Personal Blog

There’s always a lot of down time around Christmas, and I always end up getting stuck back in to Reddit. Almost like an addiction or something.

I always regret it.

The enshittification, the infinite scroll, and the sore neck I get for killing hours on there.

I also really like the format of Daily Dose of Internet on YouTube, where every few days they publish a sub 10 minute video of the best social media videos. No “reactions”, no filler.

I was thinking, well maybe I could do that for Reddit. Scrape the best 20 or so posts every day, and put it in an RSS feed.

So I built DDOR.

One limitation, the enshittification of Reddit means there is no longer an open API.

So I switched to Lemmy, and its working pretty well.

What I came to realise though, is that my ‘Top posts’ algorithm is flawed.

Because there are so many communities (subreddits), you have to take the top of each one, and combine them.

I’m taking the last 24 hours of posts, which means that posts closer the time I take my snapshot of those communities, will have a lower score, and therefore won’t be as high up.

This leads to me potentially missing high quality posts that haven’t reached their pinacle yet.

I could take quite a few approaches. I thought about looking at yesterdays posts, but then the oldest post might be 48 hours old. I could look at the trajectory of a post, how many votes it got since it’s been published, but that is potentially noisy. I could put the posts in ’top’ and ‘hot’ buckets. Y Combinators Hacker News has it’s own ranking algorithm published, so does Twitter/X.

But then I realised the rabbit hole I’m going down is the exact same thing all these social media companies are trying to do to keep you on the site for longer, and they have got in trouble for.

Not that I expect to get in to trouble, and I’m not trying to keep myself on my feed reader for longer. I just want to kill 10 minutes when I have the time to.

I’m ok with missing a few posts, and I believe that’s a healthier place to be.

I’m recording some data about the posts I am putting in my RSS feed, so I could use that later if I want. I know that there are about 140 new posts per day on the Lemmy communities I’m following, and I re-publish about 15% of the daily posts.

I have also designed DDOR to be flexible so I can implement other alogithems at some point.

But what I found really interesting was that my silly little side project has lead me to have more of an understanding of what social media companies are doing, why they are doing it.

We tend to think these algoithms are mystical black boxes designed to enslave us. But really they’re hard problems with multiple solutions, and none of them right.

Hopefully when you next think of these alogithems you, like I, will have more insights in to what that means.

So the second GTA IV trailer is out and is heading to be one of the most watched videos on YouTube in the shortest amount of time.

And it looks great.

But I would also temper that with some caution.

The trailer makes the game look perfect, but that’s exactly its job.

At the end of the trailer it says it was captured on a PS5. Presumably a PS5 Pro with full ray tracing, because why would you not?

So given the game is currently slated to be out in over a years time, there’s a reasonable assumption that the tech will have moved on and perhaps look even better. Let’s not forget the PC release will happen another year after that, at the earliest.

But what if you don’t have a PS5 Pro? Presumably GTA IV is designed to run on current generation consoles.

Will it look as good when it actually gets to release.

One of the things that bugs me is that with GTA V there were quite some differences between the trailer and the game that came out.

Perhaps Rockstar will address these, or perhaps, as is suspected with GTA V, they really had to squeeze the performance out of end of life consoles and made cuts.

The main difference is removing the density of life from the game. The number of people, the amount of vegetation.

I would argue this removed a lot of the immersion.

One thing I would so though is that GTA IV doesn’t look like it adds much in terms of graphics to the game. Trailers for Cyberpunk 2077 (2020) look comparable to GTA IV.

But before you head to the comments, where the difference lies is the density and variety in game.

So I’m interested to see if this lasts to the finished product.

I was asked to tl;dr Andor Season 1 as I was reaching it.

The gist of it is that a child called Andor is kidnapped from his home planet by scrap dealers.

Immediately ends up in prison for like 5 years.

When he leaves he immediately steals a MacGuffin from the Empire.

He then decides his next step is to go looking for his sister in a brothel for some reason.

Ends up in a fight with some security guards who he overpowers and shoots.

He then tries to sell the MacGuffin he stole but the guy who wants it decides he is the perfect person to fly ships despite never having demonstrated that ability before.

Andor is being pursued by more security guards, and is forced join a group of rebels who are going to rob a bank.

Everyone male on the bank heist dies, but they do get the money.

Because they robbed a bank the Empire cracks down on security. So Andor thinks he will retire but gets sent to prison instead.

In prison he helps gollum build the Death Star and discovers gollum can’t swim.

There’s also a bit where Ginny Weasley is annoyed about the security guys being shot so falls in love with an FBI agent.

I don’t think I’ve seen anything like it.

Doritos have a Minecraft promotion on.

McDonald’s have a Minecraft promotion on.

Even Lush have a Minecraft promotion on.

I’m certain there are others I haven’t spotted.

And the movie isn’t out yet. Tomorrow.

It could be terrible.

I haven’t seen anything this heavily promoted in many many years.

Severance is amazing. Just everything about it. I enjoy the bleak humour in it too. It’s a very funny show.

Anyway, one thing that has bothered me for a while is what is Cold Harbour? What are they doing on those computers?

I think season 2 episode 7 (Chikhai Bardo) was very telling.

But to explain my theory, I have to provide some context.

I think Lumon provides a service where people can avoid events in their lives. We’ve now seen examples of this where the woman doesn’t recall giving birth, and Ms Casey visiting the dentist repeatedly. Events that are unavoidable.

What we saw in Chikhai Bardo is Ms Casey visiting different scenarios that some people may want to avoid.

My theory is these are tests for their product, to check that they work in common scenarios.

Each of these tests have codenames next to the doors named after places. Cold Harbour is one of those doors.

So my theory is that Cold Harbour is the test for a working environment, which if successful, leads to selling it as a new product.

Within Cold Harbour they have lots of different kinds of jobs from menial data entry to art production, to shepherding.

So what I’m suggesting is that what Mark and the team are up to by moving the numbers around, is absolutely nothing. It’s work that simulates work.

And I think that is the biggest joke in the show. Lots of people do menial work with arbitrary goals that doesn’t really do anything. Macro Data Refinement is no different to that.

My only question about that, is it seems like a service that only rich people are either aware of or can afford.

So would someone in a menial office job be able to afford this product?

Or perhaps companies would pay for it as a benefit?

If you’ve known me for 5 minutes, I’ll probably have told you how much I dislike all existing Infrastructure as Code tooling.

Of all of them I hate Terraform the least. HCL is another DSL to get your head around in this cloudy world of DSLs. And given its similarities, but not, to JSON the learning curve is sharper than it should really be. But once you’re comfortable with it, it’s very quick to work with.

But the thing I really hate is having to manage the state file.

The recommended route is to put it in S3 and then to have a state lock in DynamoDB, which is additional complexity and security you have to manage, and it still finds ways to mess up.

I don’t want to manage state myself, I want something that manages state for me.

GitLab has a really nice service that automatically manages Terraform state for you. But I use GitHub and not much is going to change that.

Which got me thinking. GitLab has had this feature forever, maybe I missed something in GitHub that replicates it.

Possibly not something native, a quick Google confirms that. But maybe there’s a GitHub action that does something?

Searching the GitHub Actions marketplace did identify a couple of candidates that store and retrieve state from artifacts, which is a nice idea.

Some of them even encrypt the artifacts, I guess in case someone downloads them.

But encryption adds extra complexity when you could:

• Not put secrets in your state file
• Use a private repo

But it got me thinking, if you’re storing the state in artifacts, why not just store it in the repo. After all, it’s just a JSON file.

So that’s what I did. I created a blank branch called and stored and retrieved the state file from there.

There are some advantages over artifacts, such as being able to protect the branch and only allowing the workflow to manage that branch.

By itself it’s difficult to scale because there’s no state lock, two workflows could overwrite each other.

This isn’t something I’ve tried, but merge queues should allow you to prevent two workflows running at the same time.

So I think this is something worth exploring and trying for a bit.

If it works out I might write an action for it.

I think if it was an action it would be more elegant with a standing up and tearing down process, but it works well enough and was pretty quick to get going.

I’ve seen articles and videos explaining how GTA6 is going to have Generative AI for NPCs and other parts of the game.

This is clearly some over excitable thinking without really understanding what it means.

For one, GTA6 has been in development for at least 10 years. They will have already long recorded the audio of NPCs. ChatGPT really only caught the zeitgeist in the last 2 years, 3 if you’re generous. The game is going to be released, most likely in the next 12 months.

Rockstar are not going to throw away all their years of work to swap out some, still experimental, technology in the last rush to release.

Next, Generative AI is expensive.

I think there are three ways Rockstar could implement Generative AI for its NPCs.

It could:

• Use a third party service such as OpenAI
• It could run it’s own
• It could implement it locally

First, using a third party service would be crazy expensive. Think how many people would be using the service on the first day, how many times, and how many requests need to be made. I just cannot see how it would be worth it.

Perhaps Rockstar decide to roll their own. It’s a bold decision training a new LLM on a fictional environment. Expensive to train and still expensive to run with all those GPUs it would need.

Both these options would require single player to be an online only experience. Some game companies have tried that in the past, but we’re not there yet.

Finally perhaps Rockstar decides to run an LLM locally on your gaming device. That is compute that you are using that won’t be used to make the game look good on your screen.

Sure, lower power LLMs exist such as from DeepSeek, but this is a relatively new invention so Rockstar would have decided to become the world’s best AI company rather than a gaming company in the last few years.

Another aspect is moderation. How does Rockstar know, for certain that what the content generated is right for the environment. Will it know to generate Liberty City rather than New York for example? That could lead to lawsuits and Rockstar is generally opposed to being on the receiving end of those.

Finally, another aspect that people overlook is that Generative AI is pretty slow. It’s hidden nicely on the OpenAI UI but typically it loads in slower than you can read.

Sure the Santa AI was pretty convincing but it takes a non human beat to respond. How would that work in a game? NPCs see a car speeding past and by the time they say anything it’s long gone. That’s the expensive compute wasted.

If Rockstar does use Generative AI I think it will be for precomputed simulations of NPCs voices and stock phrases that get stored in game.

I think what is happening is that people are seeing Rockstar looking to use procedural generation, like Minecraft, and have somehow leapt to Generative AI, for clicks.

I’ve been listening to The Rest is Politics podcast for many years now. It’s still one of the most listened to podcasts in the UK.

However since the fall of the Conservative party in the UK, I’ve found the format struggles.

There are many reasons to dislike it, depending on your view it could be the 3 minutes of adverts at the start, the lack of transparency between what is the show and what is an advert, that they are constantly promoting their other stuff, the Trump Derangement Syndrome, that one half is a loudmouthed orchestrator of the Iraq war.

Whatever it is, there are alternatives out there, and some of them, I would argue are better.

What I’m listing as alternatives have central left leaning bias, that talks about UK centric news and politics, without banging on about Trump and Musk constantly.

Strong Message Here

This is one of my favourite new discoveries. Presented by Helen Lewis a journalist, and Armando Iannucci a writer and satirist, you might recognise his works such as Veep and one of my favourite films Death of Stalin.

In each episode they chose a word or phrase that has been used in the last week, and discuss the usage of that and related words mostly in politics, but also in everyday life.

It’s always amusing and brings out the hypocrisy of language.

They seem to be starting to discuss what they did last week, which annoys me, but it’s short and usually quickly linked to their main subject.

And because it’s BBC it has no ads.

Page 94

This is the podcast from the Private Eye magazine, which is typically considered a satirical paper, but you may have heard about the Horizon Scandal. Private Eye had been reporting on that for years.

The hosts are not very consistent, it usually has Ian Hislop on it, the stalwart of the great satirical panel show Have I got News for You for as long as I can remember. It also usually has Helen Lewis, mentioned above, and usually Andrew Hunter Murray, a writer for Private Eye, a question setter on the TV show QI, QI spin-off podcast No Such Thing as a Fish, and as well as other Private Eye journalists who pop in for a short interview about their subject.

Each show doesn’t really have a topic, and their focus is a mix of current events and longer running stuff that you probably haven’t heard of. It’s always entertaining and interesting.

Craziest of all it seems to barely have any ads.

Pod Save the UK

You’ll likely have heard of this one, from the same people who do Pod Save the World and Pod Save America.

Personally not my favourite, I really like Nish, but I find Coco’s voice grating and the first 20 minutes of waffle about what they did last week annoying and irrelevant. I suppose some people might like that thing.

They also interview people in their main feed which I also find annoying. Some of them are certainly interesting, but for me it should be a separate feed and stick to news and politics in the main feed.

It’s also alarmist left leaning, rather than left centerist.

But maybe some people like that thing.

I was at BSides London the other week, and there was a talk I wanted to go to on Distroless Containers. I wasn’t able to make it though as there were so many good talks I had to make a call and went to a different talk.

I have however done a bit of research on the subject of Distroless Containers over the last few days after writing my first tool in Go, and I recalled there is a way for a Docker image to load a Go binary directly and I was trying to work out if Distroless Containers were the path to that.

Turns out, no, it isn’t. What I wanted were scratch containers.

But to get to the point of this post, I learnt about Distroless Containers, and I wanted to share my understanding because there’s a lot of words written about it, and I wasn’t impressed with much of it. So I wanted to make it easier for others to understand.

The first thing is that they’re not Distroless at all. It’s Google’s whacky naming making things confusing.

Really Distroless Containers are the name for a way of building Docker images that keeps the file size small, and therefore have a reduced footprint and therefore a smaller attack surface, i.e.: more secure.

The second thing to understand is that it uses a pre-existing feature of Docker, in that a Dockerfile can contain multiple image definitions, or stages, that do different things.

In practice it means that in one stage of your definition, your image first builds the binaries it needs, then you use the second stage to run the binary. The second step throws away everything in the build step, and starts with a new clean environment, which you copy your binaries from the build step in to.

You are really just using the FROM parameter in your Dockerfile a second time. And Docker knows to discard the build step but still allows you to COPY your binary from it.

Turns out this was something I was doing unintentionally already. But I do love it when you discover a name for something that you happenee to be doing.

One thing I’ve noticed about YouTube comments is how positive they are.

YouTube comments have a reputation of being a lawless area. But, at least in my experience, over the last few years finding a negative comment is pretty difficult, to the inclusion of pretty obvious bot accounts.

Today there is a pretty interesting comparison taking place, videos from both Donald Trump and Justin Trudeau are trending on YouTube.

But the comments are the reverse of my expectations, fawning comments at the top of Trump’s post, and at the top of Trudeau, neutral to negative comments.

It seems very rare to see negative comments on YouTube videos that this was quite stark to me.

There are differences between the videos. Trumps is clearly a movie trailer for the next 4 years. The only thing missing is Redd Pepper narrating over the thing.

Whereas Trudeau is setting out a plan, a bit of data, and admitting some mistakes.

Personally I prefer Trudeau’s, it’s honest, and shows he knows how to politic. I would love it if more politicians spoke just as openly and frankly.

But that does not seem to be where the tide is moving, at least as far as YouTube comments go.

I’m not going to deny, I love Bluey. I love watching it with my son who loves watching it too.

The stories are great, the music is great, the animation is great, the characters are great, the humour is great. But it’s also really deep.

If there is one issue, it’s that Bandit (the dad) sets way too high expectations on not just entertaining your children, but the lifestyle with enormous (and impossible) houses with great weather.

But they also handle that, with their trademark good humour:

I’m not going to take advice from a cartoon dog!

It’s a kids TV show with a message, and is just as much for adults and kids.

It’s the only children’s TV show where after half a nights sleep, you get up at 5am, you watch a Bluey, and it makes you cry.

One thing that you appreciate as an adult, watching these episodes hundreds of times, is how things that happen at the beginning of the episode relate to things later in the episode.

My favourite example of this is the episode Explorers, in which Bluey actually has a very minor part, but the premise is that Jack and other kids in Blueys school are paddling a pretend boat, trying to find Australia. Jack’s dad is trying to pick up Jack from nursery, but his GPS fails and he gets lost. While he tries to find the nursery he truly gets to explore Australia. At the end Jack, exhausted from trying to find Australia is lying in the boat when his dad turns up. Jack says “is this Australia?” to which his dad says ‘yes’. The dual meaning here is that they have found Australia and explored Australia, with the audience understanding Australia a little better.

Fairy Tale

One of the concerns with Bluey is how they live such a great lifestyle. Partly it’s because it’s set in the beautiful city of Brisbane, Australia, but I also believe that Bluey is set about 10–15 years ago, when owning a large house was much more achievable.

My explanation for this is the episode Fairy Tale where Bandit tells the story of his life in the 80s. At that time he was around 10. Meaning if the show was set now, he and Chilli would be in their 50s with kids under 10.

That could be it, and that’s fine, but I think it’s more likely that the show is set in the 2010s.

Nexus Episode

One of my favourite episodes is the season 2 episode Helicopter where Bluey flies a helicopter with her different school friends.

I consider this a nexus episode, where there is interweaving of other stories by this episode.

The simplest example of this is Chloe who likes to explore the ocean, and in the very next episode Octopus, enjoys playing the Octopus game, and learns a lot about Octopus.

A more far reaching example is Winton who wants to go to his dad’s house, which has a pool.

This sets out several major plot points through this series and next.

First that Wintons dad is divorced, which leads him to dating the Terrier’s mum. Which then leads Winton’s dad to leave his house, which this episode establishes has a pool. And then at the end of season 3 episode, The Sign, the dogs that want a house with a pool, and, if you look very carefully, end up buying Winton’s dad’s house.

Favourite Episode

There’s not much to say here, but it’s Rain. The simplicity of almost zero spoken words, the beautiful story that is told, beautiful visuals, and beautiful sound track.

It’s just perfect.