Name: 人工智能與事實的關係很複雜。這個問題能解決嗎？ (AI has a tricky relationship with facts. Can that be fixed?)
Uploaded: 2025-10-15T18:21:48.000Z
Duration: 41 min 27 s

Foreign this week we're at Web Summit, the annual tech conference in Lisbon.

本週我們來到里斯本參加年度科技盛會 Web Summit。

The main topic of discussion here is, of course, AI.

這裡討論的主要話題，當然就是 AI。

轉承詞

The excitement around large language models such as Microsoft's Gemini and OpenAI's ChatGPT has, however, been tempered by concerns with bias, a lack of transparency and the accuracy of results that generative AI presents.

不過，大家對於 Microsoft 的 Gemini 和 OpenAI 的 ChatGPT 等大型語言模型的興奮感，已經因為生成式AI 帶來的偏見、缺乏透明度以及結果準確性等問題而有所減弱。

So is there a way to de risk the AI development process and build accurate, factual AI that we can trust beyond the value?

那麼，有沒有辦法降低 AI 開發流程的風險，建立我們能夠信任的、準確且事實AI 呢？

Brooke Hartley Moy is co founder and CEO of AI startup in Factory.

Brooke Hartley Moy 是 AI 新創公司 in Factory 的共同創辦人兼執行長。

She previously worked for Salesforce and Google, and

她先前曾在 Salesforce 和 Google 工作，

before starting in Factory, she was head of strategic partnerships at Humane, a company that developed an AI pin and which we'll talk more about later.

在創立 in Factory 之前，她曾是 Humane 的策略夥伴關係主管，這家公司開發了 AI pin，我們稍後會再多聊聊。

Brooke, thank you for joining us on beyond the Valley.

Brooke，謝謝你來到「Beyond the Valley」。

We're going to get into your journey, your story, and

我們要來聊聊你的歷程、你的故事，

tackle what you call the bro code of AI business.

還有你說的「AI 商業的兄弟會規則」。

But first, you wrote once upon a time, with a new frontier comes the opportunity to make new rules.

但首先，你曾經寫過：「新邊疆的到來，也伴隨著制定新規則的機會。」

We don't need to adhere to the old power norms.

我們不需要遵守舊有的權力規範。

The AI revolution gives us an opportunity to welcome a new slate of leaders.

AI 的革命給了我們一個機會，去迎接新一批的領導者。

So if we are to have a new cohort of leaders in AI, what skills or more importantly, characteristics do you believe they need to have?

所以，如果我們要有一批新的 AI 領導者，你認為他們需要具備什麼技能，或者更重要的是，什麼特質？

Yeah, you know, I've been in software longer than I care to admit.

嗯，你知道，我在軟體業的時間比我願意承認的還要長。

You know, I won't put a number on it usually, but

I've seen, you know, the conversation of move fast and break things.

我看過「快速行動，打破常規」這樣的說法。

The world of, you know, if you are disrupting, that's where the praise comes from.

所以，如果你要顛覆現狀，讚美就會隨之而來。

That's where the opportunities come from.

I think AI, we've never had a technology in anyone's lifetime that has this much potential to fully change every aspect of everyone else's lives.

我認為 AI，在我們有生之年，從來沒有一項科技能有這麼大的潛力，徹底改變每個人生活的各個面向。

And for that reason, the people that are going to be successful, harnessing its capabilities, ensuring that we have something that we are going to be proud of from an innovation perspective, not afraid of, not concerned about, need to be.

也是以，那些能夠成功駕馭其能力，確保我們在創新方面能引以為傲，而非感到害怕或擔憂的人，必須具備這些特質。

People that really understand the practical opportunities, have a pragmatic viewpoint, understand a diverse set of experiences and perspectives, not just a single sort of lane.

那些真正瞭解實際機會、有務實觀點、理解多元經驗和觀點，而不僅僅是單一思維模式的人。

And you know, it's funny that we're on beyond the Valley podcast because I think about how there is sort of this bubble type thinking specifically to the Silicon Valley set around where I can be going.

而且你知道，我們在 Beyond the Valley podcast 節目上哈拉，這還挺有趣的，因為我想到矽谷那種特定的、有點像泡泡一樣的思維，關於我能去哪裡。

And I think it's also people that ultimately want to be building AI for good.

而且我認為，這也是那些最終想為善而建構 AI 的人。

你知道，我自稱是 AI 樂觀主義者。

And for that reason, I want people that, that see that opportunity more so than they see just dollar signs or the ability to move as quickly as possible to, to see what happens and hope for the best.

所以基於這個原因，我希望大家能看到這個機會，而不是隻看到金錢數字，或是以最快的速度行動，看看會發生什麼，然後寄望於最好的結果。

Before we get into it, we're going to have Arjun do stat of the week.

在我們開始之前，我們要請 Arjun 來做本週統計。

We've already explained it to you, so let's hear it, Arjun.

我們已經跟你解釋過了，所以請說吧，Arjun。

But I'm going to do it slightly differently, you know, so I'm going to give you the start of a sentence with a stat in it at the end.

但我會稍微做點不一樣的，你知道，我會給你一個句子的開頭，句子的結尾會有一個統計數據。

我們稍後會回來討論任何事情。

All right, Arjun, I know you want to start us off with your opening questions, Brooke.

好的，Arjun，我知道你想先提出你的開場問題，Brooke。

Yeah, Well, I want to just ask a little bit, Brooke, about your backstory.

嗯，Brooke，我想問一下你的背景故事。

Yeah, you know, it's funny, I was, I was a history major in college, which actually comes up nowadays more often than not.

嗯，你知道，很有趣的是，我在大學時是歷史系學生，這在現今其實越來越常被提起。

In fact, my talk this week will have a little bit of a history theme to it because I bring sort of this sense of where are we as far as cycles and trends?

事實上，我這週的演講會有一點歷史主題，因為我帶來了關於我們在週期和趨勢方面的現況。

What does that mean for all the trends that came before and what can we learn from that?

這對我們之前的各種趨勢意味著什麼？我們能從中學到什麼？

I ended up in tech, honestly, because it was something as going through school I always was interested in, but

說實話，我會進入科技業，是因為在求學過程中我一直對它很感興趣，但是

I thought that the only opportunity is in tech were for people that were software engineers, computer scientists, which was, you know, and

我以為科技業唯一的機會是給軟體工程師、電腦科學家，你知道，而且

仍然是一個非常男性主導的領域。

It wasn't a space that I necessarily saw as welcoming or open to people that were like me or had my kind of background.

這不是一個我認為會歡迎像我這樣背景的人，或是對我有這樣背景的人開放的領域。

So I never knew really that it was an option until actually a fellow female founder, Cass Lazaro, who's still someone that I consider a mentor and I look up to greatly.

所以，我其實從來不知道這是一個選項，直到一位女性創辦人 Cass Lazaro，她至今仍是我非常敬佩和仰慕的導師。

She had a startup that was focused on software applications for Facebook, of all things, which always dates me with Gen Z.

她當時創辦了一家公司，專門為 Facebook 開發軟體應用程式，這總是讓我覺得自己跟 Gen Z 有點距離。

And I had this opportunity to take that as my first job.

我是以有機會得到了我的第一份工作。

She took a bet on me with no software engineering experience, no tech experience, and

她在我沒有軟體工程經驗、沒有科技經驗的情況下，選擇相信我，

那家公司後來被 Salesforce 收購了。

I spent a large chunk of my career there, learned the ins and outs of all things Enterprise sales and so I software sales and then was part of the founding team in Asia Pacific for Slack, which was very much a zero to one.

我在那裡度過了職業生涯的很大一部分，學到了企業銷售和軟體銷售的各種眉角，然後成為 Slack 在亞太地區的創始團隊成員，這是一個從零到一的過程。

Slack was not a mature business yet in the US and so it was even further from a mature business in Asia Pacific and figuring out how to build something from the ground up there, which is where a lot of my entrepreneurial interests I think stem from.

Slack 在美國還不是一個成熟的業務，在亞太地區更是如此，摸索著如何在當地從頭開始建立一個事業，我想這也是我許多創業興趣的來源。

Moved back to the us had the opportunity to be at Google for a number of years working on Android, which is particularly ironic now because my co founder is one of the iPhone inventors and so I like to give him a hard time.

我搬回美國，有機會在 Google 工作了好幾年，負責 Android，這現在聽起來特別諷刺，因為我的共同創辦人是 iPhone 的發明人之一，所以我喜歡拿這件事來開他玩笑。

I'm green bubble until I die and still carry my Android device with pride.

我會一直用綠色泡泡直到死，而且至今仍自豪地帶著我的 Android 手機。

And then was recruited from Android to join the humane team.

然後就被從 Android 挖角到 Humane 團隊。

Seeing that the only thing that would take me away from Google was something that was truly a generation defining technology and that was AI and that was now a few years ago.

我看到，唯一能讓我離開 Google 的，就是真正定義下一個世代的技術，那就是 AI，這大概是幾年前的事了。

And getting to have a front row seat to a company that was building foundation model Type technology before ChatGPT was a household name and getting to have sort of a front row seat to all those experiences and spending a lot of time working on what are the big challenges that need to be solved in AI before the technology is going to be able to be used in a way that people really find valuable and meaningful ultimately was what led me to what is now in factory and what I'm focused on now.

並有機會親眼見證一家在 ChatGPT 還沒家喻戶曉之前，就開始打造基礎模型技術的公司，並有機會親身參與所有這些經歷，花了很多時間研究在技術能被大眾真正認為有價值和有意義地使用之前，有哪些重大的挑戰需要被解決，這才引導我走到了現在的 In Factory，以及我現在專注的領域。

And we'll get more into that entrepreneurial journey later.

之後我們會再深入聊聊那段創業的旅程。

請說說看，請為我們介紹一下。

So the whole idea of in factory, it's even in the name, in fact, we are focused on how do you bring accuracy, determinism and facts to the AI space.

所以「in factory」的整個概念，甚至名字就已經點明瞭，我們專注於如何為 AI 領域帶來準確性、確定性和事實。

You know, I think at this point in the journey that we've been on with LMS and generative AI, I think most people have a pretty good understanding that hallucinations are a feature, they're not a bug.

我想，在這個大型語言模型和生成式 AI 的旅程中，大家現在對幻覺（hallucinations）應該都有個基本的認知，它們是「特色」而非「錯誤」。

And as much as we see the models improving, as much as we see more accuracy and more human like qualities to the models and the technology that they're able to create, they are inherently probabilistic.

儘管我們看到模型不斷進步，準確性越來越高，也越來越有人性化的特質，以及它們能創造出的技術，但它們本質上還是機率性的。

And we wouldn't want to curb that, we wouldn't want to reduce that entirely because that's partly what makes them magical, what makes them feel human, like which allows them to be creative, you know, when they are able to improve writing or being used for agentic capabilities where they can be very human, like in how they complete tasks.

我們不會想壓抑這點，也不會想完全消除它，因為這部分正是讓它們神奇、讓人感覺像人類、能激發創意的原因，像是能改進寫作，或是用於代理能力，在完成任務時能非常像人類。

All of that is part of what you want in a generative AI type application.

這一切都是你想要在生成式 AI 應用中所擁有的。

But at the same time, when it comes to enterprise and for businesses where close enough is not good enough, then there needs to be some other technology, some other innovations in the space outside of LLMs, that's going to close the gap between true deterministic accuracy and some something that just feels a little bit off, a little bit of randomness in the system, ultimately leading to likely an erosion of trust and sort of an embrace of the technology itself.

但同時，對於企業和商業來說，「差不多就好」是行不通的，這時候就需要有別於大型語言模型的其他技術、其他創新，來彌補真正的確定性準確性與感覺有點不對、系統中帶有隨機性的東西之間的差距，最終可能導致信任的侵蝕，以及對這項技術本身的抗拒。

So rtech at in factory is essentially a platform that allows anyone that is trying to build AI technology for the enterprise to apply RTech in such a way that you can get that level of accuracy, you will have citations, accuracy, attribution, deterministic answers alongside the generative AI capabilities.

所以，「in factory」的 rtech 本質上是一個平臺，讓任何想為企業建構 AI 技術的人，都能以這種方式應用 RTech，進而獲得那樣的準確性，你會得到引用、準確性、歸因、確定性的答案，同時也兼具生成式 AI 的能力。

So Brooke, just in terms of how that actually works, do you have your own proprietary LLM?

Brooke，就實際運作方式來說，你們有自己的專有大型語言模型嗎？

And then if I'm a business, for example, and

然後，如果我是一家企業，例如，

I want to build something based on your LLM, is it harnessing the data that I have within my organization to do that?

我想基於你們的大型語言模型來建構一些東西，它會利用我組織內部的數據來做到這點嗎？

所以我們沒有建立自己的模型。

What RTech is, is sort of the semantic layer, the APIs and frameworks, the processing side.

RTech 的本質是語義層、API 和框架，還有處理端。

We're sort of the middleware between the LLM and the application layer.

我們是大型語言模型和應用程式層之間的中間件。

So if you're wanting to use our technology, you're bringing your proprietary data.

所以如果你想使用我們的技術，你會帶入你專有的數據。

You don't have to feed it to a model that's going to train on it and have questions about data leakage, questions about security, just questions about trust, about whether you really want an LLM to know everything about your business.

你不必將它餵給一個會對其進行訓練的模型，並對數據洩漏、安全性，以及信任問題產生疑慮，也就是你是否真的想讓一個大型語言模型知道你公司的所有事情。

And, you know, if you're a media company, your entire archive of content and being able to then apply that in such a way using our technology, that the output is ultimately some kind of application, you know, that you can search off of, you can do calculations off of, you can look at research or articles or anything else that kind of points to the underlying data that goes into the system.

還有，你知道，如果你是一家媒體公司，你的整個內容檔案庫，並能透過我們的技術以某種方式應用它，最終的輸出會是某種應用程式，你可以用它來搜尋、進行計算、查看研究或文章，或是任何指向系統中輸入的底層數據的東西。

Just for our listeners, explain hallucinations, what that means because.

為了讓我們的聽眾瞭解，請解釋一下幻覺（hallucinations），那是什麼意思，因為。

Because we're going to get into inaccuracies and biases, but

因為我們接下來會談到不準確性和偏見，但

hallucinations may not immediately make sense to our listeners, right?

幻覺對我們的聽眾來說可能不是那麼直覺，對吧？

字幕列表影片播放

人工智能與事實的關係很複雜。這個問題能解決嗎？ (AI has a tricky relationship with facts. Can that be fixed?)

perspective

ultimately

episode

sort