For the past few days, whenever I turned on my computer, the battery would drop from 100% to 50%, even though this never happened before. I thought the battery was degrading, but it turns out the culprit was OpenCode. I usually have about 4–5 OpenCode windows open, just sitting there without running anything, yet the battery drain is still terrible. Then I realized that whenever the laptop got really hot, I kept assuming Chrome was loading some heavy webpages.
I really like OpenCode, but I might have to find an alternative... What are you all using?
Just this morning I read this post and it’s so interesting, everyone. Usually, to make a cup of espresso you have to push hot water through a packed coffee puck in the portafilter at about 9 atm pressure. But recently some researchers have created an espresso machine that uses water at room temperature. According to them, the technology uses ultrasonic waves to drive water through finely ground coffee, fully extracting its flavor. In a test with 100 people, most didn’t even realize it was espresso made with this new method 😆
I used sound waves to make espresso. It could cut coffee‑brewing energy use by 75%
GLM-5.2 was released recently and quickly took the top spot in the open‑source large language model rankings. It has drawn a lot of praise and is being compared to Opus 4.8—or even Fable 5—with many saying it delivers acceptable results at a much lower cost.
OpenCode Go has also been updated to 5.2, but I haven’t tried it yet because V4 Flash is just too great 😆.
There are many ways to turn SQL database tables into diagrams—each table becomes a box that lists its columns, and lines between the boxes show the relationships. The most important thing is to pick a tool that matches the level of simplicity or complexity you need.
If you prefer a simple, visual solution, give sqltoerdiagram.com a try.
For the past month I’ve been using DeepSeek V4 Flash regularly at work. I have to admit the model is great, balancing speed and quality well. Previously I used OpenCode’s Go subscription, which offered a very generous limit for this model— it never ran out.
Then I thought, instead of paying $10 for the Go plan, I’d try a pay‑as‑you‑go approach with my own DeepSeek API key. After just half a day I’ve spent about $0.50, which actually ends up costing more than the Go subscription 🤔
Claude Fable 5 – the newest and most powerful model from Anthropic – has just been released. It’s a fork of Mythos, a name that has been testing the security limits of software architecture, but this version has been tuned to reduce its strength significantly.
What’s noteworthy is the price: $10 for 1 million input tokens and $50 for 1 million output tokens. Fable 5 is currently available only in Pro‑or‑higher plans, and after June 22 it will disappear, requiring you to add credit to keep using it.
Has anyone tried it yet? I just ran one prompt and already used more than 20 % of the limit 🥶
Has anyone felt confused lately? When you look around, AI agents and increasingly smart large language models are everywhere, and it makes you wonder: if artificial intelligence could completely do what we do, what should we be doing?
Improving ourselves is probably the most sensible answer. If AI takes over our jobs, we’ll need to move to higher‑level work, such as managing them. But what happens when they become capable of management themselves? This creates a feedback loop that forces us to keep advancing, or we risk being left behind. Everyone can see the problem, but not everyone knows how to get past it. Sometimes we just have to sit down and reflect, friends.
LLMs are eroding my software engineering career and I don't know what to do
Anyway, getdesign.md is a website that collects Markdown files so AI agents can emulate the designs of many popular sites today. Imagine being able to create a project that carries the vibe of leading tech companies.
It sounds appealing, but does it choke creativity 🤔? And what about copyright concerns 🫣?
What do you think about AI taking part in its own self‑improving loop? At Anthropic, Claude now handles over 80 % of the company’s code, increasing productivity by about eight times. In other words, they’re using agents to develop AI and to create new agents that join the process.
Imagine, in the near future, what will happen when hardware capacity can no longer keep up with AI’s evolution 🤔
Recently Google released the Gemma 4 12B model, which many people praised for its flexibility and its ability to run on a personal computer. Users describe it as compact yet surprisingly good at answering questions. Of course, it can’t be compared with the most advanced paid models available today.
A few days later they launched [Gemma 4 QAT](Gemma 4 QAT models: Optimizing model compression for mobile and laptop efficiency) - quantized versions of Gemma 4 that aim to bring the quality of models with tens of billions of parameters to more efficient hardware. Overall, Google is making a strong effort to bring LLMs to personal devices. Have you tried it?