53
0Episode: Jeff Dean's Honest Talk: Gemini, Discovery Loop
Duration: approximately 12 minutes
Level: B1 (Intermediate)
---
[Mike]: Welcome back to "Learn English with Podcasts"! Sarah, imagine your old boss speaks in public for the first time after quitting - and says exactly what they really think. Awkward, right?
zh:欢迎回到"Learn English with Podcasts"！Sarah，想象一下，你的老上司离职后第一次公开讲话，而且把心里话全说了出来。尴尬吧？
[Sarah]: Very awkward! But also kind of refreshing. Did that actually happen?
zh:非常尴尬！但也挺让人耳目一新的。真有这种事吗？
[Mike]: It just did. Jeff Dean - the legend who helped build Google - gave his first public interview after leaving. And he did not hold back.
zh:刚刚就发生了。Jeff Dean——帮忙打造谷歌的那位传奇人物——在离职后做了第一次公开访谈，而且一点都没藏着掖着。
[Sarah]: Wait, the Jeff Dean? The one behind Gemini, Google's biggest AI model?
zh:等等，是那个 Jeff Dean 吗？就是谷歌最大 AI 模型 Gemini 背后的那个人？
[Mike]: The same. He co-started Gemini and was its first technical lead. In this talk, he admitted something surprising about it.
zh:就是他。他共同发起 Gemini，也是它的首位技术负责人。在这次访谈里，他承认了一件挺让人意外的事。
[Sarah]: Oh, now I'm curious. What did he admit?
zh:哦，我现在好奇了。他承认了什么？
[Mike]: He said Gemini was not great at coding in the early days. They cared a bit too late about making it really good at writing code.
zh:他说 Gemini 早期在写代码方面表现并不出色。他们把"让它在编程上真正惊艳"这件事，重视得稍微晚了一点。
[Sarah]: Really? But Gemini is huge now. So what did they get right?
zh:真的吗？可 Gemini 现在很厉害啊。那他们做对了什么？
[Mike]: One big thing: they built it multimodal from day one. Text, images, video, audio - all in one model. He said that was the right call.
zh:一件大事：他们从第一天起就把多模态做进去了。文字、图像、视频、音频——全在一个模型里。他说这个决定非常正确。
[Sarah]: Makes sense. If you want one brain for everything, it should understand everything. Did he talk about older projects too?
zh:有道理。如果你想有一个能应对一切的大脑，它就该理解一切。那他也聊了更老的项目吗？
[Mike]: Yes. He looked back at TensorFlow, the tool that helped millions learn machine learning. He admitted two clear mistakes.
zh:聊了。他回顾了 TensorFlow——那个帮无数人入门机器学习的工具。他承认了两个明显的失误。
[Sarah]: Two mistakes from the genius? I'm listening.
zh:天才也会犯两个错？我洗耳恭听。
[Mike]: First, they didn't have eager execution at the start. Later, PyTorch and JAX made it popular, and TensorFlow added it too late. Second, they opened a contrib folder where outsiders added ten different ways to do the same thing.
zh:第一，他们一开始没有即时执行模式。后来 PyTorch 和 JAX 让这种模式流行起来，TensorFlow 加得太晚了。第二，他们开了一个 contrib 文件夹，外部开发者往里塞了十种做同一件事的不同方法。
[Sarah]: Oh no, ten ways to do one thing? That must have confused everyone.
zh:哦不，一件事十种做法？那肯定把大家都搞晕了。
[Mike]: Exactly. He said they should have kept the core simple. But he also praised TensorFlow for helping so many people start in AI. Fair and honest.
zh:没错。他说当初应该让核心保持简洁。但他也称赞 TensorFlow 帮了那么多人踏入 AI 领域。很公道，也很诚实。
[Sarah]: Okay, but here's my big question - he spent 27 years at Google. Why leave?
zh:好，但我最大的问题是——他在谷歌待了 27 年。为什么要走？
[Mike]: He loved his time there. But he said a tiny, focused team has a special power: everyone works on one mission, with almost no distractions.
zh:他很喜欢在谷歌的岁月。但他说，一支极小而专注的团队有一种特别的力量：所有人都围绕同一个使命工作，几乎没有干扰。
[Sarah]: Like a band of friends building something in a garage?
zh:就像一群朋友在车库里搞发明？
[Mike]: Perfect picture. And today, cloud computing lets a small team rent huge machines. You don't need to build everything yourself. A team of about 10 people can move fast.
zh:就是这个画面。而且今天，云计算让小团队也能租到巨型机器。你不需要自己从头搭建一切。大约 10 个人的团队也能跑得飞快。
[Sarah]: Still, leaving Google after 27 years must feel scary.
zh:不过，在谷歌待了 27 年再离开，肯定还是会有点害怕吧。
[Mike]: He said it's a little nervous, but very exciting. The big dream is what he calls recursive self-improvement.
zh:他说确实稍微有点紧张，但也非常令人兴奋。那个大梦想，就是他所谓的"递归自我改进"。
[Sarah]: Recursive self-improvement? That sounds like a tongue twister. What does it mean?
zh:递归自我改进？听起来像绕口令。这是什么意思？
[Mike]: Simply: use AI to make AI better. His co-founder Quoc Le once built a system that designs model architectures by itself. One result, the Evolved Transformer, was about 30% more efficient.
zh:很简单：用 AI 来让 AI 变得更好。他的联合创始人 Quoc Le 曾经做过一个系统，能自己设计模型结构。其中一个成果，Evolved Transformer，效率大约提升了 30%。
[Sarah]: Thirty percent better, just by letting AI redesign itself? That's wild.
zh:提升了 30%，只是因为让 AI 重新设计自己？太疯狂了。
[Mike]: And that's only the start. His new company, Discovery Loop, wants to automate the whole scientific loop - build, test, learn, repeat.
zh:而这只是开始。他的新公司 Discovery Loop，想把整个科学循环自动化——构建、测试、学习、再来一遍。
[Sarah]: Discovery Loop... so they want AI to do science? What's the end goal?
zh:Discovery Loop……所以他们是想让 AI 来做科学？最终目标是什么？
[Mike]: He described a model with the power of 20 PhDs. No single human has 20 doctorates. But a model could understand many fields at once, and send agents to solve sub-problems together.
zh:他描述了一个拥有 20 个博士能力的模型。没有任何一个人能拿 20 个博士学位。但一个模型可以同时理解很多领域，并派出智能体一起解决子问题。
[Sarah]: Twenty PhDs in one machine. My brain can barely handle one topic at a time!
zh:一台机器里装 20 个博士。我的大脑一次都很难应付一个话题！
[Mike]: Right? And here's the crazy part: a round of real experiments that takes a day, or even a week, could shrink to a minute, or just an hour.
zh:对吧？还有更疯狂的：一轮真实的实验，原本要花一天、甚至一周，未来可能压缩到一分钟，或者仅仅一小时。
[Sarah]: A week becomes an hour? I need that for my laundry.
zh:一周变成一小时？我的衣服也想这么处理。
[Mike]: Ha! But beyond jokes, he shared a real study tip. Instead of reading one paper very carefully, skim 10 papers - even 100 abstracts.
zh:哈！不过说正经的，他分享了一个真正的学习建议。与其非常仔细地读一篇论文，不如快速浏览 10 篇——甚至 100 篇摘要。
[Sarah]: Wait, that feels backwards. Shouldn't we go deep, not wide?
zh:等等，这感觉反了。我们不该求深，而不是求广吗？
[Mike]: His point: you want to connect ideas nobody joined before. If you know many things are becoming possible, you see a hard problem in a new light. Then you find five parts are already solvable.
zh:他的意思是：你要去连接别人从没连起来的想法。如果你知道很多事正变得可行，你就能用新眼光看难题。然后你会发现，其中五个部分其实已经有解法了。
[Sarah]: Oh, so reading wide gives you puzzle pieces. Clever.
zh:哦，所以广泛阅读是给你拼图的碎片。聪明。
[Mike]: He also uses quick order-of-magnitude estimates. Like: will moving this data take 10 seconds, or 100 years? Those are completely different worlds.
zh:他还习惯做快速的数量级估算。比如：搬运这批数据要 10 秒，还是 100 年？那可是完全不同的两个世界。
[Sarah]: So he does math in his head like a human calculator. I like that.
zh:所以他在脑子里算得像人形计算器。我喜欢这点。
[Mike]: Now the serious turn. He warned that AI in cybersecurity can already match top human hackers - maybe even go further.
zh:现在说点严肃的。他警告，网络安全领域的 AI 已经能媲美顶级人类黑客——甚至可能更进一步。
[Sarah]: That sounds scary. Can AI really find holes in our systems?
zh:这听起来挺吓人。AI 真能在我们的系统里找出漏洞吗？
[Mike]: Yes, on both sides. The same tool can attack, but it can also find and fix real-world holes before the bad guys do. It's a double-edged sword.
zh:会，而且两边都行。同一个工具能攻击，也能在坏人动手前，找出并修补现实世界里的漏洞。这是一把双刃剑。
[Sarah]: So the tool that scares us is the same tool that protects us. Maybe the real power isn't the AI - it's the questions we choose to ask it.
zh:所以吓我们的工具，也正是保护我们的工具。也许真正的力量不在 AI 身上——而在我们选择向它提出什么问题。
[Mike]: Beautifully said, Sarah. And that's a good place to pause. If you enjoyed Jeff Dean's honest talk, tell us what you think - and we'll see you next time on "Learn English with Podcasts"!
zh:说得太好了，Sarah。我们就停在这里吧。如果你喜欢 Jeff Dean 这次坦诚的分享，告诉我们你的想法——下期"Learn English with Podcasts"再见！
[Sarah]: Thanks, Mike. Until next time!
zh:谢谢你，Mike。下期见！
