Dong Liang

film scholar, computer nerd, family man

[Dong Liang @dliangthinks]

The Rise of CLI and why you should care (part 1 of a series) https://t.co/HZIH21IQvy

[Dong Liang @dliangthinks]

Qwen 3.5 27B thought for an hour !! and solved the shoe arrangement problem.
First time on a model I can run locally.

[Dong Liang @dliangthinks]

AI does not generate videos https://t.co/5azogi1qLE

[Dong Liang @dliangthinks]

To Think or not to Think — How ‘Reasoning’ in LLMs Evolved https://t.co/bTpdyrA3Vb

[Dong Liang @dliangthinks]

Quick write: Routing Power: How GPT-5 Takes Control from Users https://t.co/3SLiOp3G8r

[Dong Liang @dliangthinks]

Stop treating LLMs as simple in-and-out machines! https://t.co/Lmkmihsu1F

[Dong Liang @dliangthinks]

Local Intelligence, an Important Step in the Future of MAD (Mass AI Deployment) https://t.co/8HFGjt2HP3

[Dong Liang @dliangthinks]

AI in May 2025: Vibe coding is mainstream, visual AI gains ground with GPT-4o, video advances fast while XR lags. Voice AI thrives with Sesame. Agentic models progress modestly. Local intelligence grows. Prompt engineering is yesterday’s news.

[Dong Liang @dliangthinks]

Tried Meta’s new conversational AI. Natural tone is solid and responsiveness is fast—maybe too fast. But the AI lacks real conversational intelligence and quickly gets irritating. It seems the team still has a lot to learn about true dialogue design.

[Dong Liang @dliangthinks]

Tried Dia, here is my thoughts: https://t.co/3bVnGz8tdb

[Dong Liang @dliangthinks]

Old sci-fi missed social media—proof we often misjudge what's next. Today’s AR aims to recreate screens in 3D, but Viture's Immersive 3D flips the script. It turns your desktop into a living world. This could be the breakthrough everyone’s been waiting for.

[Dong Liang @dliangthinks]

Tried a new method: pair two AIs like XP—a "navigator" for strategy and a "driver" for code. Gemini struggled solo, but with O4-mini as second coder, the solution became clear. This duo approach could redefine AI coding workflows. Humans? Likely shifting into project leads.

[Dong Liang @dliangthinks]

I just published The Tesla Ecosystem: FSD, robotaxi and Optimus https://t.co/TYyVOWzvZN

[Dong Liang @dliangthinks]

I just published Deep Dive into Deep Research: Google vs. OpenAI https://t.co/8yxr6YiqGV

[Dong Liang @dliangthinks]

AI chat partners can't yet detect accents or emotion, but by monitoring your own voice during conversations—especially with proper gear like a Rode mic—you can improve your own pronunciation. The tech may be limited, but it’s still a useful practice tool.

[Dong Liang @dliangthinks]

I just published Of Egg Cookers and AI, How Much Thinking Do You Want With That Prompt? https://t.co/CVW40XLXJ8

[Dong Liang @dliangthinks]

Imported old chats from Claude and ChatGPT into Obsidian via Basic-memory—fun to see my first convo from July 2023. Tried cleaning irrelevant PDF data using Cursor and Claude, but no luck. In the end, Gemini wrote a Python script that solved everything in under a second.

[Dong Liang @dliangthinks]

Everyone’s talking about MCP servers, but the real opportunity might be in third-party MCP clients. With tools like Claude Desktop, Cursor, and soon ChatGPT’s desktop version, we’re beginning to see the rise of the “LLM OS”—a better name than “Model Context Protocol” anyway.

[Dong Liang @dliangthinks]

After a long time of hesitation, I finally dove into creating iOS Shortcuts. What once felt intimidating turned into a breakthrough—I built several that fit my needs exactly.

[Dong Liang @dliangthinks]

Just found BasicMemory, an open-source MCP server that brings modular, flexible memory to AI chats. It stores knowledge as entities, observations, and relationships—promising, but needs refining. Biggest questions: Is this the right structure, and how seamless is access?