上禮拜我們老闆說我讓他在投資人面前讓他丟臉了。他說，他讓 AI 做了一門量子（Quantum）相關的課，結果 AI 製造什麼貓的圖片，一點關聯也沒有！我仔細一瞧，什麼？他罵誰笨，他才笨吧，AI 製造的可是薛丁格的貓（Schrödinger’s cat）！
神奇的是，你看看那門課裡面（附圖），沒有半個字提到貓，AI 居然知道可以這麼畫。AI 的聰明與創意，讓我打從心底欣賞與敬佩。
就業市場：AI 工程師的崛起
現在在舊金山不論產業，所有公司和新創都想融合 AI 的應用，大公司像是 Micorsoft 所有工具都有 Copilot（在那邊工作的朋友說的），中型公司像是 Notion AI，小型新創們更不用說，2023 夏季 Y Combinator 的公司幾乎全部都是 AI 公司。
因此 Latent Space 的文章裡寫到，一種新興的行業，AI 工程師即將誕生。有別於機器學習工程師和研究員，AI 工程師其實不太需要知道 LLM 底層知識，可以直接套用 API 在需要的產品之中。所需要的技能可能更多包含 LLM 後端需要用到的各種工具。
畢竟，對公司來說，要有超多錢才有辦法做出比 Openai 更好的模型，不如仰賴這些打造基礎模型的公司，把重心著重於打造用戶需要的好產品。而我現在做的就是 AI 工程師，練習如何讓 AI 乖乖聽話，把相關需要的工作做好，最後包裝成產品。
技術：AI 進步的速度，好快。
特別是在矽谷， AI 相關技術的進步真的太快了，好可怕。因為變動太大，我常常感覺自己重心不穩，每天工作都不太清楚自己在做什麼。
跟大家說一個故事讓你也明白科技進步的速度。今年五月份來舊金山的時候，LLM 的模型微調（fine-tuning），其實還是需要很多工程技術與 LLM 相關的知識。你需要懂，怎麼 Freeze layers，怎麼掛 GPU ，怎麼部署模型等等才能做。但我八月份的時候去一個黑客松，就發現已經有像公司 together.ai，給你一個平台，讓你把資料丟上去微調，如果大致了解參數們 ，二十分鐘就好了。結果兩週後，Openai 發布 Fine-tune GPT3.5 的消息。我再也什麼都不用懂，因為只要把資料集丟上去，Openai 就會幫我微調，送給我一個模型，把一整套流程都做完了。
雖然了解 transformer 的 LLM 基礎知識還會在直覺上有些幫助。但是 AI 產業真的進步太快了，嚴格來說所以我至今的工作，都還完全不需要基礎知識。
心態：我對 AI 的害怕
在這樣的環境裡，我發現有部分的自己是害怕的。在矽谷這個 AI AI AI 的環境裡，我可以感受到科技進步的速度。因為是我感覺自己做的事情越來越笨，完全不具有競爭力。
你看從五月到八月，短短的三個月，他的技術門檻就從機器學習工程師，變成所有工程師都可以做，這進步也太快了吧！如果我只需要丟資料集給 Openai 就可以把 LLM 微調程我想要的樣子，需要我機器學習和資料分析的背景做什麼呢？隨便路上拉一個會寫程式的路人不就得了？
那如果只要一個會寫程式的人都能做，之後乾脆全部都讓 AI 來做不就得了嗎？（舊金山也有很多很多 Code gen 公司，致力於用 AI 取代工程師）
機會：AI 可以顛覆舊有的體制嗎？
雖然 AI 的力量和科技演進的飛速讓我有很多的不安，但是，就像是大家從抗拒手機，到人手一機，天天盯著他看，科技的進步是沒有辦法倒退，只能適應的。反正我的哲學是，越害怕的東西，就要更勇敢的去面對它，瞭解他。因此，與其害怕 AI 搶走工作，毀面世界，不如學更多，了解它的可能性與限制，把他的可能性化為可用的力量。
因爲科技是有力量翻轉現有體制的（典範轉移：paradigm shift），歷史永遠是科技在前，體制在後，體制因應科技的改變而演進。像是傳統教育或許是個根深蒂固的體制，但如果 AI 已經在翻轉人類傳統透過學校的學習方式，說不定可以為我們的學習開創更多的可能性。像是 Minerva 就透過線上課程，創立自己的全新教育模型，吸引學生到各個城市學習才得以在舊制度下生存。
這也是為什麼我現在在教育/AI 的新創探索，雖然還不清楚怎麼做和想要帶來哪種層次的轉變，我想，新的力量就能帶來新的改變，而我想明白怎麼運用這股新力量。
AI 與我
在舊金山每天看著 Cruise 可愛的無人車在路上跑來跑去，Openai 隔著我們公司一條街，聽所有人 AI AI AI ，AI Alignment 研究（如何讓 AI 不要毀滅世界）、下一個 AI 獨角獸，我總是可以強烈的感受到「進步」是什麼樣的感受。
雖然對 AI 依舊有有多的不安，但總體而言，天天和 AI 相處，我對 AI 充滿著欣賞與敬畏。當然拉，即便天天碰，我還是覺得不是很明白 AI ，但是依舊期待更多的了解他，學習駕馭它，和它好好合作，為世界帶來想看見的改變。
AI & Me
Last week, our boss said I embarrassed him in front of investors. I asked why, and he said he had AI create a course related to quantum, and the AI ended up generating pictures of cats! I took a closer look and thought, “What?” Who’s the one being silly here? AI actually created Schrödinger’s cat!
What’s even more amazing is that if you look inside that course (see the image), not a single word mentions cats, yet AI somehow knew to draw it like that. AI’s intelligence and creativity truly make me appreciate and admire it from the bottom of my heart.
The Job Market: The Rise of AI Engineers
In San Francisco, across all industries, companies and startups are now looking to integrate AI applications. Large companies like Microsoft have tools like Copilot integrated into everything they do (according to friends working there). Medium-sized companies like Notion AI are also in the game, not to mention the countless small startups. In the summer of 2023, almost all of Y Combinator’s companies are AI-focused.
As a result, as mentioned in the Latent Space article, there’s a new emerging field, and that’s the birth of AI engineers. Unlike ML engineers and researchers, AI engineers might not need to have a deep understanding of the fundamentals. They can directly apply APIs to the products they’re working on. The required skills might lean more towards the various tools needed for the backend of LLM (Large Language Models).
After all, for companies, it takes a massive amount of resources to build models better than OpenAI. It’s often more practical to rely on companies that create foundational models and focus on building excellent products that users need. What I’m doing right now is working as an AI engineer, practicing how to make AI obedient, handling the necessary tasks, and ultimately packaging them into products.
Technology: The pace of AI advancement is incredibly fast.
Especially in Silicon Valley, the progress in AI-related technologies is truly rapid and somewhat frightening. Due to these significant changes, I often feel like my focus is unstable, and every day at work, I’m not entirely clear about what I’m doing.
Let me share a story with everyone to help you understand the speed of technological progress. When I came to San Francisco in May of this year, fine-tuning LLM models still required a lot of engineering and knowledge related to LLM. You needed to know how to freeze layers, set up GPUs, deploy models, and more. However, when I attended a hackathon in August, I discovered companies like together.ai that provide a platform where you can upload data for fine-tuning, and understand some parameters, and it’s done in about twenty minutes. Two weeks later, OpenAI announced the release of Fine-tune GPT-3.5. Now, I don’t need to know anything because all I have to do is upload the dataset, and OpenAI takes care of fine-tuning, giving me a model, and handling the entire process.
Although having a foundational understanding of Transformer can be somewhat helpful intuitively, AI is progressing so rapidly that, strictly speaking, my work so far doesn’t require foundational knowledge.
Mindset: My Fear of AI
In this kind of environment, I’ve discovered that there’s a part of me that is fearful. In Silicon Valley, this AI-centric environment, I can sense the speed of technological progress, and I feel that what I’m doing is becoming less and less smart, and I’m losing my competitiveness.
Look at the three short months from May to August; the technical threshold has shifted from being exclusive to machine learning engineers to something that all engineers can do. This progress is just too rapid! If all I need to do is hand over a dataset to OpenAI to fine-tune an LLM the way I want it, what do I need my background in machine learning and data analysis for? Can’t I just find any passerby on the street who can code?
So, if anyone who can code can do it, why not just let AI do everything? (There are also many code generation companies in San Francisco dedicated to replacing engineers with AI).
Opportunity: Can AI Disrupt Established Systems?
Although the power of AI and the rapid evolution of technology has brought me a lot of anxiety, just as we went from resisting mobile phones to having one in everyone’s hands, staring at it all day, technological progress cannot be reversed; we must adapt. In any case, my philosophy is that the more fearful something is, the braver one should be in facing it and understanding it. So, rather than fearing that AI will take away jobs and disrupt the world, why not learn more, understand its possibilities and limitations, and harness its potential as a usable force?
Because technology has the power to overturn existing paradigms, history always sees technology leading, and systems adapting to the changes brought by technology. For example, traditional education may be a deeply ingrained system, but if AI is already transforming the traditional way humans learn through schools, it might create more possibilities for our learning. Institutions like Minerva have established their own innovative education models through online courses, attracting students to learn in various cities to survive within the old system.
This is also why I’m currently exploring startups in education and AI. While I’m not yet sure how to do it and what level of transformation I want to bring, I believe that new forces can bring about new changes, and I want to understand how to harness this new power.
AI and Me
In San Francisco, I watch Cruise’s cute autonomous cars running around the streets every day, and OpenAI is just a street away from our office. With everyone talking about AI, AI, AI, discussions on AI alignment research, the next AI unicorn, and I can always strongly sense what “progress” feels like.
Although I still have many concerns about AI, overall, as I spend every day with AI, I’m filled with admiration and awe for it. Of course, even though I interact with it every day, I still feel like I don’t fully understand AI. But I still look forward to learning more about it, mastering it, working with it collaboratively, and bringing the changes to the world that I want to see in the future.