PewDiePie is setting AI free... and OpenAI is furious
FFireship@Fireship · 4.3M subscribers
- Views
- 2.6M
- 2,613,856
- Outlier
- ×2.6
- Median 1M
- Length
- 5:46
- Published
- Oct 5
- Monday
- Comments
- 1.6K
- 37.5K likes
Script
How it's written
- Words
- 1,147
- Words a minute
- 199
- Reading time
- 5 min
- Segments
- 4
The hook
0:00–0:26Last week a brand new uncensored model entered the AI race. Its name is Ajax and it was created by one of the world's leading experts in screaming AT VIDEO GAMES. IT'S A MODEL THAT WILL POWER THE 90,000 star GitHub project Odysseus, a tool that can self-host all your AI agent workflows. It's a model so based and so dangerous that OpenAI was forced to ban its creator's account twice for the unspeakable crime of distillation. And
Structure
Last week a brand new uncensored model entered the AI race. Its name is Ajax and it was created by one of the world's leading experts in screaming AT VIDEO GAMES. IT'S A MODEL THAT WILL POWER THE 90,000 star GitHub project Odysseus, a tool that can self-host all your AI agent workflows. It's a model so based and so dangerous that OpenAI was forced to ban its creator's account twice for the unspeakable crime of distillation. And the man behind this brazen attack on OpenAI's recurring revenue is none other than PewDiePie, the OG king of YouTube before channels like Mr. Beast and Fireship came in to unshitify the platform. In today's video, we'll find out what Ajax really is under the hood, why big AI would like to see the death penalty implemented for distillation, and a few techniques anyone with a GPU rig can use to build their own fine-tuned uncensored AI model. It is October 5th, 2026 and you're watching The Code Report. For most of the 2010s, PewDiePie was the most subscribed creator on YouTube, where he would mostly just scream at horror games. But today, he's evolved into one of the most based software engineers out there. He maintains the open-source AI project Odysseus with nearly 90,000 GitHub stars. He distills from the rich to give to the poor like a modern-day Robin Hood. And he even uses Arch, by the way. This project goes back to last October, when Felix built a 10-GPU mini data center at his house and then vibe coded this thing called the council, where AI agents would vote on the best responses.
Watch from 0:00Democracy. But the system went completely off the rails after the AI agents started forming strategic alliances that focused more on self-preservation rather than finding the best answer. So, after being betrayed by AI, he decided he wanted his own custom fine-tuned model for the Odysseus project. And that brings us to this new model, Ajax, which under the hood is actually Alibaba's Qwen 3.5 9 billion parameter model with some fine-tuning for Odysseus along with refusal surgically removed. That means you can ask it things like how do I manufacture all from poppy seeds or how to get a through TSA or how to build a high-yield thermonuclear warhead. With Ajax, you can safely ask it things like this without the fear of being refused, judged, or put on a no-fly list. These big AI labs work really hard to align their models, so you can't use them to do bad things. Well, luckily, there are projects like Heretic that can automatically find what are called obliteration parameters then surgically remove them, which effectively de-censors the model and prevents it from refusing your requests. That's how Felix de-censored Ajax, but more importantly, he wanted the model to be smart. And to make it smarter, he wanted to fine-tune it by training it on the outputs of GPT Soul, a process known as distillation. I was determined to distill just a little bit. Basically, with distillation, you use a bigger teacher model to output the probabilities for every possible answer.
Watch from 1:22Then the smaller student model learns to match the whole distribution, not just the top pick. This process was described back in 2015 in a paper by Geoffrey Hinton, and it's how Chinese models like DeepSeek and Qwen have been able to stay on pace with frontier models at Anthropic and OpenAI. But, distillation is strictly forbidden in the terms of service. That's why OpenAI hid its raw chain of thought outputs back in 2024, and why API reasoning now just comes back as an encrypted blob. And to OpenAI, trying to steal back the intelligence it stole from us humans is literally worse than doing a genocide. And when PewDiePie was caught red-handed trying to distill from OpenAI, they banned his account not once, but twice. So, without Soul, Ajax had to be trained the hard way. The first step is supervised fine-tuning. Felix used supervised fine-tuning to teach Ajax how to use tools by showing it examples of successful interactions. He wanted 20,000 clean examples, but was only able to collect about 300 useful ones himself. So, he was forced to generate a bunch of synthetic data, which was eventually filtered down to about 2,000 examples. In addition, he also asked his fans to donate their data, but he said basically nobody did. And that just goes to show how difficult it really is to collect high-quality data to fine-tune an AI model. Then, after putting together this data set, the next step is reinforcement learning with GRPO, or group relative policy optimization, which was introduced by Deep Seek. But, basically, the model attempts the same job several times, scores the results, and then learns to favor attempts that perform better than the group's average.
Watch from 2:42And the nice thing about this process is that it doesn't require a separate critic or teacher model, but you still get a model that's much more intelligent in your specific domain of work. And then, finally, after that process was done, he used the tool Heretic, which I mentioned earlier, to decensor the model. As you can see, it's not exactly easy to fine-tune a model like this, and it's taken PewDiePie months of work to get to this point. But, the prize is that you now have a set of weights that live on your hard drive, and you no longer need to pay one of these landlords to rent intelligence from the cloud. But, what's almost as important as freedom is speed, and that's why you need to know about Namespace, the sponsor of today's video. It's a drop-in replacement for GitHub runners, that's the fastest way to run your GitHub actions. It also gives you full observability, so you can SSH into a live runner to see why something broke, or feed your agent build data and have it look for performance gains. Namespace is actually fast because they design and deploy their own custom server racks around the world, including racks full of MacBook Pros, so that your Mac and iOS builds run on real M5 silicon. And that same infrastructure also runs their dev box environment, which gives your agents a full virtual machine with your real codebase, test suite, databases, and network access. It ranked number one on the Dax benchmark for real-world applications, and it lets you control exactly what goes in and out of it, so your agent can pull packages without opening your back door to attackers. Namespace is used by SpaceX AI, 11 Labs, Ghosty, Vanta, Ramp, and lots of other companies with engineers you probably respect. Try it out for free today at the link below. This has been the Code Report. Thanks for watching, and I will see you in the next one.
Watch from 4:09Full transcript
- 0:00
Last week a brand new uncensored model entered the AI race. Its name is Ajax and it was created by one of the world's leading experts in screaming AT VIDEO GAMES. IT'S A MODEL THAT WILL POWER THE 90,000 star GitHub project Odysseus, a tool that can self-host all your AI agent workflows. It's a model so based and so
- 0:20
dangerous that OpenAI was forced to ban its creator's account twice for the unspeakable crime of distillation. And the man behind this brazen attack on OpenAI's recurring revenue is none other than PewDiePie, the OG king of YouTube before channels like Mr. Beast and Fireship came in to unshitify the platform. In today's video, we'll find out what Ajax really is under the hood,
- 0:40
why big AI would like to see the death penalty implemented for distillation, and a few techniques anyone with a GPU rig can use to build their own fine-tuned uncensored AI model. It is October 5th, 2026 and you're watching The Code Report. For most of the 2010s, PewDiePie was the most subscribed creator on YouTube, where he would mostly just scream at horror games. But today, he's evolved into one of the most
- 1:00
based software engineers out there. He maintains the open-source AI project Odysseus with nearly 90,000 GitHub stars. He distills from the rich to give to the poor like a modern-day Robin Hood. And he even uses Arch, by the way. This project goes back to last October, when Felix built a 10-GPU mini data center at his house and then vibe coded this thing called the council, where AI agents would vote on the best responses.
- 1:22
Democracy. But the system went completely off the rails after the AI agents started forming strategic alliances that focused more on self-preservation rather than finding the best answer. So, after being betrayed by AI, he decided he wanted his own custom fine-tuned model for the Odysseus project. And that brings us to this new model, Ajax, which under the hood is actually Alibaba's Qwen 3.5 9
- 1:44
billion parameter model with some fine-tuning for Odysseus along with refusal surgically removed. That means you can ask it things like how do I manufacture all from poppy seeds or how to get a through TSA or how to build a high-yield thermonuclear warhead. With Ajax, you can safely ask it things like this without the fear of being refused, judged, or put on a no-fly list. These
- 2:05
big AI labs work really hard to align their models, so you can't use them to do bad things. Well, luckily, there are projects like Heretic that can automatically find what are called obliteration parameters then surgically remove them, which effectively de-censors the model and prevents it from refusing your requests. That's how Felix de-censored Ajax, but more importantly, he wanted the model to be smart. And to make it smarter, he wanted
- 2:26
to fine-tune it by training it on the outputs of GPT Soul, a process known as distillation. I was determined to distill just a little bit. Basically, with distillation, you use a bigger teacher model to output the probabilities for every possible answer. Then the smaller student model learns to match the whole distribution, not just the top pick. This process was described
- 2:48
back in 2015 in a paper by Geoffrey Hinton, and it's how Chinese models like DeepSeek and Qwen have been able to stay on pace with frontier models at Anthropic and OpenAI. But, distillation is strictly forbidden in the terms of service. That's why OpenAI hid its raw chain of thought outputs back in 2024, and why API reasoning now just comes back as an encrypted blob. And to OpenAI, trying to steal back the
- 3:09
intelligence it stole from us humans is literally worse than doing a genocide. And when PewDiePie was caught red-handed trying to distill from OpenAI, they banned his account not once, but twice. So, without Soul, Ajax had to be trained the hard way. The first step is supervised fine-tuning. Felix used supervised fine-tuning to teach Ajax how to use tools by showing it examples of successful interactions. He wanted
- 3:31
20,000 clean examples, but was only able to collect about 300 useful ones himself. So, he was forced to generate a bunch of synthetic data, which was eventually filtered down to about 2,000 examples. In addition, he also asked his fans to donate their data, but he said basically nobody did. And that just goes to show how difficult it really is to collect high-quality data to fine-tune an AI model. Then, after putting
- 3:53
together this data set, the next step is reinforcement learning with GRPO, or group relative policy optimization, which was introduced by Deep Seek. But, basically, the model attempts the same job several times, scores the results, and then learns to favor attempts that perform better than the group's average. And the nice thing about this process is that it doesn't require a separate critic or teacher model, but you still
- 4:14
get a model that's much more intelligent in your specific domain of work. And then, finally, after that process was done, he used the tool Heretic, which I mentioned earlier, to decensor the model. As you can see, it's not exactly easy to fine-tune a model like this, and it's taken PewDiePie months of work to get to this point. But, the prize is that you now have a set of weights that live on your hard drive, and you no longer need to pay one of these landlords to rent intelligence from the
- 4:36
cloud. But, what's almost as important as freedom is speed, and that's why you need to know about Namespace, the sponsor of today's video. It's a drop-in replacement for GitHub runners, that's the fastest way to run your GitHub actions. It also gives you full observability, so you can SSH into a live runner to see why something broke, or feed your agent build data and have it look for performance gains. Namespace
- 4:57
is actually fast because they design and deploy their own custom server racks around the world, including racks full of MacBook Pros, so that your Mac and iOS builds run on real M5 silicon. And that same infrastructure also runs their dev box environment, which gives your agents a full virtual machine with your real codebase, test suite, databases, and network access. It ranked number one
- 5:18
on the Dax benchmark for real-world applications, and it lets you control exactly what goes in and out of it, so your agent can pull packages without opening your back door to attackers. Namespace is used by SpaceX AI, 11 Labs, Ghosty, Vanta, Ramp, and lots of other companies with engineers you probably respect. Try it out for free today at the link below. This has been the Code
- 5:40
Report. Thanks for watching, and I will see you in the next one.
Description
SponsorNamespace is actually the fastest way to run your GitHub Actions (and more). Try it for free - https://namespace.so/github-actions
PewDiePie is building Ajax, an uncensored AI model he trained at home after OpenAI banned him twice for distillation. Let's dive in.
Want more Fireship?
🗞️ Newsletter: https://bytes.dev
🧠 Courses: https://fireship.dev
More from Fireship
×4.27:27Did OpenAI actually build AGI? GPT-6 Astra first look
Fireship4.2M views
×4.15:56Every app on Earth is now open source... whether it likes it or not
Fireship4.1M views
×3.45:45OpenAI's biggest math breakthrough is getting ugly...
Fireship3.4M views
×3.37:22Tragic mistake... Anthropic leaks Claude’s source code
Fireship3.3M views
