Yes, deep, thick for pro is here this time. There is one. I know it gets confusing.
是的，DeepSeek Pro 这次来了。确实有一个。我知道这很容易让人混淆。
The previous version was called preview, and this is called zero eight party.
之前的版本叫 preview，而这个版本叫 0.8 party。
The numbers it performs better than is much smaller flash version.
它的性能数据比小得多的 Flash 版本更好。
When building global scope. Flash did not completely understand the 3D structure of the object, lots of missing parts, lots of blackness.
在构建全局场景时，Flash 没有完全理解物体的 3D 结构，有很多缺失部分，很多黑色区域。
But the the, the property look much better understanding of structure.
但是那个、那个，Pro 版本对结构的理解看起来好多了。
And I was also surprised by by, holy mother of peers. Look at that.
而且我也被……天哪，太惊人了。看看那个。
That is, is Ching closer and closer to Faber quality?
那正在一点一点接近 Falcon 的质量？
Yeah, another challenger appeared, and it gets better.
是的，又一个挑战者出现了，而且它变得更好。
They give all this for us for free, which is absolutely incredible.
他们免费把这一切给我们，这绝对令人难以置信。
Okay, but what does it mean for us? The weights are available for free for all of us.
好的，但这对我们意味着什么？权重对我们所有人免费开放。
But you know, you have the hardway to to host at home. ID love to, but I don’t have that kind of harder.
但你知道，你得有硬件才能在家里托管。我很想，但我没有那种硬件。
Other options include lambda or using it hosted by deep ep, seek themselves.
其他选项包括 Lambda，或者使用 DeepSeek 自己托管的服务。
But they just raise their prices dramatically, about two and half to five eggs, the previous prices.
但他们刚刚大幅提高了价格，大约是之前价格的两倍半到五倍。
Now I bet you can already imagine the click bait headline saying it’s over.
现在我敢打赌你已经能想象出那些标题党说“完了”。
I think what they should also say is that deep seek has M. I, T licensed open weights.
我认为他们还应该说的是，DeepSeek 拥有 MIT 许可的开放权重。
What does that mean? Well, anyone can run the existing model at their own price, and they do a bunch of alls available, and they all compete on the price.
这意味着什么？嗯，任何人都可以按自己的成本运行现有模型，而且有一堆可用的 API，它们都在价格上竞争。
That is amazing for us. And it is very likely to push the frontier labs during the collectors something even better and quickly.
这对我们来说太棒了。而且这很可能推动前沿实验室为集体快速推出更好的东西。
And all this improvement comes from the same architecture. But how is that even possible?
所有这些改进都来自相同的架构。但这怎么可能？
The model structure is the same, yet it is maaster vely better than the pv was less than four months ago.
模型结构相同，却比不到四个月前的 preview 版本好得多。
So how their fellow colors, this is the special, special at once again, a lot of the magic happens after free training.
那么，各位同仁，这又是特别之处，很多魔法发生在预训练之后。
During post training, deep sea creates several special models for mathematics, coding and agency.
在后训练期间，DeepSeek 为数学、编程和智能体创建了几个专门的模型。
Now we have to suck here for a moment. People confuse these with the experts in mixture of experts.
现在我们需要在这里停一下。人们把这些和混合专家模型中的专家混淆了。
That’s not quite the same. Those are little pieces within one neural network. These are not. These are separately trained model checkpoints.
那不太一样。那些是一个神经网络内部的小部分。这些不是。这些是单独训练的模型检查点。
Okay. So then comes installation. Yeah, they take more than ten of the specialist teachers and train one final model, absorb b the abilities.
好的。然后是蒸馏。是的，他们用十多个专家教师来训练一个最终模型，吸收这些能力。
So the student mother says, this is what I would do. Then the teacher says, well, this is what I would have done. Then the student adapts its brain to be more like each teacher do it with ten teachers.
所以学生模型说，我会这么做。然后老师说，嗯，我本来会这么做。然后学生调整自己的大脑，变得更像每个老师。用十个老师这样做。
And you see that the student indeed improves like crazy.
你会看到学生确实疯狂地进步。
They also added this part to it. Instead of just predicting one token at a time, it drafts several tokens ahead. It does it much better than previous techniques.
他们还加入了这一部分。它不再一次只预测一个 token，而是提前草拟多个 token。它比之前的技术做得好得多。
And hold on to your papers, fellow colors, because deep sea reports up to 78% fast generation for, for, for pro.
各位同仁，请坐稳了，因为 DeepSeek 报告 Pro 的生成速度最高提升了 78%。
A real, measurable speed up in the real use that you get her right now and benefit from it.
这是你现在就能获得并受益的真实、可衡量的加速。
And here is something absolutely insane. This once a research paper, let’s see, six weeks ago and now everyone is using it.
还有一件绝对疯狂的事。这曾经是一篇研究论文，大概六周前，现在每个人都在使用它。
Let me say again, a research paper only six weeks ago, one of the best papers of the year, and it is coming alive right in our hands for free.
我再说一遍，仅仅六周前的一篇研究论文，年度最佳论文之一，现在正免费在我们手中变成现实。
Incredible full breakdown video in the description.
描述中有完整的详细解析视频。
And don’t forget, we own and can run it if one Dangeau to a different model, if we type a wrong key words, games.
别忘了，我们拥有并且可以运行它，如果我们想换一个不同的模型，如果我们输入错误的关键词，收益。
That is incredible, even if I can run it at home, which I would love to do. But there are options.
这太不可思议了，即使我可以在家里运行它，我也很乐意。但还有其他选择。
What the time to be alive even is open science and open research. It is best, and it’s important that we talk about it.
活在这个时代真是太好了，开放科学和开放研究是最好的，我们谈论它很重要。
Why? Because the future belongs to those who who understand it.
为什么？因为未来属于那些理解它的人。
Use deep, thick and use this Spark advantage of them.
使用 DeepSeek，并利用它们的这一优势。
Oh, and I plan to talk about deep sea residence. No agent tones as well. Another design really powerful, if you’re interested, consider subscribing and hitting the bag.
哦，我还计划谈谈 DeepSeek 最近的新智能体工具。还有新的智能体工具。另一个非常强大的设计，如果你感兴趣，考虑订阅并点击铃铛。
I use lambda to reproduce AI research papers, often in minutes.
我使用 Lambda 来复现 AI 研究论文，通常几分钟就能完成。
It’s also great to train your own models or ftd, an existing one, and influence or text to image or video.
它也非常适合训练你自己的模型，或对现有模型进行微调，以及进行文本到图像或视频的推理。
Easy pc running a deep sick chat bot, or as NA super fast, super reliable lambda gives you powerful and video gp s to run your own experiments.
轻松运行 DeepSeek 聊天机器人，或者作为一个超快、超可靠的 AI，Lambda 为你提供强大的视频 GPU 来运行你自己的实验。
I test ideas from the papers I cover, and moments later, results. Love it. Seriously, try it out. Now lumbered that AI slash papers.
我测试我介绍的论文中的想法，片刻之后就能得到结果。非常喜欢。说真的，试试看。现在就去 Lambda AI slash papers。