#636 – July 27, 2025
technical aspects and how to create custom cartridges from scratch
More than you wanted to know about how Game Boy cartridges work
about 1 hour by Allison Parrish
Allison describes the technical aspects of Game Boy cartridges and how to create custom cartridges from scratch. He covers how the Game Boy communicates with cartridge components through its address and data buses, the role of memory bank controllers in accessing larger ROMs, and techniques for implementing save functionality.
The CIAM Guide Vendors Don’t Want You to See
sponsored by Frontegg
Choosing a CIAM provider shouldn’t feel like rolling the dice. We created the 27-page, zero-fluff Totally Unsanctioned CIAM Vendor Guide to help you cut through the noise. No filler content. No vendor spin. Just the facts you need to make a smart CIAM decision. Download the guide today.
PostgreSQL at Scale: Database schema changes without downtime
18 minutes by James Coleman
James shares how to safely execute various DDL operations while maintaining service availability, focusing on maintaining lock times under two seconds. Key techniques include explicit lock acquisition with timeouts, using concurrent operations for index management, and breaking constraint additions into smaller steps.
TODOs aren’t for doing
6 minutes by Sophie Alpert
Sophie argues against automatically tracking or deleting TODO comments in codebases, suggesting that they serve valuable purposes beyond immediate action items. Well-crafted TODOs can provide context about edge cases, potential improvements, or the author's thinking at the time of writing, helping future developers understand why certain decisions were made even if the suggested changes are never implemented.
The big LLM architecture comparison
26 minutes by Sebastian Raschka
Sebastian provides a comprehensive overview of recent developments in LLM architectures, focusing on structural innovations rather than benchmark performance. He examines key models like DeepSeek V3/R1, OLMo 2, Gemma 3, and others, highlighting architectural innovations such as Multi-Head Latent Attention, Mixture-of-Experts, sliding window attention, and various normalization techniques that improve efficiency while maintaining performance.
Coding with LLMs in the summer of 2025
7 minutes by Salvatore Sanfilippo
LLMs can significantly enhance programmer productivity when used correctly. While these models excel at eliminating bugs, exploring ideas quickly, engaging in pair-design, accelerating work, and bridging knowledge gaps, they work best as amplifiers rather than standalone developers. Antirez recommends providing large context when working with LLMs, using top-tier models directly, and staying in the loop as the human expert guiding the process.
And the most popular article from the last issue was: