2coffee.dev

Latest threads

Recent notes, ideas and conversations.

View all
Xuân Hoài Tống

Everyone is buzzing about the news that GPT‑6 Astra has officially been launched by OpenAI.

Astra scored 99,9% on ARC-AGI-3 and 100% on ExploitBench, 98% on FrontierMath Tier 4, and has helped solve many long-standing problems in mathematics. In short, Astra has surpassed the basic human action effectiveness level at 96%, and overall is no different from an ordinary person. Sounds familiar, right, that's right, this is the legendary AGI (Artificial General Intelligence), everyone.

Many people believe that Astra's 99.9% score was actually achieved thanks to OpenAI using its own specially optimized framework (responses API harness). When run with the test organizers' neutral framework, Astra only scores around 62% - 62.7%. Astra's Intelligence Index on Artificial Analysis is only 61 points, ranking alongside Grok 4.6 or GPT-5.6 Sol and even behind Opus 5 🫩. In general, scores only reflect part of the picture and shouldn't be trusted absolutely based on advertising like that.

Once again the information above may make some people feel pressured and that it's pointless to have to constantly learn new frameworks or tools, because they risk becoming obsolete after just a few months. Many programmers are deeply concerned about their career prospects amid the rapid-fire release of new models that are becoming increasingly refined. Not to mention if they fall into the hands of "bad actors" then what will happen to humanity 😩

(0)
Xuân Hoài Tống

Huh, I looked away and back and thought only Codex was down. Turns out they all went down together, Claude, xAI are gone too, guys.

Gemini is still alive and kicking, and the providers in China don't seem to be affected either. I wonder why they all went down together? Probably AWS or Cloudflare 🥶

status.openai.com status.claude.com status.x.ai

(0)
Xuân Hoài Tống

NVIDIA to acquires Hugging Face.

One side is a chip manufacturer, the other is a platform hosting millions of AI models. Wow, what products are we going to get next 😱

(0)

Articles

Longer stories about technology and the process of building products.

View all
other

AI Is Making Me Too Lazy to Think and Too Reliant on It

For the past few days, I have been adding features to my website. As I mentioned when I tore it all down and rebuilt it last time, I used AI to recreate most of what the old version had, moving its features and interface to a rewritten version built with a new stack. Up to 99% of the code was not written by me at all, but by AI. Still, I had to understand the core because I was the one guiding AI in that direction; I did not want to let it decide everything on its own because that could easily get out of control. Article data is stored in an SQLite file, then 11ty extracts it to generate completely static HTML pages. A script...

other

New Stack: 11ty, sql.js-httpvfs, and Preact.js

Hello readers of 2coffee.dev, four months have flown by like the wind, and this time I'm back with a new article. How have you been? How are you adapting in this era of AI dominance where almost everything in work is being automated? Every now and then, a few friends ask if I'm still writing. Do I use AI to write? Hmm... definitely not. Why would I use AI to write when anyone can use it to churn out countless articles to their liking? I mean, if they want, they can use AI to generate articles and read them. But if you ever feel the need for something fresh, come back here — because all the content here is crafted by me, with many hours of reflection.

other

5 lessons from practicing game play with 4 fingers

Since I was a child, I have known about the arcade game machine with a green or red joystick that looks very much like a chicken leg, on the left are 5-6 brightly lit buttons. Living in a relatively bustling place, I rarely saw a shop like this. I still remember that sometimes when I came back from school, it felt like some invisible force pulled me inside to watch what people were playing. If I managed to save 1-2 thousand for breakfast, there was no way I would remain a spectator. It's been too long to remember every detail, but the allure of video games is real. 2008 was the first year...

Posts

Concise notes from what I am learning and making.

View all

Profiling Technique - Finding and Removing Performance Bottlenecks in Node.js

profiling

The first one or two years after starting work are often the most energetic period in a programmer's career. I could imagine many ways to solve a problem, and even think about which approach would be the most "elegant". The reasoning was simple: if an approach was recommended and used by many people, it must be the best one. A piece of logic that initially looked simple could become as thick as a burger after passing through several layers of thought, because a function had to pass through several wrappers to be ready to "cover" many future cases.

Writing code can be an enjoyable job. Solving a problem can lift my mood for the whole day, give me stories to talk about with colleagues, and let us analyze and dissect problems that seemed unknown to everyone else. Yet most conversations eventually come back to one question: "What about performance?" Getting the code to run is one thing, but have you ever asked how to know whether your code is good enough? Many people think that simply following Best Practices will automatically make it run as fast as possible. That is true, but not enough. If everyone were certain that their code was good, perfect systems would be everywhere. Whether code runs quickly depends on many factors, not just how it is written. People do not usually rely on intuition to judge whether a piece of logic is fast or slow. To prove it, they need data, or at least need to know which part of the code is taking time to process. Only after everything is laid out in front of them can they find a way to optimize it.

Profiling is a widely used

Musings about the debugger – What is DAP?

debugger

Over the past few days, I’ve switched from VSCode to using Zed for a few reasons. Zed is a code editor written in Rust, very lightweight, and its interface is similar to VSCode in many aspects. However, Zed still has its own differences and follows its own philosophy. Personally, I’ve used Zed since its early days; my first impression was that it was very fast and smooth. That said, many features I was used to in VSCode were missing in Zed, so I only stopped at the level of experimentation.

One feature I use very frequently is debugging with breakpoints: selecting positions and observing the data flow in the system. VSCode is a popular editor, so enabling debugging is much easier. It usually takes only a few steps, and especially if you use a framework, you can easily find guides online. In general, VSCode does a very good job of simplifying things for users.

Therefore, after switching to Zed,

MCP – Prompts & Resources

mcp

This may be the last post in the basic MCP series on the blog. Originally I had planned not to write any more because the documentation for Prompts & Resources in MCP is very limited, but since these two topics are interesting I decided to introduce them to readers.

I have read and studied the material about these two items on the MCP main site. After understanding the principles and use‑cases, I looked for projects that have integrated them and intended to add examples to this article, but it turned out that very few projects actually implement these two components. Why is that?

In my opinion, a few

Hoài

Software Engineer, Blogger, Mentor...

I'm Hoai - a software engineer who enjoys turning ideas into useful products and writing about the journey along the way.

Explore
Illustration of Hoai waving