We all love Reviewers. Reviewer 1, Reviewer 2, Reviewer 3 — whatever.

But what if the lovely Reviewer 1 gave you a 0 with confidence 5?

Neck

That’s exactly what happened to ICLR 2026 submission no. 21536 (See https://openreview.net/forum?id=tvDlQj0GZB).

Impressive! At least he didn’t use ChatGPT for reviewing that paper?

Or no. Maybe bro forgot to upload the PDF to ChatGPT when vibe-reviewing.

Body

The title of the paper is “Learning is Forgetting; LLM Training As Lossy Compression”.

Of course it was quite controversial. Just like how I write my blog posts.

But is it worth a 0? Not at all.

I read the work thoroughly. There’s nothing wrong in the paper that’s worth a strong reject. The literature was clear, the essay was neat, the math is justified (probably — don’t ask me).

But what is this reviewer doing????

Heart

And now let’s look at the peak moment of Reviewer 1. Even better than the 0.

Strengths: None noted. See explanation in the weaknesses. NONE.

Legs

Now let’s look at what the other reviewers gave:

• Reviewer 2: 2
• Reviewer 3: 6
• Reviewer 4: 6

Still a tough round.

But shout-out to the AC for accepting this paper.

Foot

Reviewer 1 probably wouldn’t want to receive a review like this.

So do I. And you.

So write reviews respectfully, justlike what you want to receive.

Done.

What Happened

I was simply trying to run:

pip install check50

and he greeted me with:

error: externally-managed-environment

× This environment is externally managed ╰─> To install Python packages system-wide, try brew install xyz, where xyz is the package you are trying to install.

If you wish to install a Python library that isn’t in Homebrew, use a virtual environment:

python3 -m venv path/to/venv
source path/to/venv/bin/activate
python3 -m pip install xyz

If you wish to install a Python application that isn’t in Homebrew, it may be easiest to use ‘pipx install xyz’, which will manage a virtual environment for you. You can install pipx with

brew install pipx

You may restore the old behavior of pip by passing the ‘--break-system-packages’ flag to pip, or by adding ‘break-system-packages = true’ to your pip.conf file. The latter will permanently disable this error.

If you disable this error, we STRONGLY recommend that you additionally pass the ‘--user’ flag to pip, or set ‘user = true’ in your pip.conf file. Failure to do this can result in a broken Homebrew installation.

Read more about this behavior here: https://peps.python.org/pep-0668/

Praise the Homebrew Python, with all your heart, and with all your mind, and with all your soul.

The Problem

1. I don’t want to create a venv. I would’ve created one if I needed to. I wanted to install it globally.
2. The package isn’t available in Homebrew (and so do most other packages).
3. The message is terribly written. It sounds like bypassing it will cause a heart attack (or even worse).
4. Setting user = true and --break-system-packages in my pip.conf didn’t work. I had to pass the flags manually every time. (That might be on my side, but I can’t figure it out, unfortunately.)

Why It Happened

A maintainer of pip requested Homebrew to adopt this behavior in accordance with PEP 668 (see https://github.com/orgs/Homebrew/discussions/3404).

And that caused Homebrew Python’s pip to start barking ever since.

I do understand the purpose. It’s completely acceptable:

Some Homebrew formulas rely on specific Python package versions, and letting users mess up the global environment can absolutely break those formulas.

And the Homebrew Python was intended for Homebrew packages, unlike me who was using Homebrew for Python.

But the message was still terrible, as I’ve just said.

The Solution

Don’t use Homebrew’s Python. Instead, use pyenv, which could be installed using Homebrew:

brew install pyenv

For the rest of the installation, go to https://github.com/pyenv/pyenv. Me rewriting it here would simply create more confusion.

Before We End

Readers probably know that I intentionally keep my blog posts short. But I want to quote @mlindner’s comment from the Homebrew discussion thread above:

This is a garbage… Now this even extends to making Python unable to use its own package manager.

That comment was marked as disruptive and later drew responses from multiple Homebrew maintainers. I’m not here to attack anyone, but to @mlindner: I agree.

The error message is exactly the kind of writing I hated, and most readers of this blog probably would, too.

It doesn’t state the problem, doesn’t explain why, and doesn’t give an actual solution. I’ve seen countless Reddit posts complaining about this exact same issue. And it could’ve been avoided by simply adding “use pyenv”.

Done.

You’re absolutely right! In today’s rapidly evolving digital landscape, AI-generated texts have become increasingly common across blogs, emails, essays, and social media posts.

However, despite their growing popularity, these texts often suffer from a number of recurring stylistic issues that are worth discussing in greater detail👇

AI-generated texts tend to:

• ✨ Include redundant information that restates the same idea multiple times in slightly different ways
• 📈 Expand phrasing unnecessarily, making short points far longer than needed
• ➖ Be filled with em dashes — everywhere — constantly — for no clear reason
• 📚 Use advanced vocabulary in situations where simple words would work perfectly well

Taken together, these patterns don’t merely clutter writing — they actively dilute clarity and reduce structural information density.

This isn’t abuse — this is pollution.

Would you like me to generate a comprehensive, in-depth, multi-section blog post exploring why AI-generated texts are problematic, complete with examples, historical context, actionable takeaways, and a concluding summary? 😊📌

OK. Time to stop.

I tried my best to imitate ChatGPT’s writing style. And I guess it probably fooled you (hopefully).

But this isn’t even a joke now. It’s not even supposed to be funny anymore.

AI-generated texts are actually polluting our environment: perfect grammar, long paragraphs,

AND NO INFORMATION AT ALL.

Takeaway: Don’t write like this.

Done.

P.S. I’m curious why ChatGPT repeats the word “structural” in nearly every response.

Yin–Yang and the Five Agents are the foundations of Traditional Chinese Medicine (TCM). While these concepts might seem difficult and mysterious—not only to foreigners, but even to native Chinese speakers—they are, at their core, simply two classification methods.

Tai-chi (太極)

Not Kung-fu. No.

Tai-chi is simply the thing as a whole that you want to classify. After deciding what is the Tai-chi, you can decide how to classify it, using the methods below.

Yin–Yang (陰陽)

Simply binary.

Yes. I’m not joking.

The core of Yin–Yang theory is to classify anything into two parts. For example, death is Yin, and life is Yang. Night is Yin, and day is Yang.

Five Agents (a.k.a. Five Elements) (五行)

Simply the Pro Max version of Yin–Yang. They classify things into two; we classify things into five.

The order of the Five Agents is: Wood (木), Fire (火), Earth (土), Metal (金), and Water (水). Wood and Fire are Yang, Metal and Water are Yin, and Earth sits between them.

Why this specific order? Because they generate each other in this order (the generative cycle).

Let’s use an example:

• Wood burns as fuel, which generates Fire

• Fire produces ashes, which are equivalent to Earth

• Earth includes Metal in it

• Metal condenses moisture into Water

• Water supports the growth of Wood

and it repeats.

In addition to the generative cycle, the Five Agents also restrain each other through what is called the destructive cycle (相剋). You don’t need to memorize a new order—it simply moves two steps forward each time, compared to one step forward in the generative cycle.

In this pattern:

• Wood restrains Earth (like how roots penetrate soil)

• Earth restrains Water (like how soil absorbs water)

• Water restrains Fire (like how water extinguishes fire)

• Fire restrains Metal (like how fire melts metal)

• Metal restrains Wood (like how metal cuts wood)

In TCM, the Five Agents are also used to categorize the five major organ systems:

• Wood → Liver (肝)
• Fire → Heart (心)
• Earth → Spleen (脾)
• Metal → Lung (肺)
• Water → Kidney (腎)

These are not just anatomical organs in the modern biomedical sense, but functional systems in your body.

Conclusion

Don’t study TCM. Unless you want to.

I’ve been a paid user of ChatGPT Plus ever since they released it three years ago. $20 per month is the most expensive subscription I’ve ever bought in my life. The price used to be worth it, but obviously it isn’t anymore.

OpenAI is well known for censoring—they try to censor any word that might get them in trouble. But to be honest, the censorship system has become unbearable since GPT-5 was released.

While I chose GPT-5 Instant, it sometimes goes through a mandatory, unskippable “thinking” process. I didn’t understand at first why they force this “thinking” process, given that I’m already a paid user and chose the Instant mode, obviously someone who doesn’t want to waste time.

But after browsing hundreds of Reddit posts, I found out that it’s part of the censorship system. Whenever the system “thinks” that you might have an “emotional problem”—which even includes using daily words like “knife”—it forces you into the thinking-mini model to ensure you don’t have “mental health issues”.

Obviously, the system has never actually stopped anyone with mental health issues from committing suicide, and I don’t have mental issues, of course. But it’s driving normal users nuts, and might’ve caused mental health issues among normal users like us.

I won’t stop subscribing—for ChatGPT is the only product that’s anywhere near fulfilling my needs—but OpenAI has given me every reason to hate them. They’ve never listened to suggestions, and they will continue ignoring them until a real substitute comes and kills OpenAI.

About a month ago on my birthday, my friend @ZigaoWang gave me a Keychron K2 Max with Banana switches.

However, the Fn + F shortcut which I often use for fullscreening did not work as usual.

After investigating further, I found out that Macs only treat Fn keys from Apple’s own keyboards as the Globe key (which is different from the Fn key despite its location), and ignore this signal if it comes from third-party keyboards.

The workaround is to create a keyboard macro which remaps Fn + F to Command + Control + F, a shortcut that works universally, regardless of the keyboard’s brand.

Steps

1. Connect your Keychron K2 Max to your Mac through a USB-C cable, and turn the keyboard to Cable mode.
2. Visit Keychron Launcher, choose your keyboard and connect, and then choose Macro on the left sidebar.

1. Record a macro. I use M0 as an example here, with the following content:

{+KC_LGUIX+KC_LCTLX+KC_FX-KC_LGUIK-KC_LCTLX-KC_F}

Which is equivalent to pressing the left Control key, left Command key, and F key simultaneously, and clicking Submit after releasing them.

1. Finally, choose Keyboard from the left sidebar, choose Layer 1, and map the F key to the corresponding macro (M0 in this example).

This should permanently resolve the issue. Enjoy!