Mosky Liu

tweets things.

[Mosky Liu @moskytw]

It still feels creepy that I can spin up 100+ machine learning models in a single night. 100+! 🤯

[Mosky Liu @moskytw]

@davy_0xff Beautiful!

[Mosky Liu @moskytw]

I'm looking for a new Vim color scheme. What are you all using these days?

[Mosky Liu @moskytw]

@typebrook Oh, wow, I like this one!

[Mosky Liu @moskytw]

I'm kind of good at regex, but an absolute wow moment with AI was watching it compose a complex regex in milliseconds for an rg dependency scan.

Not feeling replaced; just feeling like I can code at warp speed.

[Mosky Liu @moskytw]

One of the prompting techniques I will use is: “How would you ... ?”

This allows AI to use “a thinking unit” to plan instead of just doing. Thanks to CoT, you'll see much better results later in the conversation.

[Mosky Liu @moskytw]

People always say write detailed prompt (including myself as a teacher); however I am a lazy engineer, I always write short prompts. I've been enjoying the art to let AI write context by itself. 😎 What do you think?

[Mosky Liu @moskytw]

Since my side project is more related to data science, I still stick with Codex CLI (GPT-5.1). Hope I can find some people who have the same feeling here.

[Mosky Liu @moskytw]

So, I am wondering that people like Gemini 3 so much, so are devs switching to Gemini CLI? Or, it's more like another new tech lover hyper rather than a true productive model for devs?

[Mosky Liu @moskytw]

And, yes, you should always ask AI to fact-check its response. The AIs nowadays still are not so good at facts, even book knowledge–level facts.

[Mosky Liu @moskytw]

Looking forward to the day that I don't need to ask AI to fact-check its prev response.

[Mosky Liu @moskytw]

I just learnt that `zY` can copy a cell into sys clipboard. What the heck. AWESOME.

You can also search (/), filter (| then "), sort ([ or ]) your CSV in a decent command line viewer. A wonderful tool when you work with CLI AI.

[Mosky Liu @moskytw]

The best data format for AI is CSV, and the best tool for human reading CSV is vd (VisiData). SO GREAT!

You can ask your AI for the keystrokes, btw.

[Mosky Liu @moskytw]

I’ve been trying Warp lately. It uses the same AI engine as Cursor, but feels less capable. Interesting.

[Mosky Liu @moskytw]

The decision matrix keeps coming to mind. I feel like it was an early attempt by people in management to apply vector-like thinking—and it has ended up being one of the most effective methods.

[Mosky Liu @moskytw]

Being data nerds, we should totally have a metric for the domain: Functional Coverage Index = dot(R, C) / dot(R, R) ☺️

[Mosky Liu @moskytw]

Me: Choosing an AC brand is too hard.
Also me: Let’s make ChatGPT run cosine similarity. ✌🏻

[Mosky Liu @moskytw]

Cursor Taipei! Thank you for having me as a panel speaker.

[Mosky Liu @moskytw]

Good morning!

[AshutoshShrivastava @ai_for_success]

How true is this? 😬

[Mosky Liu @moskytw]

Love the study! He fixed the “9.11 > 9.9” LLM issue by setting the “bible” neuron activations to 0. 👍🏻

[Kevin Meng @mengk20]

why do language models think 9.11 > 9.9?

at @transluceAI we stumbled upon a surprisingly simple explanation - and a bugfix that doesn't use any re-training or prompting.

turns out, it's about months, dates, September 11th, and... the Bible?