Steam Release

I’ve been busy lately making the game ready for Steam and here it is: https://store.steampowered.com/app/4430050/SLOP_FIGHTER/

With this project I have had a lot of uncertainty as to what is actually possible to achieve from a large language model. I have constantly felt like I am walking a tightrope above complete incomprehensibility, but over time I have managed to work some very fine threads and sew together some truly impressive features I only ever dreamed were possible. I go into more length on itch.io, but to sum up, I’ve managed to better contextualise the LLM’s understanding of all the words I’ve stuffed into it. This means they can, and will, narrate the way they use themselves more effectively. A monster with camouflage, like a chameleon, will reference it when an attack against them misses, for example. The monsters are more aware of their environment, their opponent, and themselves. All the different information about the monsters is being passed properly all the way through. There is still some degree of “creative compliance”. I also added laser beams and other special attacks.

Underlying the narration are extensive semantic graphs, which are simply blocks of words sorted into categories, like body parts, movement styles, and mutation vocabulary. The LoRA adapters, which attach to the LLM, are trained on sentences constructed out of these graphs. Each word, in consideration of the sentence structure it was trained on, is given numerical weight and at runtime the LLM relies on its internal mathematics to decide which word to generate.

Its decision-making when it comes to generating words can be influenced, kind of guided, by careful prompting. Essentially, if you give the LLM enough of the information it needs in the right way it can start to build the required context to generate narration that reflects its character.

It’s a problem I’m familiar with as an author. How do I get the reader to generate an image in their head based on the words I am using? How do I make this person understand? It’s kind of fun to think about how the challenges are the same for what is just a machine that generates words.

Here’s the link again: https://store.steampowered.com/app/4430050/SLOP_FIGHTER/

— hamish, 3:17 am, June 18, 2026

Housekeeping

1. I’ve finally moved the website to a new server. My last webhost was pretty hopeless and my updates weren’t going through. Par for the course. ¯\_(ツ)_/¯
2. I’ve dropped the price on Well’s Rest. I wasn’t aware I was asking for thirty-three American dollars to purchase, that’s insane. If you’ve pirated the ebook online, great! That version is of lesser quality. The only way you can obtain the best digital version of Well’s Rest is by paying me three dollars and fifty cents.
3. SLOP FIGHTER is finalised, at least until runtime LoRAs work with Qwen3.5. I’ve made a little page for it here. If anything is updated I’ll put it on that page or in an itch devlog, not on the main blog.
4. The new book is going well. I’ve updated the earlier chapters over on the Writing page.

— hamish, 2:33 am, April 8, 2026

colorless green ideas sleep furiously

UPDATE: Did anyone see Anthropic now somehow has a Tamagotchi-style buddy system built into Claude Code that parses your input and reacts dynamically? THAT’S WILD, DOGG.

Booyakasha

I was having an issue with SLOP FIGHTER where the monsters weren’t talking about their mutations well enough. There was something missing. The monsters needed to talk about more than their physical abilities. They needed to talk about the forces that power them and the forces that emerged from them, also. It wasn’t too hard to work in a new category (of vital essences, if you’re interested) that give the monster linguistic context to feed its attacks. This solution didn’t require significant restructuring or cause smudged outputs. I’m pretty happy about that.

I also managed to (ask Claude to) solve a problem with animal parts not being connected to their semantic actions, e.g. teeth biting, claws tearing, etc. It’s one of those things where, like, I obviously don’t know how to code, but I can tell when words aren’t working. The problems get harder to find as the LLM smudges its outputs and I can’t tell if it’s just the LLM dropping the ball or if my sentences aren’t put together correctly.

_{Recently updated with transitive verbs that account for the new mutation vocabulary}.

I realise that the title of this post doesn’t quite describe what my video game does. My syntax machine belongs to a different branch of linguistic theory that developed both from and in opposition to Noam Chomsky. I’m no expert in syntax, though. I’m learning. I am reading textbooks. One of SLOP FIGHTER‘s original goals was to be educational. You’re not supposed to look at it and go “This sentence doesn’t work, this game is dumb”, you’re supposed to ask “Why doesn’t this sentence work? What is wrong with this word that it doesn’t match the action?” because we know that, despite this LLM’s inconsistency, a syntactically valid sentence can be parsed by the human mind even if the words themselves don’t make sense.

SLOP FIGHTER is done now (lol), and I think it stands as an example of what LLMs, and language as a whole, can achieve. The underlying technology is also pretty interesting. LoRA adapters are interesting. Small models that can be integrated into real-world tech are interesting. Imagine you’re out in a field driving a tractor and you just ask it exactly how many litres of diesel are left. ‘Heyo Daisy, what you got in yer?’ for a simple example. There are likely all sorts of complex questions you could ask of a tractor.

big uuup

— hamish, 1:08 pm, March 23, 2026

SLOP FIGHTER is available now

It’s finished! There’s a trailer and more screenshots etc at this link:

https://quarter2.itch.io/slopfighter

The day I release it Qwen releases new LLMs, so keep an eye out for future updates. I will be converting the adapters to work with the newer Qwen3.5 as soon as I am able. For now, here are two clips that show off the CPU battles:

The game is free to download, but you are invited to donate a sum of three dollars and fifty cents. Not too bad for a mini-Westworld that won’t ever try to self-actualise. Or do a Blade Runner.

— hamish, 8:13 am, March 4, 2026