Barry O'Sullivan

IME most of the code documentation generated by LLMs is redundant.

The dev didn't want to write docs, got lazy, and orderrd the summary machine to generate something. They never cared about the output, it was simply a box ticking exercise. The resulting docs tell you nothing useful.

a default RPG Maker forest. An anime character says "Fascinating, Mr. Data. Everything in the environment appears to be made of these hard-edged squares..." Also the anime character is Jean Luc Picard

Check out this RPG Maker game about the crew from Star Trek TNG going into a holodeck sim that Data made from default RPG Maker assets

coleo_kin.itch.io/datas-advent...

"Why is my feed full of shit?"
* Realises I'm in the discover tab *
"Ohhhhh..."

As always "just" is doing some industrial level lifting in those statements.

I will always repost when Outer Wilds is on sale, one of the absolute best games ever made

It's only fair that if I spend $15 to see a film and I like it, the director now works for me, and his/her next film must be created in accordance with my desires.

We're treating writing code as the bottleneck, but really the bottleneck is the impact of legacy decisions. Design is about minimising these impacts. Instead we're accelerating them by writing code faster and compounding these problems.

Nice! Looking forward this.

Have one more 👍. Really enjoying the game so far. The writing is top notch. I'm at the micro-worlds bit and it has me in stitches.

It's been hard (it's being hard?) to learn how to make human connections. It's still a learnable skill.

Just bounced off Notion and switched to Sublime Text for my markdown note taking needs.

Notion kept auto formatting things incorrectly and got in the way more than it helped.

Nuance evaporates when oversimplification serves someone's agenda.

Next year, we’re going to Amsterdam!! 🇳🇱

The Collaborative Software Design Workshop with @kenny.weave-it.org and @gienverschatse.com will take place on June 16 - 17 in Amsterdam and registrations are now open! 🎆 www.avanscoperta.it/en/training/...

I am trapped under a cat at at 1:15am. She's purring though, so fair trade.

I can personally recommend Lair of the Clockwork God. Great game!

I have again been asked if AI was used to make my game.

For transparency, my minions used generative AI to make all the parts of the game that didn't benefit from a human touch, were too boring to make or not important enough to craft by hand. In other words, we didn't use any generative AI.

Cryptmaster ($16.24) - mavis beacon is still teaching typing, but now she's doing it with a sword in hand at the bottom of a dungeon. do battles, cast spells, solve riddles... all by typing! type anything!

seriously, fuck around with words, see what happens. store.steampowered.com/app/1885110/...

We're using Macroscope AI for code analysis and half the time it gives us broken suggestions, including hallucinations where it makes up properties.

These tools are not production ready.

It's definitely worth revisiting the book, lots of great techniques.

I'm planning to do an information architecture session, get an idea of the different consumers, then see if LLMs can embed auto generated MD docs in those classes for different perspectives, then a separate process aggregates it.

We're gonna give this a shot.

Might be worth me re-reading @cyriux.bsky.social's Living Documentation, see if we can combine techniques to get somewhere useful.

One thing is sure, I don't trust the generated docs one bit. Everything MUST be reviewed.

Structure your docs in a way that code changes only trigger changes to specific docs. Can't rely on LLMs for this due to non-determinism. Better to have the LLM summarise the change, then a script aggregates that result with the previous unchanged summaries. Structure the doc yourself.

We're playing with LLM generated documentation. It sounds great in principle, but the non-deterministic nature of LLMs means the docs change every build and need constant reviewing for errors.

We have ideas on how to address this.

I've stopped reading PR descriptions written by LLMs. What's the point? They're usually overly verbose, don't make clear points, and are just poor rewording of the code change.

Why should I put effort into reading something you didn't write?

I'm using Notion at work for note taking and it's the wrong tool for the job. All the auto-formatting gets in the way more than it helps.

I'm gonna switch back to Sublime Text and markdown, then sync the files to Google Drive.

The Staff+ role is a fascinating place to be. Sometimes you feel accountable for everything yet nothing at the same time.

Added bonus, use the factory in the mocked classes actual tests, now your proving it actually accepts and handles that param config as expected by mocks. Contract tests for cheap!

Mocking method calls in unit tests gets so messy, especially when the method has 4+ params.

Solution: Bundle the params in a parameter object, have a factory for it in tests. BOOM, now your tests are cleaner!

Dread Delusion on Steam Dread Delusion is an open world RPG brimming with strange places and dark perils. Carve your own path through the flying continents of a shattered land. Discover curious towns, unearth occult secrets,...

This is the closest I've played. Strong Morrowind vibes, fairly recent too. Would love more like it.
store.steampowered.com/app/1574240/...

Good news: The support chat bot has finally given up and connected me to a real person.

Bad news: The real person's messages don't appear until I send a message. Looks like the polling or websocket implementation is broken.