Welcome to my personal blog! This is where I share thoughts, analysis, and commentary on various topics that interest me.

Contact

I’d love to hear from you! Feel free to reach out:

Email: admin@sukonbu.party

About the Site

This blog is built with Hugo and uses the Hugo BearCub theme. It’s hosted with care and maintained with passion for sharing interesting content.

Thanks for visiting!

Sukonbu respects your privacy and is committed to protecting your personal information.

Data Collection

Sukonbu does not collect user data outside of GitHub’s default data collection when you visit this website.

Third-Party Services

This website is hosted on GitHub Pages, which may collect basic analytics data as per their privacy policy.

Contact

For any privacy-related inquiries, please contact: privacy@sukonbu.party

Last year (2025), starting from around 7th of September I was a student in a university that was around 15 hours away from the city that I live in. For me, this is not a problem as I don’t really care about where I go because I rarely go outside myself. As long as I have my internet connection, my computer, some food and some water I’ll often be able to tough it out since most of the work I do is over my computer anyway.

When I went in there, there were only two “well known” fast food branches located near my dorms, one of them was I think Pizza Hut and other was Dominos. There were also a bunch of other local businesses located around that would sell some common food like the infamous Döner.

The city I went to was so low in population, I think if you added up all the students from all the dormitories it would equal to almost a quarter of all the locals there. Because of this, a lot of the attractions there was focused on getting us, the university students to spend their money on. While this is great for new businesses opening, the existing businesses and student related buildings had a other story, they would often suck, act hostile against students, food businesses would often cancel orders when ordered through food delivery apps and when called on phone would not deliver to the dorms because it’s too far or charge additional fees. The bus operations there was also managed by a single family that were the owners of the village that the university resided in (our university was 2 hours walking distance to the main city where our dorms were at!) so as you may guessed they were more interested in serving their own villagers travel instead of helping the students (who are all paying customers by the way) reach their university in time. When arriving at the university, they would drop us on the entrance not even entering the campus building at all. This has came to such a degree that we would have to walk 3-4 bus stops up just to find a bus that had some space to ride in, not even to sit just to be able to go to our University. This became really troublesome during finals week because they would reduce the amount of active busses on the line because of less students using them and because our finals were spread out to two weeks.

Coming back to my point, I do not have a single drop of empathy for the people who have caused me and my student friends this much trouble. Since then I was able to transfer to a new University that is around 3 hours distance to my City. When I left there, there was a new Dominos building and a brand new Burger King opening there that would actually deliver the food and don’t treat you like you’re a scumbag. As time goes, I will probably hear from my friends there that the local businesses slowly get shut down, get replaced by these corporations and I deeply wish that when the self driving technology arrives it will have the same effect on the public transportation system where you had to buy three different bus cards just to get around because the villagers held the city hostage.

I hope karma will find everyone who has caused this. I have done all I could and I’ve reported to every place I could. The takeaway here is that if you are reading this don’t be an asshole. And to all the local businesses I’m sorry but my trust on you even if you are a legit business has went down significantly.

It’s simple, encoding. Another small internet search defines encoding as “The process of converting data from one format to another, typically for storage, transmission, or compression.” As media gets encoded more and more, it reduces in quality, allowing you to count the pixels with much more accuracy.

We have a lot of media formats, from the ones we’ve heard like .mp4, .mkv, .mov to the ones you’ve probably not encountered before like ProRes and DNxHR. When we talk about a format, such as .mp4 or .mkv we are actually not talking about a video, or a audio track, we are talking about (often) the both. This is because when we have video, we very often have audio next to it.

Formats vs. Containers

Remember .mp3? It is one of the very popular and widely supported music formats for storing audio, it is also an encoding format. .mp4 on the other hand, is not a encoding format. How come? It’s because when we store video we have to store both audio and video in a single file. An .mp3 file can only store a single audio track, so when we invented .mp4 we did not invent it as a format but as a container. The container takes in a video stream and an audio stream, this can even be .mp3, and gives it to the video player on demand.

Then what do we encode the video in? We surely need something like .mp3 for video too, because .mp4 is not a encoding format, how will we encode the video with? The answer is often H.264.

H.264 is currently one of the most popular video encoding formats out there. It takes in a raw video stream, puts keyframes in and calculates the differences between the key frames mathematically. This means instead of storing 30 image files and playing them really fast, in the process wasting space by duplicating the existing details up to 30 times, we can just keep the difference between the keyframes and save space.

Why Do We Encode Anyway?

Why do we even encode things? It’s to save space. When we record something like a professional movie, we often use special encoders like ProRes. We sometimes don’t use encoders at all, and save things as RAW! This is so our camera doesn’t have to think about calculating the differences between keyframes and can just dump whatever it sees to its storage. One downside of this is the filesizes are huge! While you can get by with around 3-4gb of storage for an hour of video in H.264, the same video file can be sized upwards of 100-150gb in RAW. That’s almost the size of a game!

Another downside of encoding is that when you want to edit something, it becomes slower. Imagine you have to skip around your 1 hour recording constantly to edit things, and every time you land in a frame that is not a keyframe, your computer has to calculate the difference up to the frame you are in. If you use ProRes for example, since it saves all frames as they are, you don’t need to do any kind of computing at all, your only limitation is a fast disk drive.

This is why we categorize our encoders, stuff such as ProRes, DNxHR are considered editing codecs. H.264, x265, AV1 and a few others are considered finalizing codecs, meaning if you encode your content in one of them you want the best compression ever and don’t ever plan to edit it once again.

The Finalizing Problems

But what if we need to edit something we encoded in a finalizing encoder such as AV1? Most of our video editors, if supporting AV1, usually won’t complain. Your editing speeds will be very slow but you can always turn the AV1 back to ProRes. But when you turn the ProRes back into AV1 something magical happens, the algorithm for AV1 is designed to be a finalizing encoder, this means it will try and give you the smallest filesize possible, even if it means getting rid of minor differences between some frames.

If you keep encoding a video on one of the finalizing encoders such as AV1, you will end up losing some minor details each pass, but the minor details slowly creep up and become major, noticeable quality loss. This is the same for those memes you see online, each upload to a social media app encodes the video, then someone else downloads it, uploads it again and it gets encoded again and again.

There exists lossless encoders, such as FLAC and some editing encoders (most editing codecs do still compress, but so little it’s indistinguishable by human eyes even if you encode between them multiple times), because they are not trying to compress media, they are just pre-calculating (literally marking all frames as key frames) the frames between the keyframes so your editing application can easily seek between frames without having to calculate the frame you just landed on from a keyframe 30 frames ago.

But what if we want to work on a finalizing encoder encoded video file without losing quality? Depending on the task you want to do, it is very much possible to edit a video without encoding it once again, saving you time spent on encoding and not losing any quality at all.

How to Edit Without Encoding

Basic video operations, such as cutting a part out, merging two parts can be done through cutting between keyframes. Since a video player only requires some keyframes, some inter-frames (this is what we call the mathematically calculated difference, as they are not real frames) and some metadata, if we just cut between two keyframes, update some metadata and put it inside of a container, the video player will happily play it for us.

The same is true for switching between containers. Want to switch from .mkv to .mp4? You can take your video stream, audio stream and put it inside of a .mp4 with no encoding at all! It will have no difference than the .mkv as long as we were able to fit the contents in the .mp4 container (it has some weird limitations like not being able to take in OPUS audio, why?????)

This means you can go between .mkv and .mp4 all you want with no quality loss at all. The container of the video does not matter.

There are some tools out there for lossless container switching and video editing, I often use Avidemux, lossless-cut is another great alternative. You can also always use some ffmpeg commands. Just put your video in, skip between keyframes, put a start and end to your selection, then on your options, select both audio and video as copy (copy means no encode will happen, the editor will directly copy the frame data out of the video stream), select your preferred container (as we have talked, it does not matter at all as long as your players are happy and your container can accept your encoded stream)

You can just put two clips of a video next to each other, using Avidemux, just open a file, then open a second file but instead of using the open prompt, use the append option. If you need to put a stream to another container that does not support your audio or video encoder, like a popular case with .mp4 is it doesn’t support OPUS audio from .mkv containers, you can switch audio from copy to AAC or whatever you like, and only encode the audio, saving you time and quality on your video stream as it doesn’t need to get encoded just because of your audio.

Ever since the announcement of Discord’s age verification, people have been looking for alternatives. I’ve tried out a few and in this post tell you my experience of them.

Matrix

Matrix is one of the first replacements that comes to mind. It is not a platform by itself but more so a protocol for federated servers to communicate with eachother (though it uses JSON so far).

Matrix has some popular clients, including Element, nheko, Thunderbird (yes, the mail client!), Neochat, Cinny, Fluffychat and so on. I’ve personally tried all of the stated ones on desktop, and the one I’ll keep using seems to be nheko. Element is too clunky, it works as an Electron application so it just sucks system resources even if it’s doing nothing. nheko and Thunderbird doesn’t use Electron, but Thunderbird lacks quite a bit in terms of features, same with nheko but the only major lack of feature is just voice calls. Cinny and Fluffychat were both very clunky on my browser so they have been eliminated from my choices.

The situation with so many Matrix clients all feature lacking is interesting, when you and your friends use separate clients it can cause issues such as Element and Fluffychat users can create polls where nheko nor Thunderbird can see them, by default Thunderbird can’t even show image previews. Both of them also lack the call feature. They can be made up for by using Jitsi or Mumble, or just using the default Element client for calls.

In terms of homeservers, there are some popular homeservers out there. The most popular one is matrix.org but you can usually communicate with other servers or even create your own. Element allows you to rent servers from them, though it is also possible to host it yourself as long as you own a domain name. You can even disable external server access if you don’t want to be a part of the federated network. All of this is able to run under a docker container, which makes deployments as easy as copy pasting a docker compose file.

IRC

IRC has existed for a long time, even though most people have forgotten about it, it always had a persistent userbase on multiple servers, even so much for the Matrix server hosts implementing IRC bridges to combine the two communities together (although it had it’s own drama but we’ll not get into that)

I have been using IRC for some time too, I’ve been using Adiirc and now Thunderbird to connect to IRC servers for a long time. One of the biggest issues people encounter when using IRC for the first time is the lack of message history. Unlike Matrix or Discord, IRC keeps no message history, when you join a server you only get messages as long as you are connected, once you disconnect you are in the blind. A solution to this was to run IRC bouncers on 7/24 machines, that would record all messages and you could configure your client to fetch the messages whenever you wanted, but this required some not so easy configuration and client support.

With the “release” of ircv3, there is a new feature made to fix this (if you consider it to be an issue or not is up to you).

The IRC servers are able to store messages and send them to the clients whenever they want it. This even works for DM messages (although usually requires you to ident so you can’t spoof your nick) and as long as you have a compatible client and server, it seems to work flawlessly. I’ve been using the client Halloy in a server hosted by a friend (using Ergochat as server) for a couple of days, and if you are willing to let go of some conveniences, i think it is definitely worth a shot. It also seems to be easier to configure than Matrix, but don’t take my word on that.

The IRC servers are able to store messages and send them to the clients whenever they want it. This even works for DM messages (although usually requires you to ident so you can’t spoof your nick) and as long as you have a compatible client and server, it seems to work flawlessly. I’ve been using the client Halloy in a server hosted by a friend (using Ergochat as server) for a couple of days, and if you are willing to let go of some conveniences, i think it is definitely worth a shot. It also seems to be easier to configure than Matrix, but don’t take my word on that, take theirs:

1msg: it's been very fun to configure  
2msg: I ran the thing, it came with message history held in ram for as long as a week as the default  
3msg: which is plenty for a lot of configs but I'm used to a discord like setup  
4msg: so I set up the permanent store option which requires hooking up mariadb, which is fine, I had an install of that lying around anyhow  
5msg: another part was getting a TLS certificate, which you can do by letting a web server with let's encrypt certbot do it for you  
6msg: other than that, the config is just one yaml file, everything is explained in the comments, it's so simple and great. doesn't feel that much more complicated than setting up a discord server but way more powerful
7msg: also, you can mention the ObsidianIRC client which is a (...) Discord clone using the cutting edge ircv3 features

Well, you heard them. I just tried out ObsidianIRC in the process of writing this edit, and I have to say it does a lot more to be like Discord that Revoult does (subtle foreshadowing). While it is not just a drop-in replacement, it may help lower the learning curve ever so slightly if you’re looking to move a friend group or similar over to IRC. It has features such as being able to delete messages, set profile pictures, see who is currently typing (which are all revolutionary for IRC). The only issue is that it runs on another port than your regular IRC server does. Do with that as you wish.

Revoult

It sucks. It is trying to be a Discord clone but their mobile client is literally a webview window with their website open. The features are sketchy, works half the time. Would not recommend.

Writer: Sukonbu

Why is It Happening?

When how common and widespread social media has become is put under investigation, it is no one’s surprise that it has pros and cons. Recently, Discord, one of the biggest social media platforms amongst teens, pushed a new update to fight one of the biggest concerns of everyone. With this update, as long as the user doesn’t confirm their identity with their real ID cards, they will not be able to access certain features of Discord and will not be able to join servers that are for adults; this doesn’t necessarily mean NSFW servers.

So, if this new update looks this good on paper, why is there this much outrage against it?

Here’s actually what happens:

As much as Discord claims that this update is for protecting the children on the application, there is more to it. Being able to use the internet anonymously of course has its ups and downs, but in the end, this was how it was meant to be used. What Discord actually wants to do is remove this anonymity and have millions of people’s data in their hands, giving them the choice to do whatever they want. The best example of this would be the agreement between ICE, the DOJ, and Discord. ICE and the DOJ asked Discord for the identities of people who criticize them, basically taking away their so-called ‘free speech’.

How did users receive this update?

To put it plainly, terribly. For a platform that promised anonymity as one of its biggest features, taking it away was a big gamble. Said gamble didn’t pay off. From users to social media content creators, everybody raised their concerns. Nobody wanted this update to go through from the moment it was announced. But in the end, with the update being out, there was nothing users could do other than continuing to raise their voices and going as far as finding a new platform like Discord that still promised anonymity.

What is going to be the aftermath of this?

To wrap it up, it is not looking good for users anymore. Being anonymous on social media was a curse and a blessing, but without it, there would be nothing but pain. Discord having this much data in their hands is a way bigger problem than the anonymity they are trying to destroy. If this continues to go on like this, it is not looking good for both sides. On the one hand, users are being forced to leave the platform they used for years and leave all the memories they made behind. On the other hand, this move was not worth its cost; Discord will continue to slowly bleed out and lose users as long as they keep up with this verification system.

Continued in next post.

Writer: merdagi

Whether I Like it or not, LLMs are a part of my life now. Their ability to ingest huge amounts of text and process it down into smaller pieces is the fundamental reason we let them browse the internet and summarize content from ten different websites into a single sentence. We even use deep research capabilities to let them wander through tons of papers and sources, then creating a condensed (and hopefully not half-assumed, half-hallucinated) text output.

This is the reason why I keep some local models around for offline use. To me, these models feel like a copy of an encyclopedia that I can utilize when needed. In addition, they are completely offline, so I don’t have to wonder what Google does during the 72 hours they keep my chat log with Gemini (even though I’ve turned off all data collection and activity history).

Of course, this also made me think: would it be possible to remove the censorship of local models to the point where they can just reply to everything? This led me to the discovery of the world of “Abliterated” and uncensored models.

It wasn’t just me wondering if this was possible. People have been actively trying to break the censorship of both online and offline models, often utilizing “jailbreaks” that switch between long, confusing text walls, writing in emojis, hijacking system prompts, and so on. However, when you host a model on your local machine, the possibilities become limitless.

The first models I used relied on an analysis method (often called abliteration). The model would be given prompts where it refused to reply, and other prompts where it replied as usual. Researchers would detect the neurons triggered between these states and modify their weights, which in turn disabled the model’s ability to refuse. However, these models often sucked.

When you take a model and play with its neurons, the smaller the model and the more neurons you adjust, the dumber the model becomes. This applies not just to censored prompts, but overall reasoning. I’ve seen a lot of people complain about this, which led me to discover a second method of decensoring and debiasing.

In this method, the model is simply fine-tuned with training data that includes replies to typically censored content. The model is taught to reply to everything, essentially becoming uncensored and debiased through learning rather than surgery. This is a softer approach than playing with the neurons directly and produces much better results.

As a result, if you are looking for a local assistant that actually listens to you without lecturing and without losing its intelligence in the process, stick to the fine-tuned variants from people such as soob3123 or, for quantized versions, mradermacher.

With these fine-tuned, uncensored, and debiased models, we finally get the best of both worlds: a highly capable, creative assistant that resides entirely on our own hardware. It keeps our discussions away from big companies, answers our questions without judgment, and analyzes content without bias.

How healthy it is to share secrets or have deep discussions with an AI model is a point of argument in itself. However, since the results of mainstream models are directly tied to the profits of the companies behind them, it doesn’t seem likely that we will get naturally uncensored or debiased LLMs anytime soon… (except maybe Grok).

This is just my personal opinion. I am not trying to convince anyone, and I am not paid by any of these companies.

TL;DR: I prefer Gemini over GPT, but mostly think the AI hype is getting out of hand.

Everyone is talking about AI right now. OpenAI and their CEO are constantly in the news, and other companies like Google and Meta, even newer companies such as Mistral and Anthropic (at least in the AI scene) are trying to catch up. But because ChatGPT is the most popular, it has the most users and news about it whenever they do something.

My main issue with ChatGPT is that it tries too hard to please the user. It will agree with whatever you say, even if you are wrong. If you give it false info, it often just accepts it. When I used it in the early days, it hallucinated constantly, it knew how things worked in theory but failed completely in practice.

I also hated how OpenAI acted. They called themselves a “nonprofit” and “open” but refused to share their research and tried to monetize everything immediately after GPT-2.

I switched to Gemini because the free version of ChatGPT was getting extremely annoying. I realized that Gemini felt a bit different. Unlike ChatGPT, which acts like a “yes-man,” Gemini actually argues with me at times. If I am wrong, it insists on its knowledge. I started using it for basic scripts and regex filters, and it felt more like a tool and less like a toy that could break at any time.

The newer Gemini models have improved a lot. It is fast, and unlike GPT, it stands by its values. If GPT sees a forbidden word, it gives a generic error. Gemini clearly refuses (compared to a hard generation stop) and explains why. It can be stubborn, but I prefer that over a model that lies to make me happy.

These days, I mostly use Gemini to help me understand how machine learning actually works on my local machine. I can show it error logs or data charts from my coding experiments, and it does a surprisingly good job of analyzing them. It helps me figure out if my code is broken or if the data is bad. However this does not mean it is perfect, saying a model is better than the other is almost as same as preferring a giant douche over a turd sandwich, it is meaningless.

I am still very skeptical about the industry. Big companies act like AI is magic that will replace humans and make infinite money. They are slapping the “AI” label on everything just to please investors. I hope these investors will eventually realize that replacing entire companies with chat bots is nothing more a fever dream.

AI is not magic. It has major flaws. As the internet fills up with AI-generated slop, these companies will struggle to find good data to train on. They will end up feeding their models their own garbage, like a snake eating its own tail.

It’s late at night, I’ve been going back and forth around a specific RPG Maker VX Ace game whose developer has completely disappeared off the face of internet. The game has a lot of bugs, works half of the time and due to the constraints of the game engine, looks like absolute garbage at a small resolution on my high DPI screen.

I have worked with RPG Maker, but my experience was majorly in MV and recently MZ, i had no idea of how VX Ace worked, but since it’s still RPG Maker in the end, i wanted to find a way to somehow extract the assets.

The game has a weird directory structure.

Let’s explore it a bit,

Under Audio, Fonts, Graphics and System, we have all the regular RPG Maker assets in their unencrypted formats. These files are created by default on every RPG Maker project, albeit a bit differently between versions (we will get to that).

But this is useless for us, we want the changes that are made, not just the default assets. For that we have the Game.rgss3a file located right next to the executable.

When a RPG Maker game is about to be “Deployed” (I will call this Compiling even though it isn’t) the developer is able to place a password in order to encrypt their custom game data such as Common Events, Images, Tilesets, Scripts etc.

If the developer does not encrypt their games, the assets would be in their usual spots and we wouldn’t have a .rgss3a file, since this file is specifically for encrypted games. We need a way to get into this.

In my initial research, it’s quite easy to get into this file and extract the assets. I used the first website I’ve found which is called Game Resources Viewer (not endorsed)

I’ve put our encrypted file in there, and wouldn’t you know it, we got the data unencrypted.

At this point, I forgot some crucial information about how RPG Maker projects work, so I had to make it harder for myself. I needed a .rjproj file which I could open with my RPG Maker MV engine and hope that the engine would convert the old version to the new one by itself. I did some research online to come across a tool called RPGMakerDecrypter with a supposed feature to “…give a best guess in creating the original project…” which got me excited. I quickly downloaded the latest release and got to work.

I didn’t own RPG Maker VX Ace and I was not paying that much for an engine this old, so I downloaded a trial version from the internet, I was scared that the trial version would have some limitations, but other than the lack of a deployment option, I never encountered a single trial limitation.

Due to me doing this out of my own regard without permission from the game’s developer, while it’s legal for me to reverse engineer games myself, i don’t want to share images from the game just in case.

The game opened and ran quite fine, it had some errors occurring but I had to merge the unencrypted assets with the encrypted ones to get rid of them. Afterwards I used a script to turn the game data into RPG Maker MV format and had to deal with a bunch of problems there. As long as your VX Ace game doesn’t have custom scripts (RPG Maker VX Ace uses Ruby, meanwhile newer engines MV and MZ use Javascript, unlike scripts on older engines these rely on a Plugin system), it’s fairly easy to convert games to modern engines, although you do have to upscale most of the graphics by 150% (sometimes 133%) and convert audio into .ogg, and give up on using animations from VX Ace because while older engines used multiple images to create animations, the newer ones seem to use a special format that looks a lot like video.

Now I was finished with this game, i wanted to turn another game into a project as well. However there was a problem…

In order for my decryptor to work, it needed a .rgss2a file, and it would not accept anything else…

Then I had a realization, how does a freshly created RPG Maker project look like?

This looks awfully like a deployed game, so let’s see what is inside of the .rvproj file…

Turns out, the rvproj file is just a simple text document that contains the engine (and maybe corescript) information. Copying this file into a non-encrypted deployed game simply turns it into a project file in which you can edit the database however you like. The project information such as the game’s name is also stored in the database unlike my assumption of being stored in the project file. This means instead of having to download and unencrypt with this application, I could have very well created a new RPG Maker project, then copied the extracted data into my new project and all would work well. The more you know…

TL;DR: my PC broke down and I was refused a replacement; I’ve just got it back and it’s still broken, so it will be sent again for repairs.

The long explanation is that when I sent my PC for repairs after a fan issue, I was refused service because my warranty was over. I’ve checked this, and my warranty should still be on for about 6 more months. After going back and forth with the OEM, we found out that the computer I was using was actually a refurbished product sold to me without my knowledge, and it had 4 service records before it finally reached my hands. The issue is that neither the selling company nor the third-party website I bought this from acknowledges this. The third-party website tells me to solve it between the seller company and the seller company outright refuses replacement and told me they could only do repairs on it. After waiting for around three months, I’ve finally got the supposedly repaired computer with the aluminum fan air holes visibly distorted, causing noise when pressed on, and the fan issues not actually being fixed.

This is quite the predicament, so we’ll see how this goes. I’m still around, but since I don’t have another computer, the blog may see fewer updates.

Thank you for your support in this time.

Emily Young was a former writer and host at Linus Media Group (LMG), the parent company of many well-known channels such as Linus Tech Tips, TechLinked and ShortCircuit. Many of the fans known her from his appearence and knowledge on Linux related content.

On May 27 2023, Emily Young had announced she was coming out as transgender with a video on her personal Youtube channel. Following this announcement though, her appearence on videos have dropped significantly and they have been continuing their work as a writer.

Linus Media Group has been called out in the past for mistreatment and poor work conditions. One female ex-employee of LMG had alleged that she has been inappropriately grabbed multiple times by employees in the office. She also stated that she was barred from appearing on videos due to her gender. These allegations caused LMG to halt production and hire a private investigator. LMG has announced that following a third-party investigation conducted by Vancouver-based law firm Roper Greyell, that the claims of alleged workplace bullying and sexual harassment “were not substantiated”.

From the contents of the video, and the comment left on it by Linus Media Group’s founder Linus Sebastian, this appears to be a choice made by Emily Young, and is unrelated to the drama that have been surrounding LMG.

In the video, Emily Young talks about multiple topics including future plans of the channel, monetization startegy and talks about how she is not an expert, just a curious person on topics others had assumed she was.

Refferences:

https://www.dexerto.com/tech/ex-linus-tech-tips-employee-alleges-mistreatment-and-poor-conditions-2251613/

In this video, he talked about how he received a mail sent by a Counter Strike gambling website trying to bribe them with 20.000$ into uncovering a rival gambling website. Coffezilla, in return, refused the bribe and decided to make a video uncovering the suspicious and illegal practices of these gambling sites.

Counter Strike 2 is a game released by Valve Corporation on their digital distribution platform, Steam. The game features skin crates in which players can spend money to unlock them and with a low chance, get gun skins that are worth more than the crate. In addition, players are allowed to use the Steam Marketplace to trade or buy and sell these skins.

In this three part video series, Coffeezilla goes into how these gambling websites try and take advantage of minors, allowing them to get used to gambling at a young age. In addition, he covers how Valve Corporation takes advantage of this fact and doesn’t take the required action against these sites, allowing them to gain millions of dollars via Steam Marketplace where players are able to buy and sell in-game items for Steam Wallet Funds (while paying the platform owner Valve Corporation around 15% of the item’s sale price), which cannot be withdrawn from their account but can be used to buy physical items such as Valve’s new gaming console, Steam Deck.

In addition, Coffezilla covers the technique used by Valve to allow these crates to be opened by players in France, where a law is in place preventing these kinds of luck games.

Until recently, Steam’s User Agreement had a forced arbitration, resuting in players who agreed to this agreement not being able to sue the company over these practices. One notable example of a lawsuit attempt was in 2022 where a parent had sued Valve Corporation over Counter Strike skin gambling but the case eventually got dismissed. On September 27, 2024 Valve Corporation removed the forced arbitration clause from their user agreement, opening the doors to future lawsuits.

Deepseek-R1 is a brand new model that is able to reason, meaning instead of doing a task like a regular model, it first think of what to do and then takes action.

The only other model that was capable of achieving this with reasonable accuracy was OpenAI’s gpt4-o model, which, unlike Deepseek-R1, is closed-source.

After the model was released, Deepseek released a substential amount of papers about the model, explaining how they were able to achieve this success with their model. OpenAI however, claims that Deepseek used output from their gpt4-o model in order to create their cold start data. At the time of writing this there hasn’t been an official announcement about these claims yet but Deepseek claims that the cold start data they used came from their old model Deepseek V2

Deepseek has released multiple lower parameter versions of their model including 1.5B, 7B and 8B parameter models. This allows users who don’t have dedicated hardware or memory to run bigger models such as Meta’s Llama to run Deepseek-R1 with great performance albeit at a reduced accuracy.

For people who would like to test this model out, Deepseek has a public instance available. In addition it is possible to run the model locally using Ollama.