2coffee.dev

Google is really stepping up. Recently, they released another 68-page version dedicated to Prompt Engineering. This time, it focuses more on API users.

Prompt Engineering

Last night we went into production but for some reason I got hit with two painful issues, everyone. I chose to work at night but still ended up sweating bullets.

First, npm threw me a curveball with an error that prevented me from installing packages. Talk about bad luck. After fumbling around for a bit, I was lucky enough to have run it on my machine before, so I zipped up the node_modules and pushed it to the server 😂

Second, GCP was running fine until it suddenly crashed after a while. I couldn't ssh into it, so I had no choice but to create a new VM and plug everything back in, and it worked normally again 😩

At one point, Groq bragged that they could run Meta's open-source model at a speed of over 1 thousand tokens per second. Recently, Cerebras - a service similar to Groq, has set a new record: Over 2 thousand tokens 🫣

Just imagine, you hit Enter and it responds immediately 🥶

Have you all noticed that ChatGPT has introduced the Sora icon on the slider on the left? Free accounts can now create 3 free images daily. The quality is quite good 🥳

Or you can try clicking on this link to see Sora

Following MCP, the A2A - Agent2Agent Protocol has just been introduced by Google a day or two ago. This is a protocol that promises to explode in the near future. Why?

As you know, MCP is a protocol introduced by Anthropic. It indicates how large language models (LLMs) can retrieve your data, load it into memory, and then process it. Imagine MCP as a USB-C port; as long as it meets the C standard, data exchange is very easy.

Since the concept of Agent has exploded, many Agents have been created to handle various tasks, posing a new problem of how to combine them together. That is, how can the Agent you create communicate with the Agent I create? That is when A2A comes into play. At this point, it is clear how enticing it is, isn’t it?

Announcing the Agent2Agent Protocol (A2A)

Today there are a lot of news. And I'm still... very busy 🥹

But the most notable news is probably that Google has just released Firebase Studio to assist developers in creating applications 🙏

Website haveibeenpwned.com has the ability to alert you if your account information has been leaked online. Just leave your email address, and whenever haveibeenpwned detects that the login information for that email address has been compromised, it will alert you immediately.

As for how haveibeenpwned.com works, it collects data from sources where information is sold online. It extracts information and the origin of the attack to notify registered users.

Since I signed up for this service, I occasionally receive notifications that my account has been leaked. The unfortunate part is that I can't even remember when I registered for that service or what the password was 😑. So, did I even use it? 😅

After about a month of testing the "Featured Posts" slider at the top of the homepage, I've decided to remove it today, everyone. There are a few reasons:

• First, not many users engaged with or showed interest in it.
• Second, managing it took too much time.
• Third, its design was a bit "off-tone" compared to what 2coffee already has.
• Fourth, I want to focus on more useful features rather than showcasing a slider that takes up too much display space.

Therefore, I've moved the "Featured" section to display outside the homepage next to the "New Posts" section. You can click on it to see the list of the most popular posts from last week. The list is updated every beginning of the week 😄

Took a natural dive into this repository, and it's pretty cool, everyone: charmbracelet/bubbletea. This is like a library that helps create CLI applications written in Go. But they focus more on TUI - which means Text-based User Interface.

Imagine when making a Web UI, you often look for component libraries or rely on tailwindcss... This is similar but designed for the Terminal 😁. I wonder if there's something similar for Node.js 🤔

Meta has announced Llama 4, the latest generation of their large multimodal language model, including Llama 4 Scout and Llama 4 Maverick, designed to provide a more personalized AI experience.

These models outperform their predecessors and many competing models across various evaluation standards, thanks to the mixture-of-experts architecture and advanced training techniques. Meta also introduced Llama 4 Behemoth, a larger model that serves as a "teacher" model to enhance the quality of smaller models.

As usual, these models are open source, so you can download and experience them right now.

The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation