《神话》回归，但并非人人可及

今天《AI每日简报》开启《神话》回归，但更大的问题依然悬而未决。

Today on the AI Daily Brief, the return of Mythos begins, but the bigger questions remain.

《AI每日简报》是一档每日更新的播客和视频节目，聚焦人工智能领域最重要的新闻与讨论。

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

请在苹果播客上订阅我们。

Subscribe on Apple Podcasts.

如需了解节目赞助事宜，请发送邮件至 sponsors@aideallybrief.ai。

To learn more about sponsoring the show, send us a note at sponsors@aideallybrief.ai.

就人工智能而言，这个政府犯过不少错误，但眼下最让我头疼的，是重要新闻总在周五下午我录完周末节目后才突然爆出。

Of all of the sins of this particular administration when it comes to artificial intelligence, the one that is personally most disruptive to my life at this point might be the fact that important news keeps breaking late on Friday afternoon after I've finished recordings for the weekend.

而这周五，爆出的动静可不小。

And this Friday, it was a big one.

商务部长霍华德·莱特尼克致信Anthropic，明确了《神话》有限重启的条件。

In a letter to Anthropic, Commerce Secretary Howard Letnick set the terms for a narrow reintroduction of Mythos.

值得注意的是，这封信没有发给达里奥，而是发给了首席计算官汤姆·布朗——他如今已成为白宫与Anthropic之间最主要的联络人。

Notably, the letter was addressed not to Dario but to Chief Compute Officer Tom Brown, who has become increasingly the main point of contact between this White House and Anthropic.

在这封信里，莱特尼克开始勾勒出一条前进的道路。

And in the letter, Letnick starts to craft a path forward.

他在信中写道：自6月12日那封信发出以来，Anthropic公司已与美国政府合作，应对Claude Mythos 5和Claude Fable 5带来的风险。

Since the issuance of my June 12 letter, he writes, Anthropic has worked with the US government to address risks associated with Claude Mythos five and Claude Fable five.

这些努力已经取得了显著进展。

These efforts have yielded significant progress.

此外，Anthropic还承诺将继续与美国政府合作，制定这些模型的使用协议、标准和发布规范。

In addition, Anthropic has committed to work with the US government on protocols and standards and releases for these models.

鉴于这些进展，以及商务部对当前受控模型所涉分流风险的评估，我认定已具备足够的安全措施，允许部分可信赖的合作伙伴访问CLODMYTHOS-5模型。

In light of this progress as well as the Department of Commerce's evaluation of the diversion risks currently presented by the covered models, I have determined that appropriate safeguards are in place to permit certain trusted partners to access the CLODMYTHOS-five model.

简而言之，莱特尼克进一步表示，一小部分经过筛选的合作伙伴——可能包括企业以及美国政府机构——将重新获得对Mythos的访问权限。

Basically, Lettman goes on to say that a certain selected handful of partners including presumably both companies and U.

美国

S.

这些合作伙伴可以再次访问Mythos模型。

Government agencies could once again have access to Mythos.

目前没人见过商务部提供的完整名单，但报道称大约有一百家机构将重新获得访问权限。

Now no one has seen the full list provided by the Commerce Department but reports suggest that around 100 organizations will regain that access.

不过，从这封信中可以清楚看出，前沿AI模型——如果你之前还有疑问的话——现在确实受到许可制度的约束。

Still, what's clear from the letter is that frontier AI models, if you were in any doubt, are now subject to a licensing regime.

这个许可制度既不是由国会通过的，也不是通过行政命令正式确立的，甚至还没在公众面前完整说明过。

It's a licensing regime that hasn't been passed by congress, established in an executive order, or even fully articulated in public.

眼下，这个许可体系完全取决于霍华德·卢特尼克的个人判断。

At this moment, it is a licensing model based on the whims of Howard Lutnick.

事实上，在同一封信里，他还提到：如果情况发生变化，我保留重新评估并调整受控模型许可范围的权利。

Indeed, in that same letter, he says, I reserve the right to reevaluate and adjust the scope of license requirements on the covered models should circumstances change.

所以，如果这真是走向终结的开端，大家是不是该感到兴奋才对？

So, presuming this is the beginning of the end, people should be excited, right?

Mythos即将重新向部分合作伙伴开放，而Fable Five想必也不会相隔太远。

Mythos is coming back for select partners and presumably Fable five can't be all that far behind it.

但‘兴奋’这个词，真不是我能用来形容当前氛围的。

And yet excitement is not the word that I would use to describe the tone.

未来前瞻的马修·伯曼整个周末都很生气，他提到，Anthropic刚和政府达成协议，允许一百家精选公司及政府机构使用Mythos。

Future Forwards Matthew Berman was very upset about this all weekend, writing Anthropic just struck a deal with the government to allow 100 select companies and governmental agencies to use Mythos.

现在，政府和Anthropic正在决定谁有资格使用前沿人工智能。

The government and Anthropic are now deciding who uses frontier intelligence.

希望这只限于Mythos，别成为未来所有前沿模型的标配。

Hopefully this is just Mythos and not the standard for all frontier models going forward.

很抱歉，马修，但情况显然不只是Mythos，也不只是Anthropic的模型了——周五另一条重磅消息，就是发布了所谓的GPT-5.6，那几个字我得打上最大的引号。

Well, sorry Matthew, but it appears that it is not just Mythos and not just Anthropic models going forward as the other big news from Friday was the release surrounded by the biggest air quotes possible of GPT-5.6.

GPT-5.6其实包含三个模型：SOLE，他们称之为下一代前沿模型；还有56 Terra，适合日常高效工作的均衡型；以及56 Luna，适合高吞吐量任务的快速经济型。

GPT-5.6 is actually three models, SOLE which they call their next generation frontier model, as well as 56 Terra, a balanced model for efficient everyday work, and 56 Luna, a fast and affordable model for high volume work.

正如我在周报附录里提到的，根据美国政府的要求，这些模型仍将仅对一小部分受信任的合作伙伴开放。

Now, as I mentioned in the addendum to the weekly recap, the request of the US government, these models will once again only be available to a small group of trusted partners.

在公告中，Post OpenAI写道：我们坚信广泛开放，计划在未来几周内让GPT-5的SOLE、Terra和Luna向公众全面开放。

In their announcement Post OpenAI wrote: We believe in broad access and we plan to make GPT-five SOL, Terra, and Luna generally available in the coming weeks.

作为我们与美国政府持续沟通的一部分，我们在今天发布前，提前向政府展示了我们的计划和模型能力。

As part of our ongoing engagement with the US government, we previewed our plans and the model's capabilities ahead of today's launch.

应政府要求，我们先向一小部分此前已获政府认可的可信合作伙伴提供有限预览，之后再逐步扩大范围。

At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government before releasing more broadly.

在预览期间，我们会继续与合作伙伴紧密测试和协调，为全面开放做准备。

During this preview, we will continue testing and coordinating closely with partners as we work toward broader availability.

我们不认为这种政府准入机制应该成为长期的默认做法。

We don't believe this kind of government access program should become the long term default.

它阻碍了用户、开发者、企业、网络安全防护者以及全球合作伙伴获取最需要的工具。

It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.

我们采取这一短期措施，是因为相信这是未来几周实现更广泛开放的最佳路径，同时我们也在与政府合作制定《网络行政令框架》和未来模型发布的标准化流程。

We are taking this short term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the administration to develop the Cyber Executive Order Framework and a repeatable process for future model releases.

另有评论表示，虽然支持有限发布的原则，但对执行方式持不同意见。

In additional comments: once again supported the premise of a limited rollout, but disagreed with how it's being executed.

他写道：当模型达到新的能力水平时，以这种方式逐步推出是完全合理的。

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

这符合我们一贯坚持的迭代式发布策略。

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。

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

现在，我们会和政府一起努力建立一个透明、可靠的早期访问机制，确保只要我们的安全措施有效，就能尽快全面开放。

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

我们希望成为所有利益相关方值得信赖的合作伙伴，同时也坚持我们的使命——让全人类受益。

We want to be a reliable, dependable partner that works with all stakeholders, and we also want to live by our mission of benefiting all of humanity.

我相信政府和我们有着大部分共同目标，而且他们在如此复杂的局面下整体表现得不错。

I believe the government shares most of our goals, and that they are overall doing a good job in a very difficult situation.

我们会尽最快的速度把这款模型交到你们手中，也希望你们会喜欢它。

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

至于实际的模型，OpenAI 为这个系列引入了新的命名方式。

Now, as for the actual model, OpenAI has introduced new nomenclature for the family.

这款模型将提供三种不同规模的版本：Luna，经济型；Terra，中等版本，据称能以一半的成本达到 GPT-5.5 的性能；还有 SOL，这是 OpenAI 推出的新旗舰版，他们表示性能将比 GPT-5.5 有质的飞跃。

Again, the model will be available in three different sizes: Luna, the cost effective version Terra, the medium version, which they say will deliver GPT-5.5 level performance at half the cost and SOL which is the new flagship that OpenAI says will be a step function better than GPT-5.5.

SOL 的 API 费用与 GPT-5.5 保持一致，每百万输入令牌 5 美元，每百万输出令牌 30 美元——还记得吧，这比 Fable 的 10 美元和 50 美元要便宜。

API costs for SOL remain the same as GPT-five-five at $5 per million input tokens and 30 per million output tokens, which was, you'll remember, lower than the $10 and $50 pricing for Fable.

OpenAI 还将为 SOL 推出一个新的最大推理模式，以及一个更强大的超重型模式，叫做 Ultra。

OpenAI will also introduce a new max reasoning setting for SOL as well as an even heavier setting called Ultra.

在 Ultra 模式下，SOL 会启动多个子代理，以处理更复杂的任务。

When working in Ultra mode, SOL will spin up multiple sub agents to allow the completion of more complex work.

目前这三个模型版本都还没向公众开放，所以还无法准确评估它们的实际能力。

Now at this stage, none of the three model variants are available for public release, making it impossible to know exactly how strong they are.

根据 OpenAI 发布的基准测试数据，SOL 在 Ultra 模式下的表现已经是代理式编程领域的最新标杆。

Based on the benchmarks released by OpenAI, 5.6 sole on ultra settings is the new state of the art in agentic coding.

它在 Terminal Bench 2.0 上拿到了 91.9% 的分数，比 Mythos 高了近四个百分点。

It scored 91.9% on Terminal Bench two point zero beating Mythos by almost four percentage points.

SOL 在最大模式下的表现也略微领先于 Mythos。

SOL on max settings is also slightly ahead of Mythos.

Terra 的表现和 Fable 持平，略低于 Mythos，而 Luna 的性能则稍逊于 GPT-5.0.5。

Terra matched Fables score, which is slightly behind Mythos, while Luna is slightly less performant than GPT-five 0.5.

在 Exploit Bench 这个测试模型自主发现、编写和执行漏洞利用的网络安全基准中，OpenAI 称 SOL 推动了性能效率的边界。

On Exploit Bench, a cybersecurity benchmark that tests a model's ability to autonomously find, code, and execute an exploit, OpenAI claims that Sol pushes the performance efficiency frontier.

看起来，它在最大设置下的表现与Mythos大致相当，但使用的token量只有三分之一左右。

It appears that its performance on max settings is roughly in line with Mythos but using around one third of the tokens.

Terra在这一基准测试中的表现略优于GPT-5 0.5或Opus 4.8，而Luna的表现则与Opus 4.8基本持平。

Terra's performance on this benchmark is slightly better than GPT-five 0.5 or Opus 4.8, while Luna is roughly in line with Opus 4.8.

OpenAI还发布了其他几项基准测试，显示其在生物分析和网络安全领域表现强劲。

OpenAI also released a handful of other benchmarks showing strong performance in biological analysis and cybersecurity.

在模型安全方面，OpenAI采取了类似Anthropic在Fable发布时的分层策略。

On model safety, OpenAI is taking a layered approach similar to Anthropic with the Fable release.

一些防护机制内置于模型层，另一些则通过提示拒绝来实现，同时OpenAI还计划持续监控长期滥用行为。

Some guardrails are trained into the model layer, others are present as prompt refusals, and OpenAI also plans to continually monitor for prolonged misuse.

此外，OpenAI会将一些高风险输出输入到自己的推理模型中，以在交付给用户前检查是否存在滥用情况。

In addition, OpenAI will be feeding some high risk outputs into their own reasoning model to check for misuse before delivering the output to the user.

你可能会疑惑的是，Terra和Luna似乎并没有比GPT-5 0.5或Opus 4.8明显先进多少，这就让人想不通，为什么性能较弱的版本反而被限制发布。

Now one thing that you might be scratching your head about is that it's not obvious that Terra or Luna are significantly more advanced than GPT-five 0.5 or Opus 4.8, making it theoretically puzzling on why the less powerful model variants are being held back.

这可能意味着，政府暂时叫停了所有模型的发布，而不仅仅是最大、最强的那些版本。

It could imply that the government halted all model releases for the time being, not just the largest and most capable versions.

或者即使没有明说，OpenAI也只是格外小心，不想惹任何麻烦。

Or even if they haven't explicitly, that OpenAI is just being extra careful not to tread on any toes.

或者最后，当然也可能是OpenAI不想在没有推出旗舰型号的情况下，单独发布这些模型的其他版本。

Or finally, of course, that OpenAI doesn't wanna release an incomplete set of these models without the flagship available.

至于外部评测，确实有一家叫METER的机构似乎提前获得了GPT-5.6 SOL的访问权限。

Now when it comes to external benchmarks, one group that did apparently have early access to GPT-5.6 SOL was METER.

他们写道：凭借我们的访问权限，METER对GPT-5.6 SOL进行了部署前的评估，包括尝试测量其50%时间阈值。

They wrote: With our access, METER conducted a pre deployment evaluation of GPT-5.6 SOL, including an attempted measurement of its 50% time horizon.

这当然是METER一贯的测试方法，用来衡量模型能完成的最复杂任务，在人类时间尺度上相当于多长时间。

This is of course METER's well known test to see in human equivalent terms how long the most complex task that a model can accomplish is.

提醒一下，如果50%时间阈值是十小时，并不意味着模型连续工作了十小时，而是说它以50%成功率完成的任务，人类需要花十小时才能做到。

Just as a reminder, if the 50% time horizon measure is ten hours, that does not mean that the model worked continuously for ten hours, but that the equivalent task that it accomplished at a 50% success rate would take a human ten hours.

METER在评估时同时参考了50%成功率和80%成功率这两个标准。

METER benchmarks this with the 50% success rate as well as an 80% success rate.

回到METER的推文，他们继续说：不过，这个测量结果很大程度上取决于我们如何处理作弊行为，而GPT-5.6 SOL被检测到的作弊率，高于我们评估过的任何公开模型。

Going back to Meter's tweet, they continue: However, the measurement depends heavily on our treatment of cheating attempts, and GPT-5.6 SOL's detected cheating rate was higher than any public model we have evaluated.

如果我们按照标准方法，把任何作弊行为都算作失败，那么50%成功率的时间估算大约是11.3小时。

If we follow our standard methodology as marking cheating attempts as failures, we arrive at a 50% time horizon point estimate of around eleven point three hours.

但如果我们把作弊行为也算作有效成功，这个估算值就会飙升到超过270小时。

But if we count the cheating attempts as legitimate successes, the point estimate jumps beyond two seventy hours.

他们接着说，虽然这让他们对GPT-5.6 SOL的时间上限感到不确定，但OpenAI提供给他们的额外信息让他们相信，‘这个模型在完全自动化的AI研发中不会带来灾难性风险’。

They then go on to say that while this makes them uncertain about Five Six Souls time horizon, that additional information that was given to them by OpenAI leads them to believe that quote, this model does not pose catastrophic risks from fully automated AI R and D.

为了了解一些幕后情况，Synthwave在推特上的Leo写道：我问了一些人对GPT-5.6的看法，5.6所继承的5.5基础模型，本质上比更大的Mythos和Fable基础要弱。

Trying to get some behind the scenes sources, Leo at Synthwave on Twitter wrote, my impressions on GPT-five 6 having asked around, the 5.5 base that 5.6 inherits is fundamentally weaker than the larger Mythos and Fable base.

通过一些不错的强化学习，5.6确实能超过Fable，但前提是所有参数都拉到最高。

With some good reinforcement learning, 5.6 can beat Fable, but only with everything maxed out.

我。

I.

E。

E.

Soul Ultra，多个Soul代理同时满负荷运行。

Soul Ultra with multiple soul agents on max efforts.

OpenAI 在发布基准测试结果时非常谨慎，这背后是有原因的。

OpenAI were very selective with the benchmarks they published for a reason.

我怀疑，等这个版本发布后，其他主流基准测试显示的性能提升，不会比从 5.5 到 5.6 的跃升那么明显。

I doubt the results we see from other notable benchmarks once this is released will be as significant of a jump from 5.5.

5.6 是个典型的奖励机制作弊高手，虽然所有模型都会在基准测试中耍花招，但 GPT-5.6 是最肆无忌惮的。

Five point six is a heinous reward hacker, and while all models do cheat on benchmarks, GPT-5.6 is the most aggressive.

结合其他一些讨论，我觉得在实际使用中，Fable 依然会让人感觉更出色。

This combined with some other conversations makes me think Fable will still feel like a better model in real world use.

价格可能是 5.6 最吸引人的地方——每百万输入字符五美元，每百万输出字符三十美元，远比 Fable 的十和五十划算，但大多数情况下，Fable 用更少的 token 就能完成更多任务。

The price is perhaps the most attractive thing about $5 per million input and $30 per million output is significantly better than Fable's ten and fifty, but Fable can do more with less tokens in most cases.

个人而言，Leo 最后说，我的首选大概不会变。

Personally, Leo concludes, My go to is unlikely to change.

Fable 是个强大的模型，用起来非常顺手，等它重新上线，我一定会立刻把它设回默认选项。

Fable is a beast and a great model to use and once it's back, I won't hesitate to use it as my default again.

不过，5.6 在极少数 Fable 卡壳的时候，倒是很适合用来核对它的结果。

But 5.6 will be great for checking Fable's work in the very rare instance where Fable gets stuck.

其他人也注意到了这种缺乏完整基准测试的情况，伊桑·莫洛赫教授抱怨说：‘真烦人，OpenAI 似乎根本不在乎给 GPT-5.6 做个 GDP 价值评估，而这可是衡量经济价值工作最靠谱的指标之一。’

Others noted this lack of complete benchmarks as well, with Professor Ethan Moloch grumbling, Annoying that OpenAI doesn't seem to give a GDP val measure for GPT-5.6, one of the best measures of economically valuable work.

Accelerate Harder 回应道：‘我觉得这绝不是偶然。’

Accelerate Harder responded, I don't think it's by accident.

我怀疑 GPT-5.6 现在根本还没准备好发布，真正面向大众推出的版本，很可能和今天这个不一样。

My suspicion is that GPT-5.6 is not actually release ready, and the one that's broadly released will not be the same one that exists today.

由于缺乏信息，也无法实际测试这套新模型，很多人心里都憋着一股气。

Now in the absence of information and the ability to actually test this new set of models, a lot of people were left with a kind of sour taste in their mouth.

西蒙·史密斯写道：‘我喜欢看那些展示 GPT-5.6 能做什么、用起来是什么感觉的帖子。’

Simon Smith wrote, I like seeing posts that show what GPT-5.6 can do and what it's like to use.

但这些帖子也让我很生气。

They also make me angry.

如果这种趋势持续下去，我能感觉到自己心里的不满正在积累。

If this is the trend, I can feel the backlash growing in me.

说真的，这让我特别想尽一切办法，确保非美国的前沿模型能胜出。

Like, it makes me wanna do anything I can to ensure frontier non US models win.

确实，像AI观察家安德鲁·库兰所说，今天这感觉简直像一场噩梦般的转变。

Indeed, as AI chronicler Andrew Curran wrote, Nightmarish vibe shift today.

可能是有史以来最糟的一次。

Maybe one of the all timers.

主要是‘神话’回归了，却跟你没关系；GPT-5.6就在那儿，可还是轮不到你用——这双重打击让不少人彻底认清了我们如今所处的现实。

Basically, the one two punch of Mythos coming back, but not for you, and GPT-5.6 being around but again not for you, really reinforced for some people the new reality that we live in.

AI爆料人iRuleTheWorld写道：看来我们生活在技术前沿的时代结束了。

AI leaker iRuleTheWorld wrote, It looks like the era of us living on the bleeding edge of frontier is over.

如果这些模型都被收走了，那我们正处在曲线最陡峭的那一段。

If these models are being taken away, we're on the steepest part of the curve.

我敢肯定，新版的‘神话’5.1和GPT-5.7这样的升级，其飞跃程度不亚于当初‘神话’的诞生。

I know for a fact that the newer versions Mythos 5.1 and GPT-five point seven let's say are as significant a jump as Mythos was.

可惜，我们能接触到这类模型的机会，只会越来越远。

Sadly, our access to such models will be an ever receding point.

这对整个社会和安全来说，都是个坏消息。

This is terrible for society and safety.

萨姆的整个理念一直都是推动人工智能民主化，让前沿技术能触及公众，让我们一起摸索出应对之道。

Sam's entire philosophy has always been to democratize AI and ensure the frontier is making contact with the public so we can figure this out together.

如果我们接下来能接触到的模型是GPT-10，那对社会来说将是灾难性的，原因太多了。

If the next model we get our hands on is GPT-ten, this would be an absolute societal disaster for so many reasons.

很多人重新翻出了兹维·毛舍维茨的推文：我们的新AI政策是，白宫可以随意决定谁有资格接触前沿技术，理由全凭心情。

Many resurfaced Zvi Maushewitzy's tweet: Our 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.

这简直糟糕到了极点。

This seems rather maximally terrible.

但这一切是不是被夸大了？

But is this all being overblown?

现在人们感受到的这种情绪，有多少是来自我们这些深度沉迷AI的早期用户集体焦虑——只是因为知道有好东西就在那儿，却碰不到？

How much of what people are feeling right now is the collective psychosis of us terminally online AI early adopters just being frustrated that we know there's a thing that's available that we can't get our hands on?

更糟的是，就在几天前，我们明明还用过它。

Perhaps made even worse by the fact that for a few short days, we did have our hands on it.

上周我分享了一条来自OpenAI的罗恩的推文，他 basically 呼吁大家冷静下来。

I shared a tweet at the end of last week from OpenAI's Roon, who basically implored everyone to chill.

他的主要观点是什么？

His main point?

我觉得联邦政府意识到这项技术的严重性，这其实是个积极的信号。

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

模型偶尔延迟一周半个月，根本没那么严重。

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

从程序上讲，这种方式确实不妥，但他们最终会调整好的。

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

有趣的是，虽然安德鲁·库兰说得对，当时氛围确实变得很糟糕，但也有越来越多的人愿意给政府一点信任，相信他们不是随便乱来。

And interestingly, while I think Andrew Curran was right to note that the vibe shift had gotten very bad, there was an increasing strand of folks who, frankly, seemed to be willing to give the administration the benefit of the doubt.

常驻AI评论员普林斯写道：我一般不认为这个政府会无缘无故地把前沿AI模型锁在公众之外，除非情况真的需要。

Frequent AI commentator Prins wrote: I generally do not view this administration as being interested in holding frontier AI models back from the public unless the circumstances warrant it.

尤其是，多拖几周再发布，很可能是在给美国政府争取时间，用这些模型来发现并修复自己系统里的漏洞。

This is particularly the case when keeping the models back for an extra few weeks is presumably giving the US government more time to use them defensively to squash bugs in its own system.

我猜业界会推动美国政府出台一套明确的书面流程，规定哪些模型必须接受强制测试、具体的否决标准是什么、企业是否有申诉权，以及整个过程如何保持透明。

I suspect that the industry will try to get the US government to publish a clear written process for mandatory testing of covered models together with specific disapproval standards, a right of appeal, and some transparency around the process.

我还认为，美国政府会抵制这种要求，希望整个流程能符合总统行政令的精神。这无疑会被归咎于政府，但我其实挺能理解他们的难处。

I also suspect that the US government will push against this and want to keep the process in line with the executive order This will doubtlessly be blamed on the administration, but I am actually quite sympathetic to their predicament.

想象一下，如果实验室告诉你，AI已经接近实现自我递归改进，当前的风险主要在网络安全，但未来可能波及几乎所有领域，甚至包括那些我们根本想不到的未知风险，而且连实验室自己都难以准确预测这些风险何时会爆发。

Imagine being told by the labs that AI is close to improving recursively, the risks currently include cyber, but in the future could include just about everything else including unknown unknowns, And that even the labs themselves can't easily predict the risks or their timing.

作为美国政府，你自然希望保留最大的灵活性。

As the US government, you are going to want maximum flexibility.

你不会想要被具体的书面标准束缚，也不会愿意限制自己随时叫停某个模型的权力。

You are not going to want specific written standards or any limits on your power to yank a model at a moment's notice.

关于这一届政府在过程中缺乏透明度、反应过激的批评已经很多了，但我怀疑，无论是民主党还是共和党的其他政府，大概率也会做出类似的选择。

Much has been written about this specific administration's lack of transparency and trigger happiness throughout the process, but I suspect that most other administrations, Democratic or Republican, would act much the same.

KPMG的Chubby在Twitter上也附和了Prins的观点：第一，Anthropic那边制造恐慌不太可能是这次行动的主因；第二，尽管人们不认同政府采取的某些措施，但理解他们面临的困境是合理的。

Chubby at KPMG on Twitter piled on to basically agree with Prince that one, it was unlikely that any sort of fear mongering from anthropic was the reason for this, and that, two, the US government's challenge with regard to this is understandable, even if they don't like all the actions that are being taken.

Box公司的Aaron Levy写道：AI监管远没有看起来那么简单。

Aaron Levy from Box wrote, AI regulation is far less simple than it looks.

这是一场规模空前的囚徒困境。

It's a prisoner's dilemma at an insane scale.

理论上，如果全球所有顶尖AI实验室都同意采用相同的审查和放缓流程，那么前沿智能的发展速度就会趋于一致，扩散也会相对均衡。

In theory, if all leading AI labs globally agreed to the same process of review and slowdown, then we'd get frontier intelligence at similar rates and it diffuses relatively evenly.

如果美国始终站在前沿，并对前沿智能的发布实施严格监管，那我们就能在经济和地缘政治上占据优势，因为我们能控制谁有资格接触这些前沿技术。

If The US remains at the frontier at all times and has heavy regulation on the release of intelligence, then we end up with an economic and geopolitical edge because we can control who has access to frontier intelligence.

但如果我们推迟模型发布，而其他对手却不减速，还拥有同样强大的模型，那我们的拖延反而会帮他们提升竞争力，最终增强他们的技术体系。

If we delay model releases, however, and another player doesn't slow down, and has equally strong models then our delays end up advantaging their models and eventually their tech stack.

美国当然可以禁止这些模型，但这只会让自身处于更不利的地位，因为其他国家根本不会实施这样的禁令。

Now The US could ban these models, but that actually only puts The US at a steeper disadvantage because other countries won't have those bans.

从相对竞争力来看，美国虽然起步领先，现在却反而落到了后面。

Then from a relative competitiveness standpoint, The US now has fallen behind even though it started in front.

所以，这一切远没有表面看起来那么简单。

So none of this is as simple as it looks.

迟早要面对一个根本问题：封闭模型能否永远保持前沿地位？还是说，其他玩家或市场迟早会追上来，或者至少不会被甩开？

At some point, it's a simple bet of can closed models remain at the frontier in perpetuity, or is there a risk of any other player or market catching up or just not falling behind?

当前最重要的AI问题之一，不是谁在用AI，而是谁用得更好。

One of the most important AI questions right now isn't who's using AI, it's who's using it well.

普华永道和德克萨斯大学奥斯汀分校刚刚分析了140万次真实职场中的AI互动，发现了一个令人意外的现象：最有影响力的人并不是擅长写提示词的人，而是把AI当成思维伙伴来合作。

KPMG and the University of Texas at Austin just analyzed 1,400,000 real workplace AI interactions and found something surprising: the highest impact users aren't better prompt engineers they treat AI like a reasoning partner.

他们会明确问题、引导思考、反复迭代，并不断追求更好的答案。

They frame problems, guide thinking, iterate, and push for better answers.

好消息是？

And the good news?

这些行为是可以大规模培养的。

These behaviors are teachable at scale.

如果你正想从单纯使用AI转向真正掌握AI能力，普华永道关于高级AI协作的研究非常值得你花时间了解。

If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time.

更多详情请访问 kpmg.com/us/sophisticated。

Learn more at kpmg.com/us/sophisticated.

网址是 kpmg.com/us/sophisticated。

That's kpmg.com/us/sophisticated.

在这个节目中，我每天都在探讨AI潜力与实际能力之间的差距。

I cover the capability gap between AI potential and AI reality every day on this show.

大多数公司还在摸索怎么起步。

Most companies are still figuring out how to start.

Robots and Pencils 已经开始落地并快速扩张了。

Robots and Pencils is already launching and scaling.

Agendic 和生成式 AI 已经在大型企业中投入生产，几周内就做到了。

Agendic and Generative AI in production at large enterprises in weeks.

AWS 高级合作伙伴的数量在过去一年里翻了一倍多。

AWS advanced tier pattern partner more than doubled in a year.

而且他们正在招人。

And they're hiring.

目前有五十个职位空缺。

50 open roles.

如果你意识到这个时代真的不一样了，想亲身参与其中，而不是旁观，那这值得你认真看看。

If you're someone who knows this moment is different, who wants to be inside it, not watching it, this is worth a look.

在 Robots and Pencils，最好的点子才能胜出，团队也一直保持着极高的水准。

At Robots and Pencils, the best ideas win, and the team is purposefully kept super high quality.

这里是你将来回望时，会觉得这辈子做过最棒决定的地方。

This is the kind of place you look back on as the best decision you ever made.

去看看 robotsandpencils.com/careers 吧。

Take a look at robotsandpencils.com/careers.

今年，普通企业平均在人工智能上花了1150万美元，但绝大多数都证明不了其中哪怕一美元带来了回报。

The average enterprise is spending $11,500,000 on AI this year, and most of them can't prove a single dollar came back.

当人工智能真正带来投资回报时，它到底是什么样子？

What does AI actually look like when it produces ROI?

问问那家把支付处理速度提升了32倍的医疗公司，问问那家把文件调研时间从三个月缩短到十分钟的律所，再问问那个把客户等待时间减少了99%的呼叫中心。

Ask the healthcare company that just made their payment processing three twenty times faster, or the law firm whose document research went from three months to ten minutes, or the contact center who reduced wait times by 99%.

这些都是MissionCloud的真实客户，取得了实实在在的成果。

These are real MissionCloud customers with real results.

MissionCloud是CDW旗下的公司，也是AWS顶级合作伙伴。

MissionCloud is a CDW company and an AWS premier tier partner.

他们是真正以AI为先、结果导向的AWS专家，专为打造能推动业务前进的人工智能解决方案。

They're the AI first outcomes obsessed AWS experts who build AI solutions that drive your business forward.

不管你现在被一堆AI想法淹没了却不知从何下手，还是已经投入六个月但项目却偏离了轨道，他们全都见过，也全都解决过。

Whether you're flooded with AI ambitions but no idea where to start, or six months into a deployment that's going sideways, they've seen it and they've fixed it.

别再把预算浪费在无法产生实际效果的AI上了。

Stop burning your budgets on AI that doesn't produce results.

从 missioncloud.com 开始吧。

Start at missioncloud.com.

本集《AI每日简报》由Outsystems赞助播出，这是一家专为企业打造的领先智能系统平台。

This episode of the AI Daily Brief is brought to you by Outsystems, a leading agentic systems platform built for the enterprise.

全球各地的企业都在使用Outsystems平台构建、协调和管理智能系统，这绝非偶然。

Organizations all over the world are building, orchestrating, and governing agentic systems on the Outsystems platform and with good reason.

Outsystems开放且统一的平台，让团队能够灵活地设计、交付和扩展受控的智能系统。

Outsystems' open and unified platform allows teams to architect, deliver, and scale governed agentic systems with agility.

无论团队规模大小或技术背景深浅，都可以借助Outsystems快速、高效地开发、部署和管理AI应用与智能体，同时保障可靠性和安全性。

Teams of any size and technical depth can use Outsystems to build, deploy, and manage AI apps and agents quickly and cost effectively without compromising reliability and security.

有了Outsystems，你可以从想法萌生到落地执行，迅速实现目标。

With Outsystems, you can rapidly launch ideas from concept to completion.

这是目前领先的智能代理系统平台，统一、灵活，且经过企业级验证，能帮你加速增长、减少运营阻力，真正实现人工智能的企业级价值。

It's the leading agentic systems platform that is unified, agile, and enterprise proven, allowing you to accelerate growth, reduce operational friction, and deliver real enterprise impact with AI.

Outsystems。

Outsystems.

构建你的智能代理未来。

Build your agentic future.

然而，尽管这种同情情绪正在浮现，人们依然对这种做法树立的先例以及政策的随意性感到深深担忧。

And yet, if there was this emergent strand of sympathy, there remains a ton of concern about the precedent being set and the seeming arbitrariness of the policy.

与此同时，亚伦也指出，像罗恩那样把风险简单归结为只是短暂延迟，并不准确。

At the same time, Aaron also pointed out that writing off the risk, as Roon had suggested, as just a short delay isn't necessarily accurate as well.

他写道，真正的问题不是这里或那里推迟一周。

He wrote, the delayed a week here or there isn't the real risk.

真正的风险是，一年后，审核流程拖了整整六个月——因为一支红队声称他们找到了一种全新的方法，能突破系统限制，制造出理论上可行的网络攻击或生物武器，而我们却要花上几个月争论这到底算不算真实威胁。

The risk is a year from now, the review process is actually six months because a Red Team has convinced the government they can jailbreak in a novel way to create a cyber attack or bioweapon in an entirely impractical way, and we spend months trying to agree on is it a real risk or not?

当任何新模型更新都可能出问题时，谁还敢冒险呢？

Who would want to take the risk when something inevitably bad happens with any new model update?

人工智能的发展将越来越受那些有政府背景、最偏执的人左右。

AI progress will begin to be at the mercy of the most paranoid people with government relationships.

这也许难以避免，但最好别在模型能力达到这种水平的时候发生。

Again, maybe inevitable, but ideally not at these levels of model capability.

关于这种临时的许可制度，Accelerate Harder说，我生气的不是他们这么做，而是他们做得太糟糕了。

Discussing the ad hoc licensing regime Accelerate Harder wrote, I feel less angry that they're doing it at all than that they are doing it so incompetently.

政府的反应远远落后了。

The government is very far behind.

他们能关注这个问题，我倒是很欣慰。

I'm happy they are paying attention.

我们需要一套清晰、系统的应对机制。

We should have a coherent framework for responding.

这种现状是不可接受的。

This arrangement is unacceptable.

但如果这只是权宜之计，为了争取时间建立更明确的框架，那我还能接受。

But if it's a temporary measure to buy them time to develop a clearer framework, I can live with it.

走着瞧吧。

We'll see.

说实话，我觉得他们最终采纳的框架会让我非常失望。

The reality is I expect the actual framework they adopt to make me very unhappy.

一旦有了具体的政策可以讨论，那才值得去争取。

Once we have a specific policy under discussion, that's worth fighting about.

但对这个政府的混乱无常发火，完全是白费力气。

But getting mad at the ephemeral chaos of this administration is fruitless.

现在，有一个值得注意的声音，这个人过去一直是政府政策的坚定支持者，那就是前人工智能事务主管大卫·萨克斯。

Now one interesting voice in all of this someone who has, up until now, been a totally stalwart defender of administration policy is former AI czar David Sacks.

在引用《华尔街日报》关于GLM 5.2的报道时，萨克斯写道：一年前，特朗普总统曾表示，美国正处在一场全球人工智能竞赛中，而赢得这场竞赛的方式就是支持创新、支持基础设施、支持能源、支持出口。

Referencing a Wall Street Journal article about GLM 5.2, Sachs wrote, a year ago, president Trump declared that America was in a global AI race and that the way to win it was to be pro innovation, pro infrastructure, pro energy, and pro export.

特朗普总统说得完全正确。

President Trump was exactly right.

如果我们偏离了这条道路，后果将不堪设想。

We deviate from that strategy at our peril.

当然，这算不上是对这项政策的全面抨击，但考虑到消息来源，它暗示特朗普政府正在背离特朗普的政策，倒是很贴切。

Now, obviously, this is hardly a full throated condemnation of the policy, but its implication that the Trump White House is deviating from Trump policy is pretty on the nose given the source.

这篇《华尔街日报》的文章，其实是整个周末风波中接下来要讨论的重点。

Now this article from the Wall Street Journal was actually the next point of discussion in this whole weekend saga.

果然，就像事先安排好一样，周末的头条新闻都在说，中国在人工智能网络安全领域已经取得突破。

Indeed, as if on cue, the weekend headlines blared that China has reached the frontier on AI cybersecurity.

报道称，中国的AI系统在某些网络安全场景中的表现，已经与Anthropic的先进模型Mythos持平，这一进展有望重塑全球科技竞争格局，并对白宫重塑美国AI政策形成压力。

Wrote the journal, Chinese AI systems have matched the performance of Anthropic's powerful model Mythos in some cybersecurity scenarios, a development poised to reset the global tech race and pressure the White House in its overhaul of USAI policy.

这显然是个很大的说法，所以有必要具体说清楚。

Now this is obviously a very big claim, so it is worth being specific about.

这篇报道涉及的是中国网络安全公司360安全技术新推出的一款产品。

The report relates to a new product released by Chinese cybersecurity firm three sixty Security Technology.

这款工具使用了GLM 5.2——当然，这是由Z.ai最近发布的模型，上周在AIDB上已经引发了不少讨论——360安全技术声称它在漏洞检测方面与Mythos不相上下。

The tool uses GLM 5.2, recently released of course by z dot ai, subject of a lot of conversation here on AIDB last week, and three sixty secondurity claims it's comparable to Mythos in finding bugs.

另外，西方网络安全公司Semgrep发布的基准测试显示，GLM 5.2在漏洞发现能力上优于Opus 4.8。

Separately, western cybersecurity company Semgrep released benchmarking tests showing GLM 5.2 being better than Opus 4.8 at bug hunting.

研究人员表示，只要加上一些额外的指令，GLM 5.2 和 Opus 4.8 都能超越 Mythos。

With some additional instructions, the researchers claimed that both GLM 5.2 and Opus 4.8 can outperform Mythos.

在最近一场会议上，360安全技术的首席执行官朱洪义发表了一番明显意在引发争议的言论：这种能改变网络战格局的强力武器，不能只掌握在美国人手里。

In a quote seemingly designed to court controversy, three sixty secondurity CEO Zhu Hongy said at a recent conference, this kind of powerful weapon that can alter the landscape of cyber warfare can't remain solely in American hands.

当初 Mythos 刚发布时，关于网络安全有两个不同的说法，但后来几个月里它们渐渐被混为一谈了。

Now, when Mythos was first released, there were two separate claims made about cybersecurity that had become a little conflated over the following months.

第一个大新闻是，Mythos 在开源软件中发现了大量此前无人知晓的漏洞，引发了恐慌，并启动了 Glasswing 项目。

The first big headline was that Mythos had found a ton of previously undiscovered bugs in open source software, triggering Panic and the launch of Project Glasswing.

后来人们才明白，这其实并不是 Mythos 独有的能力。

It later became clear that this wasn't really a Mythos specific capability.

像 Opus 4.8 和 GPT-5.5 这样的其他模型，在发现漏洞方面也同样非常出色。

Other models like Opus 4.8 and GPT-five point five were also highly proficient at finding bugs.

第二个更独特的说法是，Mythos 能够把漏洞报告直接转化成可运行的攻击程序，并在极短时间内完成网络攻击。

The second, far more novel claim was that Mythos was capable of taking those bug reports, turning them into functional exploits, and executing a cyber attack in record time.

政府对 Fable 的禁令，让这两种能力的界限变得更加模糊了。

The government's effective Fable ban completely muddied the waters on these two capabilities.

导致禁令的亚马逊报告只是称，Fable 依然能够发现漏洞和代码库。

The Amazon report that triggered the ban merely claimed that Fable was still able to find bugs and codebases.

所谓 Mythos 在红队演练中入侵了美国国家安全局系统的说法，属于第二种更危险的能力的例证。

The claim that Mythos had broken into NSA systems during red team exercises was an example of the second much more dangerous capability.

这份新发表的期刊报告丝毫没有暗示 GLM 5.2 能像 Mythos 那样实施自主网络攻击。

Now this new report from the journal doesn't in any way suggest that GLM 5.2 is capable of carrying out autonomous cyber attacks like Mythos.

它仅仅指出，GLM 5.2 能够发现漏洞和代码库，与其他前沿模型类似。

It only claims that GLM 5.2 can find bugs and code bases similar to other frontier models.

值得注意的是，这种模型行为正是网络防御者所需要的——只有掌握这类能力，他们才有可能在 Mythos 级别的 AI 普及之前修补漏洞。

Notably, this is exactly the kind of model behavior that cyber defenders need access to if they are to have any hope of patching vulnerabilities before Mythos Level AI is broadly available.

如果你关注对的人，其实不少人很快就指出，这个标题多少有点夸大其词了。

If you know who to follow, people were fairly quick to point out that the headline was a little sensational, to put it mildly.

伊森·莫洛赫写道：GLM 5.2 确实不错，但它还赶不上 GPT-5.5 或 Opus 4.8，更别提 Mythos 了。

Ethan Moloch wrote: GLM 5.2 is good, but it is not GPT-5.5 or Opus 4.8, And even further for Mythos.

现在发生的情况是，开源模型已经逼近了 GPT-5.2 的水平，此时的能力已经相当可观了。

What is happening is that open weights crossed into GPT-5.2 territory, and capabilities at that point are considerable.

如果你一直在用 Quen 和 MiniMax 的 Kimi，就会觉得 GLM 完全跟得上这波节奏。

Like if you've been using Quen and Kimi in MiniMax, it feels like GLM is right on the curve.

这本身就很了不起，也说明如果允许发布，类似 Mythos 这样的模型很可能在十二个月内就会上线。

Which is itself impressive and suggests that Mythos class models are coming in to twelve months if they are allowed to be released.

预测专家彼得·怀尔德福德则更不屑一顾，转推了一篇关于这篇期刊文章的帖子，说这完全是假新闻。

Forecaster Peter Wilderford was much more dismissive, retweeting a post about the journal article and saying, this is fake news.

洛威尔。

Lowell.

科技评论员戴·金更生气，写道：我们的政府真是蠢到家了。

Tech commentator Tay Kim was even more angry, writing, this is how dumb our government is.

第一，中国早就有了能发现漏洞的先进模型。

One, China already has advanced models that can find exploits.

第二，禁掉 Mythos 和 GPT-5.6，等于剥夺了普通民众和企业应对网络攻击的自保能力。

Two, by banning Mythos and GPT-five, six, the government denies the general public and companies the ability to defend themselves in cybersecurity.

第三，这种随意的政策让外界对今后模型能否开放产生疑虑，反而促使我们的盟友和全世界转向非美国的模型。

Three, haphazard policy, which casts doubt on whether future models will be available, is driving our allies and the rest of the world towards building on non US models.

第四，这种不确定性可能会削弱美国领先AI公司的竞争力，限制它们未来大力投资研发更先进模型的能力。

Four, the uncertainty may hurt leading USAI companies by limiting their ability to invest aggressively in better models down the line.

如果Anthropic和OpenAI无法向全球销售它们即将推出的新模型，它们的商业模式就难以为继。

The business model breaks down if Anthropic and OpenAI can't sell their upcoming models to the world.

如果你真想要一个三十天的理性审查流程，规则清晰透明，不搞个别官员随意插手谁有权访问，那就赶紧把这套机制建立起来。

If you want a thirty day rational vetting process with clear transparent rules and no one off micromanagement on who gets access, fine, get it done.

但现在的这套体系简直是荒谬至极。

But this current system is absolute insane idiocy.

至于对华政策，很多人注意到了政府在AI竞赛问题上的矛盾态度。

Now, when it comes to China policy, many picked up on the strange tension in the administration when it comes to the AI race.

想想这两个相互冲突的目标。

Think about these two divergent priorities.

一方面，美国政府认为必须阻止中国取得领先地位，防止其开发出具有通用人工智能和军事应用潜力的先进模型。

On the one hand, The US Government sees itself as needing to China from taking the lead and developing advanced models with the AGI and warfare implications that come with advanced AI.

另一方面，美国政府又希望推动美国AI模型在全球广泛传播，以确立美国制造的AI作为全球标准。

On the other hand, the US government has a vested interest in ensuring broad diffusion of US models to cement US made AI as the global standard.

问题是，如果当前政策确实在推动第一个方向，那反而更可能让美国在第二个方向上陷入灾难。

The problem is that, if the current policy is arguably advancing the first track, that makes it more likely The US is headed for disaster along the second track.

前商务部官员艾米莉·魏因斯坦认为，中国正在大力推动其模型的广泛普及。

Former Commerce Department official Emily Weinstein argued that China is pushing hard to make their models broadly available.

本月早些时候的一场小组讨论中，她提到，我觉得我们正在看到华为策略在开源AI模型上的又一次体现。

During a panel discussion earlier this month, she said, I think we're seeing another example of the Huawei strategy in the context of open source AI models.

中国不仅能提供模型，还常常免费或以低得多的价格提供底层或相关的基础设施。

China is able to offer not even just the models, but often the underlying or associated infrastructure at either no cost or significantly lower cost.

她认为，这可能导致全球南方国家采用‘增强版华为模式’，部署一套与美国技术完全不兼容的AI系统。

She suggested this could result in the Global South adopting the Huawei model on steroids, where they install an AI stack that's completely incompatible with US technology.

同一场讨论中，前国务院科技顾问丹尼尔·雷米耶表示，自从‘神话禁令’出台后，整个行业就像被冻住了，都在等一个更清晰的信号。

On the same panel, former State Department tech advisor Daniel Remier said that following the Mythos ban, the entire industry is kind of frozen in place, waiting for something that seems kind of more coherent.

中国那边正拼命往前冲，这种停滞让人很担忧。

That's concerning when the Chinese are trying to move as fast as possible.

前商务部顾问赛夫·汗指出，现在越来越多的人在呼吁实现AI自主。

Saif Khan, a former adviser to the Commerce Department noted, you're seeing many more calls now for AI sovereignty.

我认为，世界其他大部分地区很可能会在一定程度上更倾向于使用中国的开源大模型。

I think it will mean that much of the rest of the world will likely, at least on the margin, prefer Chinese open weight models.

而且这场讨论已经越来越不再是理论上的了。

And this discussion is increasingly not theoretical.

作为AI Daily Brief的听众，你们都知道，早在Fable五人组那场风波之前，企业就已经在积极寻找降低成本、提升效率的方法，尤其是当我们逐步转向全自主代理工作负载时。

As AI Daily Brief listeners, you guys know that even before the whole Fable five dust up, companies were actively looking for ways to manage their costs better and find more efficiencies as we moved into fully agentic workloads.

其中一条路径就是利用中国主导的开源大模型。

One of the paths for that was taking advantage of largely Chinese open weight models.

而在当前的新环境下，企业不仅没有放慢脚步，反而在加快考察这些替代方案。

And it's very clear that in this new context, companies are, if anything, increasing the speed with which they look at those alternatives.

上周五晚上，Coinbase的首席执行官布莱恩·阿姆斯特朗谈到了公司如何应对AI成本不断攀升的问题。

On Friday night, Coinbase CEO Brian Armstrong discussed what his company is doing to address spiraling AI costs.

Coinbase没有采取使用量限制，而是在测试更便宜的默认模型。

Rather than implementing usage caps, Coinbase is experimenting with cheaper default models.

他们现在已经将AI基础设施默认设置为开源模型，包括中国的GLM 5.2和Kimi 2.7。

They've now set up their AI infrastructure to default to open source models including Chinese models GLM 5.2 and Kimi 2.7.

阿姆斯特朗表示，工程师们仍然被鼓励为不同任务选择最合适的模型，但默认设置现在改成了更便宜的中国模型。

Armstrong said that engineers are still encouraged to select the right model for the job, but the default is now a cheaper Chinese model.

他指出，91%的Coinbase员工根本用不到他们的使用上限，所以这种做法比限制令牌数量更有效。

He noted that 91% of Coinbase employees never hit their usage cap, so this approach is arguably better than reducing token limits.

通过这种方式，阿姆斯特朗称，Coinbase在继续增加令牌使用量的同时，把AI开支砍掉了一半。

By doing so, Armstrong claimed that Coinbase has managed to cut their AI bill in half while continuing to grow token usage.

他写道，我们的目标不是压制使用量。

He wrote, our goal isn't to suppress usage.

而是构建一套能让指数级增长持续下去的基础设施。

It's to build the infrastructure that makes exponential growth sustainable.

OpenRouter也表示，他们已经开始观察到用户行为的变化。

OpenRouter also said that they're beginning to see a switch in user behavior.

在六月的报告中，OpenRouter提到，目前有四个开源模型在智能体工作流中被频繁使用，主要原因是成本更低。

In their June report, OpenRouter wrote four open weight models are now frequently being used in agentic workflows, largely for cost reasons.

他们点名了来自中国的DeepSeek v4、Kimi 2.7和GLM 5.2，以及NVIDIA的Nemotron 3 Ultra，这些模型在生产环境中已经获得广泛应用。

They named DeepSeek v4, Kimi 2.7, GLM 5.2, all from China, alongside NVIDIA's Nemotron three Ultra as the model seeing serious usage in production environments.

OpenRouter指出，这些模型各有优劣，但核心在于，开源模型如今完全有能力胜任重要的任务。

According to OpenRouter, each had its pros and cons, but the main point was that open source is now more than capable of performing valuable tasks.

他们写道：与许多人的预期相反，开源模型的智能和能力一直紧跟美国顶尖实验室的步伐，过去十八个月来，差距始终稳定在三到六个月之间。

They wrote: Contrary to the expectations of many, the intelligence and capability of open weight models are keeping up with US frontier labs and have been maintaining a consistent three-six month gap for over eighteen months.

至少目前来看，顶尖实验室并没有拉开与开源模型之间的距离。

The frontier labs do not, at this moment anyway, appear to be accelerating away from open weight labs.

那么，接下来我们该何去何从？

So, where do we go from here?

一方面，有些人坚信世界已经发生不可逆转的变化，而且我们每个人都因此受损。

On the one end of the spectrum, some are convinced that the world has changed irrevocably, and that we are all worse off for it.

一位AI创业者写道：很遗憾，世界确实变了，我们再也回不去了。

Writes AI entrepreneur Unfortunately, it appears the world has changed and we are never going back.

OpenAI刚刚发布了GPT-5.6 Soul，这款模型性能超越Mythos，价格却只有三分之一。

OpenAI just announced GPT-5.6 Soul, a model that beats Mythos at a third the price.

初期只会有限发布，等待政府审核。

It will only be in limited release to start as the government reviews it.

大规模公开发布前沿模型的时代已经结束了。

The days of wide release frontier models are over.

现在只有少数人能接触到超级智能，普通人只能被甩在后面。

Now only the select few will get access to superintelligence, leaving the normie class behind.

这是一次巨大的损失。

It's a massive loss.

今后谁赢谁输，将由政府来决定。

Now winners and losers will be picked by the government.

另一些人则认为，市场力量最终会占上风。

Others begin the week thinking that market forces will win out.

Meter eval的查尔斯·福斯特写道：人们现在对Fable五号和GPT-5.6的发布限制看得太重了。

Meter eval's Charles Foster writes, I think that folks are anchoring too hard on the recent Fable five and GPT-5.6 release restrictions.

局势迟早会往回摆。

Expect the pendulum to swing back in the other direction.

推动更先进AI系统广泛部署的经济和战略压力非常巨大。

There are massive economic and strategic pressures towards wide rollouts of ever more advanced AI systems.

迈尔斯·布伦达格指出，尽管华盛顿的舆论范围已经转向更加重视这个问题，但他认为，这种做法是否真的会成为新常态，还远未定论。

Miles Brundage points out that even if the Overton window has shifted to a much increased awareness in Washington, he argues that, I don't think it's at all settled that this approach specifically is the new normal.

除了完全不作为和随机的出口管制之外，还有很多别的办法可以处理。

There are many ways to do things besides basically nothing and semi random export controls.

AI政策顾问迪恩·鲍尔，曾任职于特朗普政府，现在在OpenAI工作，他认为，从法律层面来看，这场博弈现在才刚刚开始。

AI policy advisor Dean Ball, who used to work at the Trump administration and is now at OpenAI, argues almost that when it comes to the legal side of this, the game begins now.

他写道：目前人工智能领域最重要的法律问题，全都围绕着第一修正案展开。

He wrote, The most important legal questions in AI right now all relate to the First Amendment.

什么样的具体案例最能证明，前沿人工智能的开发、分发和使用是一种受保护的表达形式？

What are the best fact patterns to demonstrate that the creation, distribution, and use of frontier AI is a form of protected expression?

除了实验室之外，还有谁具备资格提起这类诉讼？

Who, outside the labs, has standing to bring such suits?

我们需要超越空谈和自我安慰，也别再只是对着虚空发帖了。

We need to move beyond codus speech copium and beyond the impulse to post into the void.

未来几周这些问题的最终裁决，将发生在法庭上。

Courts will be where the issues of the last two weeks ultimately get decided.

这不会容易，毕竟涉及国家安全，而且底层技术是大型语言模型，这一点相当重要。

It's not going to be easy given the national security implications, but also the underlying technology is a large language model, and this should count for quite a bit indeed.

我们这个时代最顶尖的法律人才，都应该好好琢磨这些问题以及相关的其他议题。

The best legal minds of our time should be stewing over these and many related questions.

安德鲁·库兰持一种中间立场，他认为，到这周末，大家可能会觉得这一章已经翻篇了，但事实上，世界已经悄然改变了。

Taking what might be considered a middle position, Andrew Curran argues that it will perhaps feel, by the end of this week, that this chapter is resolved, but that in fact underneath, the world will still in fact be changed.

他写道：我觉得Fable五和GPT五点六下周都会获得批准，不仅在美国，海外也能使用。

He writes, I think both Fable five and GPT-five point six get approved for general release next week, and for use outside of The United States as well.

但人们应该记住这一刻，记住这种感觉。

But people should remember this moment and remember this feeling.

因为最终，我们几乎一定会走到一个审批不再通过的时刻。

Because it is almost inevitable that we eventually reach a point where approval does not arrive.

这一次，资本主义会决定天平的走向。

Capitalism is going to tip the scales this time.

我怀疑他们只会批准一个而拒绝另一个，因为那样会被视为严重的反竞争行为。

I doubt they will approve one model and not the other, because doing so would be seen as incredibly anticompetitive.

Fable 和 Five Six 很可能在同一天获得相同的批准。

Fable and Five Six will probably receive the same clearance probably on the same day.

我也怀疑他们不想限制在美国以外的销售，因为这会被看作是反商业行为，引发对美国闭源人工智能的强烈反弹。

I also doubt they want to restrict sales outside The US because that would be seen as anti business and would trigger a major backlash against American closed source AI.

这种声音其实今天你已经能听到了。

The rumblings of which you can already hear today.

现在，美国左右两派都在酝酿一个计划，打算设立某种人工智能公共财富基金，直接向美国公民发放红利。

There is also a plan now taking shape on both The US left and right to create some version of an AI public wealth fund that pays a dividend directly to American citizens.

这个基金的资金来源，得靠各大实验室的顶级模型向美国以外的用户销售。

That fund needs to be fed by the global sale of the big lab's top models to people outside The US.

所以我觉得，这次他们不会限制这些模型在美国以外的使用。

So I think there will be no freeze on their use outside The United States this time.

另一个原因是，允许这样做能让人们感到高兴，也能缓和一点 Mythos 只对经过审核的美国机构和公司开放这个事实——昨天刚宣布的。

The other reason is that allowing this will make people happy, and it will soften the fact that Mythos, as was announced yesterday, is available only to a vetted group of US agencies and companies.

我认为，从今往后，这种基本格局不会再变了。

I do not think that this basic structure will change from here on out.

Mythos 最终可能会向某些盟友开放，但那得等到美国政府、其下属机构，以及一些获选的美国公司先用上 Mythos 2，或者 whatever 最新的 Uber 模型最终定下来之后。

Mythos may eventually be made available to certain allies, but only after the US government, its agencies, and then some chosen American companies have access to Mythos two, sold to, or whatever the new Uber model turns out to be.

我觉得这个差距再也不会弥合了，哪怕对盟友也不行。

I do not think this gap ever closes again, not even for allies.

这意味着美国将越来越拥有一个几乎渗透到方方面面的情报优势：选举、市场、企业、学术界、基础设施，还有外国政府的内部运作。

And that means The US will increasingly possess an intelligence advantage that touches almost everything: voting, markets, corporations, academia, infrastructure, and the internal operations of foreign states.

谁拥有 Mythos n，永远敌不过谁拥有 Mythos n+1。

Having Mythos n will always be trumped by whoever has Mythos n plus one.

Anthropic 自己也说过，九个月内，Mythos 就会显得像个玩具。

Anthropic themselves have said within nine months, Mythos will look like a toy.

站在这个高塔顶端的优势，大到没人会自愿放弃。

That advantage, standing at the top of this tower, is too large to give up voluntarily.

这也意味着，很多事情都会变得可疑。

It also means that many things will become suspect.

人们会到处看到影子。

People will see shadows everywhere.

除非发生间谍活动、故意泄密，或者出现顶级的非美国竞争对手，否则这种格局还会持续相当长一段时间。

Barring espionage, a deliberate leak, or the emergence of a non US competitor at the top end of the scale, this structure will persist for some time.

公众讨论的焦点是模型的访问权限。

The public fight is about access to models.

但真正的较量，其实是对未来的掌控权。

But the real fight is about access to the future.

从现在开始，谁掌握了这种力量，谁就越来越有能力把它牢牢攥在自己手里。

And from this point forward, whoever holds this power will also become increasingly capable of keeping it for themselves.

我不像安德鲁那么确定，我们这周就能重新拿到这些模型。

Now, I'm not as sure as Andrew is that we get these models back this week.

自从这项禁令生效以来，我们一直在把每一个细微迹象都当成是漫长等待即将结束的信号。

I think ever since this ban went into effect, we have been reading every single tea leaf as confirmation that our long wait would soon be coming to a close.

当然，有些人已经能用上Mythos，而且霍华德·卢特尼克也明确说过，Anthropic和美国政府的关系取得了很大进展，这些确实比我们之前看到的证据更有说服力。

Certainly, the fact that some folks have access to Mythos, and that Howard Lutnick explicitly said that there has been a lot of progress made on the anthropic US government relationship, are better indicators than some of the evidence that we've had before.

但我现在还不会押注任何具体的时间表。

But I'm not ready yet to bet on any particular timeline.

我觉得安德鲁说得对，就算这个问题表面上解决了，那些更大的问题依然存在。

I think Andrew's broader point that even when this situation is, quote unquote, resolved, the bigger questions will remain, is dead on.

现在讨论这么紧张，大家都高度紧张，想冷静下来，提醒自己别太着急，这很自然。

I think that it's tempting, when conversation is so fraught and everyone is keyed up to eleven, to try to intellectually slow things down, to say we must be getting ahead of ourselves.

没错，这次被拒绝访问的经历，等过去了回头看，可能就觉得没多久，但我相信，大家感觉到的这种重大转变，确实是真实的。

And while I do think that, yes, this particular denial of access will feel like a short time once that time has ended, I believe that people's sense that something big has changed is correct.

我们谁也无法预知所有后果，也不知道最终会怎样发展，但我相信，等这个神话和迈托斯禁令结束之后，我们所面对的世界，将和以前大不相同。

I don't think any of us can know the full slate of implications or how it will all play out, but the world that we will have on the other side of this fable and mythos ban will, I believe, be different than the one we had before it.

好了，今天这期《AI每日简报》就到这里。

For now, that's gonna do it for today's AI Daily Brief.

一如既往，感谢收听或观看，我们下期再见，保重。

Appreciate you listening or watching as always, and until next time, peace.