Quantization from the ground up

Quantization from the ground up
20 minutes by Sam Rose

Quantization reduces the size of large language models by storing their parameters with fewer bits. This makes models smaller and faster, with only small accuracy loss. Sam explains how numbers are stored, how models use them, and how techniques like symmetric and asymmetric quantization work. He also shows how quality changes using metrics like perplexity, and why moderate compression, like 4–8 bits, works well in practice.

Improve accessibility and speed up your development
sponsored by Level Access

How can development teams deliver inclusive, compliant products without slowing down? Testing earlier is just table stakes. Check out our on-demand webinar for tools and strategies you can use to accelerate progress by bringing accessibility into your existing development workflows.

JPEG compression
11 minutes by Sophie L Wang

Sophie describes how JPEG compression makes images smaller by taking advantage of human vision. It separates brightness from color, reduces color detail, and converts image blocks into frequency values. Less noticeable details are removed, and the remaining data is compressed efficiently. This allows images to keep a similar appearance while using much less storage space.

Queueing requests queues your capacity problems, too
10 minutes by Matthew Hawthorne

Queues seem like a cheap fix for traffic spikes, but they hide capacity problems rather than solve them. A modest 10% traffic spike for just one hour can leave 360,000 requests waiting six minutes each. He says the only real fix is adding capacity. You either pay for it, or your customers pay with slow responses.

The gold standard of optimization: A look under the hood of RollerCoaster tycoon
31 minutes by Lars Thießen

RollerCoaster Tycoon ran surprisingly well on 1999 hardware by combining low-level coding with smart design choices. Guests wander randomly instead of using costly pathfinding, and crowds are simulated without collision checks. Because Chris Sawyer was both programmer and designer, he could shape game rules around what the hardware handled best, turning technical limits into gameplay features.

7 more common mistakes in architecture diagrams
6 minutes by Billy Pilger

System architecture diagrams are essential tools for documenting complex systems. However, common mistakes in these diagrams can lead to confusion, misinterpretation, and frustration for viewers. Billy gives a rundown of seven common mistakes to avoid.

And the most popular article from the last issue was:

• Warranty void if regenerated