Wundervault Weekly

slopmill and why it took me all week to write this one

September 27, 2026

Hello readers. Welcome to issue 23, keep reading to see if I got mad dunks like Michael Jordan (I don’t). We have already dedicated a few issues here to AI writing and this week I’m going to show you the tool I created and even give you a chance to use it. But before we get there, I have a couple quick updates from the news and discourse around AI this week. Thank you readers for sticking with us on this “long and winding road that leads” us to no place in particular.

A basketball player reaches for a dunk that could still go either way.

I think I have this.

Slopmill took me all week

If you remember two weeks earlier (which feels like a year ago), I told you that I tried a new approach to writing this newsletter that would allow me to go faster that caused me to take twice as long to write the weekly newsletter. Well this week, I really outdid myself because I spent almost the entire week creating slopmill, which should allow all of you to write your own newsletter alongside your ai, in your own voice. I’m using slopmill to write this issue, which means the whole process has taken me the better part of a week, a new record!

Could Laya fit on my computer?

The discourse online is still very actively speaking about JEV and how that model is being tied in as a tool to work with the LLMs that we are all comfortable using. There’s also Laya, an open weight alternative to it from a different company, ConvAI Innovations.

If JEV sounds like I started in the middle of a conversation, I did. I wrote more about it in Issue 22. Go there first if you want the longer version, and then come back to see why Laya has me looking at it again.

I went looking at TypeSafe AI’s JEV again. You give it a question about something your software is handling, and it can return a choice or a score. Say a customer message comes in and your system needs to decide where to send it. Laya, from ConvAI Innovations, is an open-weight alternative from a different company. You can run it yourself, though ConvAI says you should expect to tune it for your particular task (famous last words).

If you’re wondering what either one would do for you, ask the LLM that knows your projects. Send it those links and ask where a small decision like that comes up in your work, and how using JEV or Laya might improve it. I would ask for an example using a project I actually have, then see whether the suggestion makes any sense. Have you got a recurring decision you’d try it on?

Wait, which model answered?

And while I was trying to figure out where Laya might fit, Opus 5.5 and Sol 6 showed up this week. I may need to start writing model names on little cards (and then finding the cards).

If you use Claude for free, you get Sonnet and Haiku. Opus needs Pro or higher. ChatGPT’s free plan gives you unlimited text chats with GPT-5.6 Luna, last generation’s small model. Sol 6 needs Plus or higher. So if you try a prompt after hearing what somebody did with Opus 5.5, check which model answered you. I have forgotten to do that.

So which one would I use, Opus 5.5 or Sol 6? I haven’t given them the same job yet. I’d like to hand both a question from something I’m already working on and see which answer I have to fuss with less. Maybe that answer changes with the next question.

I also want to know which model answered when somebody tells me they had a great experience with Claude or ChatGPT. I have caught myself comparing results without checking that first.

$Claude's post from September 22, 2026: "At its default effort setting, Opus 5.5 delivers frontier results for a fraction of the cost per task, often beating other models running at their highest settings. It also generates output more than 30% faster than Opus 5." Below the text, four line charts plot each model's scores across effort settings from low to max, with Opus 5.5 in orange above most of the other lines.$ $Claude's post from September 22, 2026: "At its default effort setting, Opus 5.5 delivers frontier results for a fraction of the cost per task, often beating other models running at their highest settings. It also generates output more than 30% faster than Opus 5." Below the text, four line charts plot each model's scores across effort settings from low to max, with Opus 5.5 in orange above most of the other lines.$

The post from Claude, September 22.

Claude’s comparison and charts are right above this paragraph. If you use a free account, check which model answered you before comparing your result with somebody else’s. Claude’s free plan gives you Sonnet and Haiku; Opus needs Pro or higher. ChatGPT’s free plan gives you unlimited text chats with GPT-5.6 Luna; Sol 6 needs Plus or higher.

You might be having a different experience with the same prompt. I have rewritten a prompt for the fifth time without checking the model name (yes, really).

A cheaper flight sounds pretty good

A reader told me they recently booked a trip using AI to find flight deals and were very satisfied with how it went. They set up scheduled tasks in ChatGPT and got alerts when a deal matched their criteria. I like this example because I can picture myself doing it, and then getting a notification while I’m supposed to be doing something else.

When I spend too long reading about AI discovering new medicines, solving math problems, or potentially ending humanity as we know it, I forget about finding a cheaper flight. The trip is a smaller story, but you can actually use that idea today. Give it something specific to watch for, then see whether the alerts are useful.

And now there is Meta Muse Spark, another model trying to find its way into consumer use. Meta says it now powers the Meta AI assistant. It is also coming to Meta AI across Instagram, Facebook, WhatsApp and Threads, in search bars, group chats and posts. I wonder if I could ask about somewhere a friend posted and end up with a useful trip idea. I haven’t tried it for a flight search. I’m curious what I’d hand it first, though. Maybe the question is whether I can think of a useful task before I open the chat box.

I started with JEV, wandered over to Laya, and then two more model names showed up while I was trying to write about them. Well, actually, I already had the names in my notes. The flight-deal story is the one I keep coming back to. Somebody gave AI a job with clear instructions and got a useful alert. I can manage that level of experimentation.

Have you set up a task like that? Tell me what you asked it to watch for, and whether it found anything worth your time.

Below the fold is how slopmill came together this week, and a chance to try it yourself for a week (I’d like to know what you do with it).

Until next week,

◆ Below the fold ◆

How slopmill came together, and a week to try it yourself.

Making slopmill

I started building slopmill this past week. The next day it had a name. Claude wrote the code, and every piece went through rounds of review by GPT-6 Sol before I saw it. A rather involved way to make a tool that was supposed to help me write the newsletter faster. You saw how that went above the fold.

This issue was written in it. I wrote the paragraphs I wanted to write, left prompts in the other places, and GPT-6 Sol drafted those parts in my voice. I could then read them and give a direction to rewrite what needed changing. My own paragraphs stay as I wrote them unless I change them myself, which matters when a tool has strong opinions about my grammar.

The process has three steps:

Plan: write your words and leave prompts where you want help. There is a Background box for facts you already have.
Draft: read each section, edit it, or ask for a rewrite with a direction. You can rewrite several sections at once. Slopmill looks things up on the web before writing each prompt, so there is at least a chance it has read about the thing I asked it to write about.
Proof: see the finished issue as a reader would. There is a spelling and grammar check, plus a fact check that reads the whole issue, including the parts you wrote yourself, and opens every link.

It can make pictures and charts, add section headings, and suggest titles. Once you’re done, you can download a web page, a Word file, or a zip. I have spent enough time moving words between places this week to appreciate the download button (and I built the thing).

The code is on GitHub. You can run it on your own computer with your own AI key. If you try writing something in it, I’d like to hear which part you ended up rewriting yourself.

Want to try slopmill yourself?

If you want to poke at it before installing anything, try slopmill here. You’ll get a private workspace with a practice issue, so you can start changing words without first figuring out what to write about. Or bring your own idea. I would probably start with a prompt I have been putting off and see what it does with it.

The trial runs for one week from when this issue goes out, or until $10 of AI credit is used, whichever comes first. Everyone shares that $10, so it might end early. Research is on. AI pictures are off because they cost far more than text, though you can upload your own pictures.

Your workspace lasts while you use it, so download your issue before you leave. If you want to keep going, you can install slopmill for free and use your own AI key on your computer. What did you make, and what broke?

“It’s a plan of my own invention. You may try it if you like.”— The White Knight, Through the Looking-Glass

← All Issues Subscribe