今天实际上是在早上 6 点才睡着的。之前一直辗转反侧睡不着，感觉压力很大。MATH 263、COMP 579 都没有学完，周四中午要考试了，还有很多没学，感觉很有压力。就在床上辗转反侧睡不着。接着在 6 点的时候想起来学，但是算了一卦说还不如继续睡，就继续躺下了。过一会果真就睡着了。

睡到定的闹钟 12 点起来，起来收拾一下去吃饭，吃了 Opiano。吃完之后下午开始 meeting，先是 MATH 463 的 meeting，一起讨论了一下分工。因为实际上这个 proposal 只是为了让我们确定 meet 了一次就行，所以我们就是用各自的笔记证明我们大家坐到一块、一起写了个字。简单地分了一下工，每人负责一个 algorithm。接着一起学了一会 463。

接着下一个 meeting 是 COMP 550 的 meeting。实际上也是分工和 proposal。我对这个 meeting 直接看到说我们可以对 LLM 进行研究，所以我就提了一堆跟我那个 Skip Lecture 项目相关的研究。比如说研究模型对图像的分析能力，研究模型对于长 slides 的文本进行分析和长间隔的关联性逻辑分析能力等问题。队友提出了一个对于错别字的容忍度的问题，我说这个问题几乎测不出来什么效果，不如改成 Prompt Injection，他也同意了。我们就写了大概四个问题上去。

实际上我当时特别想提：让 AI 作为其训练数据来源群体的代言人，研究 AI 的政治倾向，以此分析其主要训练文本的数据来源人群的政治倾向或各样的性格图谱。但是大家一致认为这个可能不是 NLP 的问题，更像是一个社会学问题，所以就没有写。

接着我们一起去六楼继续学，我在这段时间基本上学完了 MATH 463 的 Chapter 5，然后在大概六点半左右去吃饭，最后吃了披萨。我当时看到一个黑黑的披萨，觉得很感兴趣点了一个，结果他告诉我那是奥利奥巧克力披萨。当时想着就尝尝鲜，结果拿上来那个小饼非常小，估计有个四寸到六寸的大小，闻起来像蛋糕一样，上面抹满了巧克力。咬了一口贼甜，但是对我来说不算很腻，因为我口味比较重。但是吃了两片之后感觉不对，这吃的根本就不爽，没有肉，全是糖，又撑肚子。吃不爽，所以我扫旁边的二维码，买了炸鸡。披萨店的员工也没有说什么，就看着我让我吃了。

吃完之后，我们又回去继续学，我把第六章给学完了，来不及看所有的定理和推导了，只把定理抄了一遍，理解了一下定理。真的感觉定理好多，并不是很能记得住。而且在这个时间，我又打了两局游戏，又打输了，所以心情很烦躁，于是我问了 Gemini 和 GPT 同样的问题：打游戏打赢了觉得对手也没有成就感，打输了又觉得自己很菜，感觉非常自怨自艾；然后学习，遇到简单的又不想学，让 AI 帮我做，遇到难的又不想学，怎么办？然后他们都告诉我，这只是一种对心流状态的非常严格的要求，导致我不管什么情况都感觉不好。

然后我又问了 GPT 一句，我说我觉得现在各种事情都没有什么意义，学习上又觉得很难、不想学，什么都不想做，这种情况怎么办。然后可能触发了什么机制，直接告诉我”你这样不行，要去看心理医生”，给了我一堆 McGill 和蒙特利尔相关的电话，让我去打，又给我说什么自杀帮助热线。给我整害怕了。

我现在感觉要记录日记，就是因为万一我有心理问题，那如果后面真的有问题了，也可以通过日记来发现我到底从哪里开始心态有了明显的变化。如果以后真的有心理问题了，这也是一个记录。同样，我也是因为在做 AI 相关的东西，用 Remote Lab 结合现在非常好用的豆包输入法，能非常好地输入中英文结合的文字，还理解得非常好，还能去除噪音词，真的太棒了。感觉比我现在电脑上那个 WhisperFlow 要好得多，而且它不需要停顿识别，中英文结合的语言特别强，我真的好喜欢这个输入法。希望它赶快有电脑端，这样我就能把电脑上这个收费的 WhisperFlow 给换掉了。

然后我回到家之后，开始收拾东西，本来说想要玩杀戮尖塔二的，但是一直到现在凌晨一点四十八都没玩上，因为我看了两个数学简史的视频。这个数学简史真的好好看，又诙谐幽默，又真的讲了很多历史相关的东西。真的感觉数学的发展是一个凝聚人之力的感觉。同时我又想到，为什么数学当时有一个人类群星闪耀时？真的是因为数学每一次危机都是数学发展的一个窗口期吗？那如果以此类推的话，难道每一次的危机，实际上都是人类科技，或者说人类进步的一个大爆发阶段吗？比如说一战、二战实际上是科技进步大爆发的阶段，冷战又是航天科技进步大爆发的时间段。这可能是一个合理的理由，以后可以研究一下。

看到数学简史的终章，真的到中后期煽情的时候就哽咽了，有点想哭，眼泪都在眼眶里打转。但是正好在被窝里，所以也感觉不到是否真的哭了。但是我感觉已经很久没有这种哽咽、眼泪打转的情况了，可能也是一个发泄吧。

然后接下来我一边看数学简史的上篇，一边让 Claude 帮我写一个日记页面。接下来我就开始用豆包输入法输入这篇日记，准备上传第一篇日记了。让我试试用 Remote Lab 把这篇日记传上去吧。

I actually didn’t fall asleep until 6 in the morning. I’d been tossing and turning, feeling a lot of pressure. I haven’t finished studying for MATH 263 or COMP 579, the exam is Thursday at noon, and there’s still so much left. I was lying in bed unable to sleep. Around 6 I thought about getting up to study, but did a quick gut-check and decided I’d be better off staying in bed. A little while later, I actually drifted off.

I woke up at noon when my alarm went off, got ready, and went to eat at Opiano. After lunch, the afternoon was all meetings. First was the MATH 463 meeting, where we discussed dividing up the work. Since the proposal really just requires us to confirm we met once, we basically used our individual notes as proof we’d all sat down together and written something. We split it simply — one algorithm per person — then studied 463 together for a bit.

Next was the COMP 550 meeting, also about splitting work and the proposal. I saw right away that we could research LLMs, so I pitched a bunch of ideas related to my Skip Lecture project — things like analyzing a model’s image comprehension ability, or its capacity to handle long-form slide text and make logical connections across large gaps. A teammate suggested studying a model’s tolerance for typos, but I said that would barely show any interesting results and proposed Prompt Injection instead. He agreed. We ended up writing down about four research questions.

I’d actually really wanted to pitch another idea: using AI as a spokesperson for the communities that trained it, studying AI’s political leanings to analyze the political tendencies or personality profiles of its primary training data sources. But everyone felt it was more of a sociology question than an NLP one, so we dropped it.

After that we moved up to the sixth floor to keep studying. I managed to finish MATH 463 Chapter 5 during this stretch, then headed out for dinner around 6:30 and ended up getting pizza. I spotted a very dark-looking pizza, got curious, and ordered it — only to be told it was an Oreo chocolate pizza. I figured, why not try something new. When it arrived, the little thing was tiny, maybe four to six inches across, smelled like a cake, and was slathered in chocolate. The first bite was intensely sweet, but not unbearably rich for me since I have a strong palate. After two slices, though, it felt all wrong — no meat, pure sugar, just sitting in my stomach. Completely unsatisfying. So I scanned the QR code nearby and ordered fried chicken. The pizza place staff didn’t say a word, just watched me eat.

We went back to study afterward. I finished Chapter 6, though I ran out of time to go through all the proofs — I just copied down the theorems and tried to understand them. There are so many theorems, and I honestly can’t retain all of them. On top of that, I lost two more games during this time, which left me in a foul mood. So I asked both Gemini and GPT the same question: when I win I feel no satisfaction because the opponent wasn’t impressive, and when I lose I feel bad about myself — constant self-pity either way. And when studying, if it’s easy I don’t want to bother and just have AI do it, and if it’s hard I don’t want to deal with it either. What do I do? They both told me it’s essentially an extremely high demand for flow state, which means nothing ever feels quite right.

Then I asked GPT one more thing: I said everything feels meaningless right now, studying feels hard and I don’t want to do it, I don’t want to do anything — what should I do? It must have triggered some safety mechanism, because it immediately told me I needed to see a mental health professional, gave me a whole list of McGill and Montreal phone numbers, and even mentioned a suicide helpline. That genuinely scared me.

I think the reason I want to start keeping a diary now is precisely because of that — what if I do have some mental health issues developing? If something does go wrong later, I could look back at these entries and figure out exactly when my mindset started to noticeably shift. It would be a record. I’m also motivated because I’m working on AI-related stuff anyway, and using Remote Lab together with the Doubao input method — which handles mixed Chinese and English really well, understands everything accurately, and even filters out filler words — it’s genuinely great. Way better than WhisperFlow on my computer, and it doesn’t need pause recognition. The mixed-language support is so much stronger. I love this input method. I’m hoping they release a desktop version soon so I can ditch my paid WhisperFlow subscription.

When I got home, I started getting things in order. I had been planning to play Slay the Spire 2, but here it is, 1:48 in the morning, and I still haven’t touched it — because I watched two episodes of a math history series. It’s such a great series, witty and funny while genuinely covering a lot of historical ground. The development of mathematics really does feel like a story of collective human effort. It also made me think: was there really a “Moments of Greatness” era in mathematics? Is every mathematical crisis actually a window of opportunity for mathematical progress? And by extension, is every great crisis actually a period of explosive human advancement? World War I and II drove enormous leaps in technology; the Cold War was the golden age of space exploration. That might actually be a reasonable theory — something worth looking into sometime.

The final episode hit hard. Midway through the emotional ending, my throat tightened. I felt like crying, and my eyes were full of tears. I was in bed so I couldn’t really tell if I was actually crying. But it’s been a long time since I’ve felt that — that tightening in the chest, eyes welling up. Maybe it was a release of some kind.

After that, I watched the first episode of the series while having Claude help me build a diary page. Then I started dictating this entry with the Doubao input method, getting ready to upload my very first diary post. Let’s see if I can get this up through Remote Lab.