There are many evaluations of one-shotting working video games.  This evaluation is about how well the AI handles all the follow up edits, because that is a better test of understanding.  

After you dial up the context window, Qwen 3.6 is better than ChatGPT.

If you have 32 GB of RAM, you can run it.

Curious if this applies to other domains that are not as easy to test, or did the Qwen team just do a really good job fine tuning this on video games after seeing game development as a trending AI test.  Watching Gemma 4 vs Qwen 3.5 tests, it's clear the ability to one-shot anything resembling a working game is new for any model this small.

Limited testing is already suggesting the 35 billion parameter model is outperforming the 122 billion parameter model.  This is further proof that intelligence is going to be small, distributed, able to run on garbage hardware, and impossible to embargo, or monopolize.  This will not be winner take all network effects of social media.  This will not generate super normal profit for anyone but a few Nvidia/Google types who already have healthy moats and margins of physical scale, and a tech lead that doesn't depend on IP law.  Anyone still dependent on those types will at best become a hot restaurant with immediately jacked up rents that leave the proprietor at subsistence remuneration.

Ollama is probably the easiest way to run open source AI on your own computer.  Its biggest gotcha, is a default context window of 4k which effectively lobotomizes it without being obvious.  The model was "self aware" enough to answer why it was generating clearly incomplete files when asked,  but fixing persistently it in Ollama required making a new modelfile, and new model.  But once you change it, you can reliably edit, and add to an existing code base.

The main performance hit appears to show up in response times almost doubling every turn, as the entire conversation gets reprocessed every turn.  For an example like this, continually editing 500-1000 lines of code, version 1 takes an hour to finish, version 5 takes six hours to finish.  It looks like you're often better off starting a new conversation for each feature request, or bug fix, certainly in response time, possibly on response quality as well.

Racer game made entirely with Qwen.  Started with 3.5 122b, until turn five failed to fix a restart button requiring "loop();" at the end of the restart function.  Fixed with 3.6 35b, along with a few suggestions, and a difficult level 3.  3.6 turn one was a copy paste request to fix the RACE AGAIN button.  It generated a working snippet to show where to place loop.  Turn two was telling it the fix worked, and asking for "any other suggestions" to which it gave a few, including a pause system with P, a screen shake on crash, persistent high scores, mobile touch controls, and some non obvious code cleanups in snippet form.  I was unable to paste its snippets into the existing code, get anything to work, so I told it that on turn three, and asked for a complete .html which it generated, while shrinking lines of code from 608 to 574, and growing to file size from 18.9 to 19.9 kB.  Turn four asked for a Level 3.

Context Window is a control knob ChatGPT, Grok, Gemini, and Claude use to turn up, and down compute resources.  The details are a bit of a mystery, but it is becoming common knowledge that all the hosted models use some equivalent of rolling brownouts that feel like rolling lobotomies to manage peak demand.  At the free tier level, ChatGPT, and Gemini are unable to coherently edit a bare minimum game framework.  Asking for one thing, loses another, or sometimes several.  Forward progress in a coherent environment is impossible.  ChatGPT burns tokens talking up "visual enhancements" that transform squares into rectangles.  Grok appeared capable, consistent, and coherent for fourteen turns, and hit a wall of not being able to fix a win screen after seemingly infinite attempts.  Starting a new Grok conversation with the most recent working version, and asking for the fix did not work.  

For the time being, if I had to pick a tool to hammer out a complete working game of roughly 8 bit Nintendo Quality, it would be Qwen 3.6.

The cult of complexity strikes again.

Gone on for hours, and quoting hours more to restore service.  We don't know root cause because they don't know root cause.  No one does anymore.  They subordinated themselves to the machines, and have already injected a terminal amount into their own code, and configuration.

Even before AI, Microsoft had entered the realm of permanent chaos with the Windows 10 forced update policy, but with their all in bet on AI, nothing from them can be trusted anymore.

It has been said by cat breeders, that an early sign of genetic collapse is when the mothering instinct is bread out.

Relatively early Saturday night, just before 10pm, with a little more human noise than usually, Ninja, Rusty, Rocky, Fireball, and I ventured out, across the street, and down the hill.  We have about a 300 foot stretch of unusually green urban frontage, due to a setback requirement applied equally to apartments, and houses.  Lots of grass, and trees, that cats run wild in, playing as if in a giant jungle gym obstacle course.

It is effectively our front yard, and everyone has walked with me fully around the block, and some even a bit beyond, so it's normal now for some to hang behind, while others venture farther out.

Fireball had decided to stay around the top of the block in a brick, and iron gated garden, and Ninja decided to stop about halfway down the hill at a 40 foot almost spherical tree with multiple large, low angle climbing branches, and dense leaves that effectively create its own ecosystem of crickets.  Inside this foliage dome exists a different world.

Rusty, and Rocky continued to chase, and fight each other down about another 100 feet to almost the bottom of the hill.

That's where about a 30 pound raccoon emerged from the shadows.

Rusty, and Rocky were taking a bit of breather in front of a house where satellite imagery shows a small forest in the backyard. A little over 1000 square feet with 6 trees, one a 50 foot oak, the other a 100 foot redwood.  It might as well be the wild Sierra Nevada back in there.

I have never in my life seen such a large, healthy raccoon, and he was very clearly making a territorial claim advance towards Rusty.

I immediately raised my voice, while remaining in calm, and friendly mode announcing to our local area, "wow, you're a biiiiiiigggg raccoony......" as I walked directly toward Rusty, and Rusty held is position.

Rocky was practicing the art of invisibility.

Before I could even say the word raccoony a second time, I see Ninja sprinting down the sidewalk towards us.  She knew exactly what was happening, and had locked target on the raccoon before I could even say "Niiiiiiiiinjaaaaaaaa!!!!!!"

Ninja took about a 90 degree triangulated position relative to where Rusty and I were standing, and we both advanced toward the raccoon, at which point he turned, and did a slow retreat.  

I encouraged everyone to leave with me, and Rocky immediately appeared, and started sprinting uphill, and so did Ninja.

But Rusty held his ground there for probably another minute before he continued up with us.

Balls.

Rusty is almost a 14 pound cat, and his lengthiness, and fluffiness make him look even bigger.  Ninja is not even quite 8 pounds.  But she lives in an awareness field a step ahead of reality, and routinely sizes up animals 3-4 times her size as inferior.  I watch her stare down medium sized dogs as if they are prey animals.

In my mind, the boys have become The Queen's Guard, but she is still more Ninja than all of them.

The challenge to human mothers is some threats, often the worst threats, are not so obvious.  

If you do the small scale thought experiment of 100 people trying to survive on an island, food, fuel, and shelter become a clear priority.  You would more or less want to max these things out until you were pretty certain you had that taken care of.

Beyond that, you would start to think about how to make improvements to your life.

What you would probably not do is tell half the group to actively interfere with the other half of the group, invent conflict, or encourage people to destroy, and rebuild what they have already done.

But that is actually a lot of the modern workforce, particularly within the realm of information technology.  Almost every tech company, and software vendor is guilty of designed obsolescence, constant change, and security theater.  The three crimes feed each other, and take a toll on society far greater than dollars alone.  

They rob us of our time, and distract us from living life to its fullest, and moving forward rather than paying a tax to stay in place.

Fuck these criminals.

Saturday I thought it would be a good idea to take Rusty to the lake for dinner, so just before 7PM, I put him in a backpack carrier, and rode my bike 6 blocks down the hill to the lake, with half a can of cat food, and a spoon.

It was a little too noisy.  

While he did well on the ride, and didn't hardly complain at all, before I even unzipped the carrier, I could see he was a little perturbed, but thought letting him out would help.  We had set up on top of a small, fenced porch that was the entry to a rec center beside the lake, about 6 feet above the ground.  Lots of birds come out at this hour.  We had the local high position, and cover, and concealment, against the side of a closed building.

As soon as I let him out of the carrier, he flips into panic, evade, escape, and elude mode.  I imagine losing an eye if trying to physically intervene at this point, so I calmly say "it's OK" repeatedly in my high pitched talking to cats voice.

Within 30 seconds of getting out of the carrier, he has upgraded his position to on top of a support joist underneath a two foot overhang of the first floor over the foundation.

I'm thinking, OK, this works, we can eat here.

And about five seconds after that he discovers a small hole, barrier bigger than a cat's head, that leads into whatever is behind the foundation wall.

I immediately know we are in for a longer night than I planned, and pin 1 hour after whenever the music stops as our best case scenario for recovery.  The initial hope is maybe by 11pm, he crawls out the way he crawled in, and we go home.

Assuming he is not coming out for a while, I figure I can walk around a bit, and examine the perimeter of the building enough to be mostly confident that there's no other way out, and spot a clear human sized access panel that's secured with a small padlock, and four Philips head screws.

Before what looks like the last people for the night leave the park, I ask if perchance any of them have a Philips head screw driver I could borrow for two minutes.  This person who I had never seen before says, "you know, actually I have an entire tool kit I don't need, and you can just have it."  He gives it to me, and drives off.

It took nearly two hours to unscrew three, and a half of these four inch screws that were in two inches of the densest wood I have ever seen.  The last half of the last screw stripped before I could finish, and I had to bend, and break with the crescent wrench, and the pliers that I insisted I didn't need.

After imagining all manner of worst case scenario for what might be inside this crawl space next to the lake that has a literal parade of rats, and raccoons come out at night, I decided to enter with my phone light after making enough noise, and repeatedly singing "Rusty.......Big strong Rusty......"

The space started with about four feet of headroom, which became three, and then two, and then less feet of headroom as you made your way towards the back of the building.  There were multiple rooms, and a divider wall where he came in, but by dropping a string of lights from my bike down the entry hole from the outside, I could confirm he probably didn't get stuck between a wall, and probably came in on the larger main room I was in, and not the 18 inches of clearance section that was on the other side of the wall.  All around the perimeter, and upper sections were tiny sections that I hoped he was hiding in, and just not responding to my calls.  Crawling around looking for him would be insane.  But before I left, I estimated that due to gravity, vertical distance, and tight spaces, getting out looked more difficult than getting in, so I built him stairs out of old paint cans, closed the access panel, and put the screws partially back in.

It was probably about 11:30 by now, and I was about 98% confident he was in there, and OK, and about 60% confident that if I went home right then, he'd find is own way before sunrise.  But I also didn't want to leave him.  I didn't want him to think I was abandoning him.  I wanted him to understand we could go somewhere else, explore, and come back.  The grand vision of an adventure cat.  

My mind moves to what is a reasonable amount of time to wait?  If I were telling this story to someone, and I said I went home at 8, what would they think about how much I cared about my cat?  I figure midnight is a time I can live with, and anything beyond that, this cat has seriously decided to be on his own, and does not trust me one bit.  Midnight comes, and goes without notice.  One of the many inner voices says 1AM, stay till 1, then you can go home.  1AM arrives, and I stick my face in the hole one last time to call "Ruuuustyyyyyyyyyy....", and I see a single whisker extending an inch beyond that support joist.  Now I know I can't leave.  About five minutes of coaxing later, and I see the tip of a nose, and maybe 15 minutes after that, I see his face for a second.  But he is still hiding in this thing so much that I'm not even 100% certain he isn't stuck on something.  By about 1:30AM, he has his entire head, and one paw out of the hole, and I'm thinking he has to be physically able to get out, he's now just observing.  I considered trying a snatch, and grab, but thought that would be the quickest way to get him to flip out, and lose trust, so I placed a scoop of cat food on the joist in front of him, and watched him eat it slowly.  By about 1:45 he climbed out, hopped down, came towards me, and I was easily able to pick him up, and put him in the carrier.  

We finished the can of cat food outside in more familiar territory 6 blocks away.

As soon as the smell of cat pee is gone from the backpack carrier, we will try exploring again when it is much quieter.

Cats are perhaps the most fascinating of creatures to study closely.  I have five of them.  Mom, three sons, and another male foster swap for the sister, who wound up living in a palace in San Francisco.

The four boys still have their balls, which causes quite the consternation to many.

These are their stories.

👋 Welcome, it's great to have you here.

This site is built on perhaps the best web publishing platform ever made, the best version of Ghost ever released.  

The original vision of Ghost was a core of simplicity, and minimalism, particularly as opposed to the bloat of WordPress.  The version this site is built on is still secure to this day, while later versions introduced a severe security vulnerability in a feature that normal people didn't even want.  Everything about the project bloated, and strayed, but with Docker, it is fairly easy to run a specific version forever.

Security nuts will reflexively say that's a terrible idea, but ask them how many published vulnerabilities require some combination of local access to the computer, and working credentials.  "Yeah, after you get admin access, you can do this thing that you shouldn't be able to do, and so now you need to run my newer software."  That is almost verbatim their racket.  The only vulnerabilities that really matter are exceedingly rare, end up becoming front page news for anyone paying any kind of attention, and are typically introduced by bloat that nobody but a self important product manager, or portfolio builder asked for.

Quantum computing will be commercialized, and drain your bank account before this site gets hacked.

The Ghost development team's original welcome message:

We know that first impressions are important, so we've populated your new site with some initial getting started posts that will help you get familiar with everything in no time. This is the first one!

A few things you should know upfront:

1. Ghost is designed for ambitious, professional publishers who want to actively build a business around their content. That's who it works best for.
2. The entire platform can be modified and customised to suit your needs. It's very powerful, but does require some knowledge of code. Ghost is not necessarily a good platform for beginners or people who just want a simple personal blog.
3. For the best experience we recommend downloading the Ghost Desktop App for your computer, which is the best way to access your Ghost site on a desktop device.

Ghost is made by an independent non-profit organisation called the Ghost Foundation. We are 100% self funded by revenue from our Ghost(Pro) service, and every penny we make is re-invested into funding further development of free, open source technology for modern publishing.

The version of Ghost you are looking at right now would not have been made possible without generous contributions from the open source community.

Next up, the editor

The main thing you'll want to read about next is probably: the Ghost editor. This is where the good stuff happens.

By the way, once you're done reading, you can simply delete the default Ghost user from your team to remove all of these introductory posts!