Hey there!

Here's what's in this email:

• Spark: Integration with Kafka, Rabbit, and SQS course 100% published on the Standard plan.
• Why does AI keep getting worse?
• New edition of the AI Workshop.

Estimated reading time: 3 minutes.

🐰 Spark: Integration with Kafka, Rabbit, and SQS Course

We've just published the course on the Standard plan.

Apache Spark's true power is unlocked when you can integrate it with different messaging systems to process data in real time.

In this course, you'll learn how to connect Spark with the most popular tools in the ecosystem: Kafka, RabbitMQ, and AWS SQS.

Among other things, the course covers:

• 🚀 Kafka and Spark: Learn how to integrate Kafka with Spark to handle real-time events with a production-ready setup.
• 🏗️ AWS SQS and Spark: Discover how to use Spark to process data from AWS SQS, an ideal solution for serverless architectures.
• 🐰 RabbitMQ and Spark: Master the integration with RabbitMQ and build robust, scalable data flows in distributed architectures.

🤷 Why does AI keep getting worse?

I remember when Claude 4 Opus came out. It felt like a model that worked AMAZINGLY for coding tasks. I don't have that feeling anymore today.

Today it feels like a good model, but not as good as when it first launched. This is likely due to 2 factors (one technical, one not).

🔢 They've quantized the model Quantization is a compression technique used to make models take up less memory.

This is achieved by converting the model weights (the brain of how they work) that were originally stored as float32 into smaller representations like int8 or even smaller.

When new models are released, they're put out at full power to learn how people use them, and from there, once the data has been collected, the model is quantized so it can scale.

Quantizing models makes them somewhat less precise, especially for complex tasks.

👓 The degradation illusion The phenomenon known as "the degradation illusion" also plays a role.

As we become more familiar with the model, our expectations increase, and errors that previously went unnoticed become more visible.

So, freshly released models are probably better than the same ones after they've been around for a while. Time to make the most of them while you can!

Tomorrow on Cafe con Codely, at 9 AM CEST, we'll be discussing it. You can follow along on Twitch and YouTube.

✨ 4th edition of the AI Workshop now open

And yet another week (I've lost count at this point), we want to thank you once again.

In the last newsletter we announced the final spots for the third edition. Shortly after sending it, the edition filled up, so we've opened a fourth edition. 🥹

And since you've made it this far into the newsletter, here's the joke of the week, which I know you've been waiting for: > What did the Java code say to the C code? -- You have no class 😂

See you around!