These past few days I've gotten "hooked" on helix editor again, everyone. I have to admit, every now and then when I come back to it, my "skill" at using keyboard shortcuts gets a little better.
This time I'm more determined than before, so I've managed to use it for longer, and I'm also gradually getting more used to having to cobble together a few tools to view "diff" in helix.
Không viết bà i nữa à ad?
Vẫn viết bạn ơi, xin lỗi mấy hôm nay mình có chút việc cá nhân.
I bookmarked the lazy-tmux.xyz repository quite a while ago, but only today did I remember it and get a chance to try it out.
This tool helps back up and restore sessions that are open in tmux. If you accidentally shut down the tmux server, all the data will be lost, for example when restarting the machine. It can be configured to save everything that has been output to the console, across all sessions and all windows, as you choose. I tried it out, and it can restore things to some extent, though it’s not 100% perfect yet. But at least the data in the console is still there, so you can see what you were doing before.
Anthropic just introduced Claude Opus 5.5 and then just a few hours later OpenAI also introduced GPT-6 Sol & Luna.
Because they launched so close together neither one seems to compare scores with the newly released one. But the commendable point is that both are cheaper than their predecessors.
I just discovered this pretty cool website, everyone, but it's only for MacOS.
macos-default.com compiles a lot of handy commands for tweaking MacOS settings without having to use the interface.
People who are used to using commands will probably like it because it's convenient, and you can even automate things too. And it has quite a few interesting settings that I thought weren't configurable anymore. For example, turning off the "shadow" when taking screenshots.
So awesome!
Qwen just opened an image generation model with only 7B parameters, the images it produces are very impressive, even better than Google's Nano Banana 2.0.
There are currently no hardware specifications announced, however from what I've found it requires at least 12-16GB VRAM, corresponding to around an NVIDIA RTX 3060 or higher.
Qwen-Image-2.1: Compact, Efficient, and Unified Image Creation
Do you know the website CSS-Tricks? It turns out DigitalOcean bought it and has been running it since 2022. Back when LLMs weren’t booming yet, it was already one of my favorite resources. The articles there are very well edited and extremely detailed.
But it seems the site is now in a shaky state, as there are reports that content publishing has stalled. That's sad news 🥲
The recent Anthropic report details the detection and mitigation of AI misuse, focusing on its large language model Claude. It covers activities from December 2025 to August 2026 across seven areas: cyber, influence, surveillance, fraud and scams, biological misuse, conventional weapons, and illicit distillation.
You can read the full report for details-it's lengthy-but the listed areas are familiar, and many sound dangerous without safeguards. Once again, AI proves to be a double‑edged sword.
Anthropic doesn't shy away from calling out those who are distilling their model. This activity happens daily on a massive scale, with numerous tricks to bypass strict technical barriers. Afterwards, the models are released as open source or sold at prices many times lower. It's hard to tell whether to feel happy or sad 🥲
Basing it on a Hacker News post complaining that the front page is now flooded with AI news, while other topics get drowned out and ignored, I read some comments saying that's normal because HN has gone through many phases where a single subject dominates for a long time. But that's not the only interesting part.
I discovered that unslop.news is a site created to filter AI‑related stories out of the feed, which makes the news board look noticeably different. In addition, lobste.rs is a community similar to HN but with much stricter moderation and posting rules; its articles are also heavily focused on computer science and are just as high‑quality.
Shopify has just announced that its mobile app has moved from React Native (RN) to native development. In 2020, Shopify engineers chose React Native for their main apps, aiming to write code once and run it everywhere. At that time, RN attracted a lot of attention from the mobile‑app community, but performance was a trade‑off. RN can speed up development, yet it requires significantly more effort to optimize.
Now they are switching back to native for a fairly straightforward reason: large language models have made cross‑platform coding almost obsolete. With the same design, AI can generate code for both platforms simultaneously, solving both performance and multi‑platform challenges.
DeepSeek has just released Flash 4.1, which is smarter and faster than the previous Flash V4.
Some users say the performance of 4.1 falls between Flash and Pro v4. It can reach over 400 tokens per second, while the earlier version was only around 100 tokens per second. This version also adds image‑understanding capabilities. The model is already accessible via API at the same price.
I've been using OpenAI for a long time, so I haven't tried DeepSeek yet 😅. I'll give it a look in a few days.