A Byte of Coding Issue 367

Sup?

As part of my recent foray into self-hosting a bunch of services, I finally got a cron job monitor set up (using healthchecks). I know I could just set up cron to email me the results of any job, but I’ve messed around with it in the past and it’s just a pain the ass. Appending && curl to the end of the command being executed is so much simpler.

If anyone has any recommendations for open source postgres database monitoring tools, let me know!

Anyway, here’s the issue.

PostgreSQL Index Types

Published: 12 March 2024

Tags: postgres, database

Vlad Mihalcea dives into the different index types in postges, focusing a lot on the b+tree index while also reviewing hash and gin index types (include json for gin).

Some highlights:

• “The B+Tree index type is the default option in PostgreSQL and many other relational database systems”

• hash index type can only be used for equality operations

• gin index can be used on json column related searches

40 years of programming

Published: 12 March 2024

Tags: philosophy

Lars Wirzenius reflects on 40 years of programming, including core skills, productivity, governance, politics and ethics, maintenance, working, planning, and a number of other topics.

Some highlights:

• “To do good work you need to take care of yourself” → I always have to remind myself of this

• writing software is more about communicating and working with other people, than actually typing code

• “Take care of yourself. Sleep. Eat. Exercise. Rest. Relax. Take care of other people, as best you can. People are important. Software is just fun.”

ArtPrompt: ASCII Art-based Jailbreak Attacks against Aligned LLMs

Published: 22 February 2024

Tags: ai, machine learning, infosec

Fengqing Jiang, Zhangchen Xu, Luyao Niu, Zhen Xiang, Bhaskar Ramasubramanian, Bo Li, and Radha Poovendran published a paper on how they used ASCII art to make LLMs do things they weren’t supposed to be able to do.

Some highlights:

• checking the model’s output for harmful content or rephrasing input can mitigate these kinds of attacks

• GPT-3.5, GPT-4, Gemini, Claude, and Llama2 were all vulnerable

• model security is hard yo