PewDiePie Googles Himself: Retirement, His Custom AI Model Ajax, and Linux Life

PewDiePie recently posted a video of himself "Googling himself," which on the surface is a lighthearted, self-deprecating Q&A but actually reveals quite a bit of first-hand information about his current technical life: from the training progress of his custom AI model Ajax, to the Linux distribution he uses daily, to his weariness with the modern YouTube content ecosystem.

The Real State Behind the Self-Search: Retirement Is Fake, Exhaustion Is Real

At the start of the video, PewDiePie directly addresses the search suggestion "Is PewDiePie retired?" His answer is candid: "Yes and no." He admits that he is indeed doing YouTube in a more relaxed way—only posting when he feels like it, or uploading because of contracts and promotional obligations.

He attributes this to being tired of "trying too hard." PewDiePie says he has spent 15 years addicted to YouTube, and another 15 years "would be too sad and too stupid." He even uses the word "sweaty" to describe the current YouTube ecosystem, complaining that the platform A/B tests even opening lines and forces creators to shoot two versions of an intro. This distaste for platform mechanics is the real context behind his repeated mentions of "retirement."

Custom AI Model Ajax: 60 tokens/s, BF16 Version, But Training Isn't Done

The most informative part of this video is his update on his AI project. PewDiePie says he uses Odysseus every day and reveals that his custom model Ajax can already hit 60 tokens/s, and is currently a BF16 version, so there is room for further speed improvements.

He specifically emphasizes that he has achieved a "breakthrough," but because the model is still training, he doesn't plan to elaborate for now and "won't release it before it's done." This cautious attitude contrasts with the casualness of his past videos, showing that he has invested a considerable degree of seriousness into this project.

On the installation experience, he admits that Odysseus is currently "too hard to install," and the team is working on the problem with the goal of making the next release "easy to install and ready out of the box." This is actually the most common hurdle for custom AI tools: model capability is one thing, distribution and the barrier to getting started are another.

token and tokens/s

In the context of language models, a token is the basic unit of text that the model processes, usually corresponding to a word or part of a word. tokens/s (tokens per second) is the core metric for measuring a model's generation speed: it indicates how many tokens the model can output per second on average. For locally running models, this number is affected by model size, quantization precision, hardware compute, and memory bandwidth. 60 tokens/s is roughly several times the human reading speed, which is fairly smooth for interactive conversation, but still lags behind cloud-based large models that can reach hundreds of tokens/s in throughput.

BF16 (BFloat16)

BF16 is a 16-bit floating-point format. It has the same bit width as the common FP16, but more exponent bits and fewer mantissa bits, so its dynamic range is close to 32-bit floating point. It is better suited for deep learning training and inference, and can significantly reduce VRAM usage and increase computation speed with almost no loss of precision. When PewDiePie mentions that Ajax is currently a BF16 version "so there is room for speed improvements," it usually means that INT8/INT4 and other quantization schemes may be adopted later to further increase speed, at the cost of a slight drop in precision.

OpenAI Bans and the Switch to DeepSeek: A Creator's AI Tool Choices

One easily overlooked but interesting detail in the video is PewDiePie mentioning that he "got banned by OpenAI three more times," and saying he has given up and switched to DeepSeek^Ad. He jokes, "I'm such a nice person," then quickly adds, "I shouldn't say that."

This content is a personal experience and cannot be taken as a general conclusion about platform policy, but it reflects a reality: for creators who need to frequently try different models and conduct local experiments, account-level stability and access restrictions directly affect tool choices. PewDiePie's response is straightforward—switch tools and keep doing his own thing.

Linux and Open Source: Arch for Daily Use, Ubuntu for Servers

When asked what Linux distribution he uses, PewDiePie's answer is crisp: "Arch all the way," but he uses Ubuntu on servers. He also takes the opportunity to question Windows: as AI makes software easier to rebuild and the open-source ecosystem continues to expand, he sees no reason to stay on Windows—"you can game on Linux, you can do anything."

This combination is actually quite typical: the desktop pursues the latest and controllability, while the server pursues stability and peace of mind. His view may not apply to everyone, but as a user who has long switched between content creation and custom tools, his choices have some reference value.

The Positioning Difference Between Arch Linux and Ubuntu

Arch Linux is a rolling-release distribution where packages are always up to date. The system is built by the user from a minimal base, offering extreme controllability, but users must handle compatibility issues caused by configuration and updates themselves. It suits users who are willing to tinker and pursue the latest software. Ubuntu, by contrast, is a fixed-release-cycle, out-of-the-box distribution. Long-term support versions provide years of security updates, and community and documentation resources are abundant, so it is widely adopted in server scenarios. The combination of Arch on the desktop and Ubuntu on the server is essentially a division of labor between "latest and controllable" and "stable and worry-free."

Religion, Language, and Self-Awareness: Another Side of the Q&A

Beyond technical topics, this self-search also covers some personal questions. Asked whether he is religious, PewDiePie says he has never been asked this question and answers "agnostic." He also mentions that he has always wanted to make a video comparing the Bible, the Pali Canon, and the Quran, and that he has already started reading the Bible and finds it "very interesting."

Another topic he proactively admits fault on is Japanese. He says bluntly that he has lived in Japan too long but is not fluent enough, feels embarrassed about it, and says he will spend more effort practicing going forward instead of continuing to pile time into "technical stuff." This kind of self-correction says more about his current priorities than any search result.

Conclusion: A Creator's Technical Turn

Judging from this "Google himself" video, there is an obvious gap between PewDiePie's public image and his real state: search suggestions care about whether he is retired or a billionaire, while he himself cares more about how far the model training has progressed, whether the installation experience can be simplified, and whether he should practice Japanese more.

For readers interested in AI tools and the creator ecosystem, the value of this video is not gossip but that it provides an observational sample: when a content creator starts training models himself, switching operating systems, and adjusting his toolchain, the problems he faces are not fundamentally different from those of ordinary developers—if training isn't done, you can't release it; if installation is too hard, no one will use it; if a tool isn't good, switch to another one.