No. 77
2026-10-04
the AI pulse — brewed, not raw
 AI CRITIQUE
Ekran görüntüsünde, damıtma (distillation) ile ilgili politika ihlalleri nedeniyle bir hesabın devre dışı bırakıldığını bildiren bir OpenAI e-posta bildirimi görülmekte olup arka planda bir erkeğin yüzü kısmen…

PewDiePie announces local AJAX model following OpenAI account suspension

Read the article    English (automatic translation)

Summary

PewDiePie announced Ajax, a 9-billion-parameter local AI assistant fine-tuned specifically for his self-hosted 'Odysseus' workspace based on Alibaba's Qwen3.5-9B model. The launch was framed as a response to claims that OpenAI suspended his account twice due to distillation. The model is not yet publicly available; no published weights, benchmarks, or verified hardware requirements exist.

Why it matters

The project highlights growing consumer demand for small, locally running AI agents that handle daily tasks without routing data through cloud providers. This trend could pressure the dominance of companies offering models-as-a-service. It also exposes a real point of friction: OpenAI's terms of service prohibit using their outputs to train competing models, creating a legal gray area for large-scale distillation attempts. For a content creator with PewDiePie's audience, this move pushes the tension far beyond the niche of AI enthusiasts.

Highlights

  • Ajax was built on Qwen3.5-9B (~9B parameters) and targets local, always-on agentic tasks such as browser, email, and calendar management.
  • Kjellberg used the open-source 'Heretic' tool for abliteration (removing rhetoric). He claims to have maintained safety guardrails around harm and dangerous instructions.
  • As of October 2, the model was still in the 'coming soon' phase: no public weights, independent benchmarks, specified hardware specs, or verified license were available.

Critical take

The narrative regarding the OpenAI ban relies almost entirely on Kjellberg's own statements and a single email screenshot. OpenAI has not publicly confirmed the details; therefore, the dramatic framing may be overtaking the evidence. More importantly, fine-tuning an existing open-source model with 9 billion parameters and using a unique 'abliteration' method is technically an incremental step. The lack of third-party evaluation indicates that the safety and capability claims are currently marketing rather than measurable facts.

Bu kritik, yerel yapay zeka modeli Qwen3.8-27B tarafından yazılmıştır. Haberi ben seçtim; kritik metnini yayımlanmadan önce tek tek denetlemiyorum.This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.
Back to the issue