where the hallmark was there was at no point any one huge story in AI, but instead every single part of AI, from politics to markets to models, had something happen that inched us along towards a future that, in many cases, not everyone is loving.

当然，这段时间最大的背景音还是‘Fable Five’的传闻。

Certainly, the big background noise continues to be the Fable Five story.

上周末，有人开始传播参议员马克·沃纳的一段话，他说NSA告诉他，Mythos已经入侵了他们的机密系统。

Last weekend, people started chattering about a quote from Senator Mark Warner, who reported that the NSA had told him that Mythos had been able to break into their classified systems.

但很多人误以为Mythos真的攻破了NSA的机密系统，其实真相是：在一次红队演练中——也就是为了测试Mythos到底有多危险而做的可控实验——NSA发现它的能力远超以往。

Now many people took this as Mythos literally breaking into NSA's classified systems when, of course, what it actually was was that during a red teaming exercise, in other words, a controlled experiment to figure out just how dangerous Mythos could be, the NSA saw much more significant capabilities than they had seen before.

这为整周关于Mythos的传闻定了调：大家只要听到一点风声或零碎评论，就迫不及待地当成它即将回归的信号。

And this really set the tone for the week when it came to fable and Mythos, in that everyone wanted to latch on to any small rumor or comment as an indicator of when we were actually going to get this thing back.

这一周市场预测也大起大落，一开始大家普遍不看好，但有一天，预测‘Fable Five’在7月1日前回归的概率突然飙升到60%以上。

One big whipsaw throughout the week was in the prediction markets, which started very bleary, but then had a single day where the likelihood of Fable five returning by July 1 jumped all the way up above 60%.

后来发现，这个跳涨的源头是报道称，白宫在CEO达里奥·阿马德被边缘化后，与Anthropic联合创始人汤姆·布朗的沟通变得顺畅了。

Now it appears that the source for that were reports that the White House was having a better time interacting with Anthropic co founder Tom Brown after CEO Dario Amade had been sidelined in the conversations.

但这一周最重要的、也是最实质性的进展，是最终确认：这根本不是白宫和Anthropic之间个人矛盾那么简单，而是一个全新的、临时的、非正式的、缺乏问责、且明显技术能力不足的许可机制。

And yet the week closed on the most important and most substantive news in all of this, which is that, yes, indeed, this is not just a personality clash between the White House and Anthropic, but a new ad hoc, informal, unaccountable, and seemingly not particularly technically competent licensing regime.

最新消息是，GPT-5.6 也被特朗普政府推迟了。

The newest news is that GPT-5.6 is indeed being delayed by the Trump administration as well.

周三的一场公司问答中，萨姆·阿尔特曼告诉 OpenAI，GPT-5.6 不会向公众发布，而是先以限量预览形式提供给一小部分合作伙伴。

In a Wednesday company Q and A, Sam Altman told OpenAI that GPT-5.6 would not be coming to the public, but would instead be released in a limited preview to a small group of partners.

原因是政府要求 OpenAI 这样做。

The reason was that the government had asked OpenAI to do it like that.

周四的一份备忘录里，阿尔特曼补充说，政府将在预览期间逐个审核客户的访问权限。

Now, in a memo on Thursday, Altman added that the government would be, quote, approving access customer by customer during this preview period.

不过阿尔特曼在周四的备忘录里也明确表示，我们已经向美国政府说明，这并不是我们理想的长期模式，我们会与政府和业内其他伙伴合作，寻找更可持续的后续发布方式。

Now Altman did say in the Thursday memo, we've made clear to the US government that this is not our preferred long term model and will work with them and others in industry to achieve a more sustainable approach for future releases.

可事实就是，我们眼下正处在这个局面。

But here we are.

如果你觉得我语气里带着点讽刺、不满和火药味，那你确实没听错。

Now if you are sensing a bit of snark vitriol acrimony in my tone, you are indeed picking up on that correctly.

从外部来看，我觉得虽然 Anthropic 在应对政府时表现得不够妥当，但这件事的根源，其实是美国政府之前压根没当回事，直到另一家公司——亚马逊——提醒他们，Mythos 能做点 Anthropic 早就说过它能做的事。

The best I can tell from the outside is that while Anthropic didn't handle themselves particularly well with regard to the government, mostly this whole thing started by the US government not really having paid attention until a different company, Amazon, told them that Mythos could do a thing that Anthropic had been clear that Mythos could do.

而现在，我们迎来了最糟糕的结果，正如尼尔·奇尔森所说，这种随意、模糊、不透明的许可要求，比繁文缛节还要糟糕。

And now we have the worst possible outcome, which is, as Neil Chilson put it, arbitrary, unknown, nontransparent license requirements are far worse than red tape.

兹维·毛舍维特说，新的AI政策就是：白宫想什么时候决定谁有资格接触前沿技术，就什么时候决定，完全看他们心情。

Zvi Maushewitt said, The new AI policy is that the White House decides ad hoc for whatever reasons it likes, who does and does not get access to frontier intelligence.

这简直糟糕透顶。

This seems rather maximally terrible.

安德鲁·柯兰提醒大家：那些说这是暂停或安全胜利的人，完全想错了。

And Andrew Curran reminds, For the people saying this is a pause or a victory for safety, it is not.

这根本没让研发慢下来。

This does not slow development in any way.

它只是让实验室发布模型的速度变慢了，但训练模型的速度一点没减。

It only slows the rate at which the labs can release the models, not how fast they can train them.

从今天起，公众能接触到的东西和实验室内部掌握的东西之间的差距，只会越来越大。

The gap between what is available to the public and what the labs have internally will steadily widen from this day forward.

说到底，这谁都不满意。

This actually makes no one happy.

而且这不只是让没人开心的问题。

And it's not just that it makes no one happy.

这对谁都没好处。

It is good for no one.

我真心希望白宫能尽快理顺局面，或者国会能挺身而出，好好履行职责，别让我们陷在这种反复无常、让所有人都受损的境地里。

I very, very much hope that the White House can get its act together and or that Congress can step in and, you know, do its job so that we're not left with this incredibly capricious scenario that makes literally everyone worse off.

与此同时，其他公司的人还在照常做事，我也是。

Meanwhile, other people in other companies are doing what they do, I.

嗯，就是适应变化。

E, adapting.

这一周里，各个机构，尤其是中小型企业和初创公司，继续在测试Z dot AI的新版GLM 5.2模型，之前那些积极的反响还在持续。

Throughout the week, organizations, especially smaller organizations and startups, continued to experiment with Z dot AI's new GLM 5.2 model, and all of the positive buzz that we've seen before seems to be continuing.

再来看看美国本土，谷歌也宣布，Gemma 4的下载量刚刚突破两亿，进一步证明大家确实在寻找成本更低的模型架构。

Now a little bit closer to home in The US, Google also reported that Gemma four had just hit 200,000,000 downloads, providing more evidence that people really are looking for lower cost alternative model architectures.

这周我们还再次意识到，谈到AI进展，现在已不只是模型本身的问题了。

We also got a reminder this week that when it comes to AI advancements, it's not just a model question anymore.

我们使用AI的方式，以及那些逐渐成为常态的界面和交互设计，也会极大影响AI能为我们带来的价值。

The harnesses through which we use AI and the UI UX patterns that become normalized have a big impact on the value we get from AI as well.

这周最引人注目的例子就是Claude Tag。

The most interesting example of this this week was Claude Tag.

Claude Tag本质上是把Claude直接集成到Slack里，你可以在任何频道或对话中@它，让它在后台自动帮你处理任务。

Claude Tag is basically a native integration of Claude into Slack, where you can tag Claude from any channel or conversation and have it initiate work in the background.

这不只是简单地调用Claude，而是调用一个完整的Claude代码实例，能大大降低非技术人员使用Claude代码的门槛——而技术团队和早期采用者在过去半年多里早就从中获益了。

But it's not just calling upon Claude, it's actually calling upon a full instance of Claude code, which has the potential to dramatically lower the barrier to people on the team who aren't technical using Claude code in ways that both the technical team and more early adopter knowledge workers have been getting value from it for the last six months or more.

更进一步的是，由于Claude Tag直接嵌入到工作流程中，它能获取的上下文信息也变得更多了。

What's more, by embedding Claude Tag directly into the context flow, it has the potential to change just how much context it has available.

Anthropic表示，现在他们65%的代码都源于人们在Slack里讨论需要做什么、要开发什么，这简直是一场巨大的行为转变。

Anthropic says that 65% of their code now originates by people talking about what needs to happen or what they need to build in Slack, which is just a massive behavior pattern shift.

我觉得未来几周，我们会看到更多关于‘AI进入多人协作模式’的实验和讨论。

I think we're going to see a lot more experimentation and discussion around the idea of AI as moving to multiplayer in the coming weeks.

但有意思的是，Fable 5的背景上下文，加上之前就已存在的令牌效率和成本考量，让人们对各大实验室推出的最新功能没那么急于跟进了。

But interestingly, the Fable five background context, along with the token efficiency considerations and cost concerns that had been growing even before that, have made people a little bit less willing to just jump in to the latest feature from one of the big labs.

正如Prime Intellect的Will Brown所总结的，最近几周确实发生了明显变化，越来越多大型企业开始希望锁定算力，自己内部训练模型，而且经常基于GLM 5.2。

As Will Brown from Prime Intellect summed up, Something has definitely shifted in the past few weeks, seeing a huge uptick in large enterprises wanting to secure compute and post train their own models in house, frequently on top of GLM 5.2.

现在大家都开始明白开源为什么能赢了。

Everyone is starting to understand how open source wins.

这其实可能和我们本周最值得关注的一个数据有关，来自普华永道第二季度全球AI脉搏调查：由CEO主导的AI项目，产生投资回报的可能性是CEO参与度低的项目三倍。

Now this, I think, actually might dovetail with our most interesting stat of the week, which comes from KPMG's Global AI Pulse Survey for quarter two that found that CEO led AI efforts were three x more likely to produce ROI than efforts where the CEO was less involved.

如今，直接选用各大封闭实验室的最新前沿模型，已经成了默认选择。

Always going with whatever the state of the art model is from the closed source labs is sort of the default option these days.

更复杂的、集成开源的、甚至经过微调的架构，能带来更强的数据主权和成本优势，这类方案，我认为才是由CEO主导的AI项目更有能力落地的类型。

More complex, open source integrated, or even post trained, the sort of novel architectures that can provide more data sovereignty and cost efficiencies are the types of things that I think organizations with CEO led AI efforts are going to be a lot more capable of implementing.

最后今天，如果你在找市场里明确清晰的信号，我先说声抱歉。

Lastly today, if you're looking for any strong clear signals from markets, I apologize in advance.

本周初，大家觉得AI泡沫要破了，这个说法一度很流行。

The beginning of the week looked like the AI bubble was popping, and in fact, that was a popular narrative.

但后来美光科技的财报彻底爆了，而且他们还明确指出，AI供应链各个环节的结构性短缺短期内根本不会缓解，这下所有人都重新回来了。

But then Micron absolutely destroyed its earnings, and what's more, reinforced with their projections that the structural supply chain shortage for every single aspect of the AI supply chain wasn't going away anytime soon, which got everyone back on board.

据最新报道，OpenAI 现在倾向于推迟到明年再进行上市。

OpenAI, it is now reported, is starting to lean towards waiting until next year for IPO.

说实话，眼下这情况，谁又能怪他们呢？

And honestly, with everything going on, who can blame them?

接下来一周值得关注的，还是那五大厂商的回归动态。

In terms of what to watch for next week, it remains the Fable five return watch.

但我觉得这周我们已经看到，局势变得越发严峻了。

But I think we saw this week that the stakes are getting even higher.

我们现在所处的这种状态——比如霍华德·卢特尼克和萨姆·阿尔特曼之间那种临时性的短信式授权模式——我再说一遍，对所有人都没好处。

The situation that we're in now, the ad hoc text messages between Howard Lutnick and Sam Altman type of licensing regime is, I will repeat, bad for everyone.

我希望这种情况能尽快得到改变。

And I hope that this has changed as soon as humanly possible.

好了。

Alright.

在录完那期节目之后，又发生了两件事。

So two things happened after the recording of that episode.

第一件事是，媒体开始报道，虽然我们没能重新获得Fable Five的访问权限，但美国已经解除了对Mythos的封锁，允许一批选定的合作伙伴使用。

The first is that news outlets started reporting that while we were not getting Fable five back, The US had lifted its block on Mythos for a handful of selected partners.

报道称，大约有一百家机构，包括多家美国大型企业和政府机构，现在可以获得访问权限。

The reporting was that it was about 100 institutions, including major US companies and government agencies, that would now have access.

正如安德鲁·库兰所写，当时整个氛围变得极其诡异。

As Andrew Curran wrote, The vibe shift was nightmarish around this.

作家贾斯汀·墨菲捕捉到了这种情绪，他说：如果他们真的开始控制谁能使用最顶尖的模型，那无异于宣战。

Writer Justin Murphy captured the sentiment saying, If they really start to gatekeep who gets to use the best models, that is a declaration of war.

这个前景让我涌起一种前所未有的、发自内心的赛博朋克式权力意志，至少从我青少年时代以来从未如此强烈。

This prospect fills me with the most sincere bodily cypherpunk will to power that I've ever felt, at least since I was a teenager.

如果他们真要走这条路，我会全力以赴，打造最疯狂的开源模型和微调集群，专门用来搅乱整个体制化的公共领域。

If they really go down this route, I would go all in on building the most psychotic swarms of open source models and fine tunes possible, all geared toward a chaotic good jamming of the entire institutional public sphere.

如果我们不这么做，思想市场的政治生活还没等我们反应过来，就会彻底终结。

If we didn't do that, all of political life in the marketplace of ideas would be over before we know it.

但那还不是唯一的消息。

And yet, that wasn't the only news.

我们终于也拿到一些关于GPT-5.6的更多信息了，果然如之前报道的那样，美国政府一开始会限制它的使用频率。

We also did finally start to get a little bit more information about GPT-5.6, and exactly as the reporting had suggested, it is going to be rate limited at first by the US government.

萨姆·阿尔特曼在推特上总结得挺到位。

Sam Altman summed it up this way on Twitter.

先说个好消息。

Good news first.

SOL是个聪明、高效又重要的进步。

SOL is a smart, efficient, and significant step forward.

它的定价和GPT-5.5一样。

It is the same price as GPT-five point five.

GPT-5.6系列还推出了Terra，性能达到5.5水平，价格却只有一半。

Also launching in the GPT-5.6 family is Terra, with 5.5 level performance at half the price.

再说个坏消息。

Bad news.

应美国政府要求，今天它不是按原计划开放使用，而是以有限预览的形式上线。

At the request of the US government, it is launching today in limited preview instead of the open access launch we were planning on.

我们正在和政府密切合作，尽快推动模型全面开放。

We are working with the government to get general availability as fast as we can.

我觉得，尤其是当模型达到新的能力水平时，采取这种逐步推进的方式是相当合理的。

I think it is quite reasonable to roll out models, especially as they reach significant new levels of capability in this way.

这和我们长期坚持的迭代式发布策略是一致的。

It fits with our long held strategy of iterative deployment.

但这个流程并不是我们认为最理想的方式。

But this isn't quite the process that we think is optimal.

接下来我们会和政府沟通，争取建立一个透明、可靠的早期访问机制，只要我们的安全措施按设计运行，就会尽快全面开放。

Now we will get with the government to attempt to get a transparent, reliable process for early access and to ensure that as long as our safeguards work as intended, can release widely.

我们会尽全力尽快把这款模型送到你手上，相信你会喜欢它。

We will work as quickly as we can to get this model in your hands, we hope you will love it.

显然，下周初我们还会就这件事展开更多讨论。

Now, obviously, there is going to be a lot more to discuss about this at the beginning of next week.

但为了给大家一点积极的展望，我们来看看OpenAI的Roon怎么说的：现在很多人说非正式的AI许可制度拖慢了创新，但没人注意到，这项影响重大的技术实际上正在以多快的速度前进。

But maybe to try to end on a note of some optimism, we turn to Roon from OpenAI who writes, Today it's popular to say the unofficial AI licensing regime is slowing down innovation or whatever, but people are not looking at the big picture of how quickly this enormously consequential technology is moving.

Mythos 的具体情况可能稍微加快了这一切，但这件事迟早会发生，早总比晚好。

The particular circumstances around Mythos may have accelerated all this slightly, but it was inevitable, and earlier is better than too late.

任何明智的决策，都会在指数曲线的早期就行动。

Any good choice will look early inside the exponential.

我觉得联邦政府能意识到这项技术的严重性，这是个积极的信号。

I think it's a positive development that the Feds understand the gravity of this technology.

模型只是被推迟个把星期，根本算不上什么大问题。

Models being publicly delayed by a week here or there is really not the end of the world.

从程序上讲，现在的做法并不妥当，但他们最终会找到办法的。

Procedurally, this is not the right way to do it, but they'll figure it out.

最让人痛心的结果，就是非美国人从此被彻底甩在前沿技术之外。

One very sad outcome will be if non Americans are just left behind from the frontier forever.

自由世界的科技秩序，以及将来非自由世界也该维持这样的秩序。

The Pax Technologica of the free world, and frankly later on, the unfree world, should be maintained.

今天就先说到这儿，但请周一一定回来。

We'll leave it there for now, but please do come back on Monday.

到时候我们还有更多内容要聊。

We are going to have a lot more to talk about then.

好了，今天这期AI周报就先到这里。

For now, that is gonna do it for today's AI weekly brief.

如果你觉得这节目有帮助，推荐给那些想了解AI最新动态的人吧。

If you wanna help the show, send this to someone who you think needs to know what's going on with AI.

我们下期再见，保重。

And until next time, peace.