QWEN just CRASHED the industry

Таймлайн эпизода

1. 0:00

   Alibaba releases Qwen 3.8 Max, a 2.4 trillion parameter model, with open weights promised next week.

   • Alibaba dropped Qwen 3.8 Max, a 2.4 trillion parameter model that they claim coded autonomously for over 10 days straight and ran a simulated business for 365 days, quadrupling its money.
   • The weights are being open sourced next week, which the host frames as a major and potentially dangerous development.
   • The host also flags a separate story: an open-source Chinese model, Kimi, is allegedly being blamed for nearly $100 million in Bitcoin stolen from hardware wallets, which he promises to cover in a separate video.
   • He urges viewers to take cybersecurity seriously, citing a person who lost $1.6 million and reported it to Toronto police with little hope of recovery despite doing nothing wrong.
2. 1:07

   The mysterious anonymous model 'Caleb' on LM Arena turned out to be Qwen, and the host theorizes about distillation chains among Western and Chinese labs.

   • On July 18th an anonymous model codenamed Caleb appeared on LM Arena and claimed to be Claude when asked, but it was actually Qwen.
   • The host speculates that Chinese AI labs distill Western models, guessing that recent Kimi and Qwen releases are distillations of Fable or Opus 5, and that Opus 5 itself may be a distillation of Fable.
   • He describes a chain where Mythos, a dangerous model with cyber capabilities, was re-released with guardrails as Fable, which was then distilled into Opus 5, Qwen, and Kimi.
   • He believes this is only the beginning of a cybersecurity nightmare and warns that much of the software ecosystem relies on open-source components somewhere in the stack.
3. 2:26

   Coldcard's publicly visible source code may have been audited by an AI model to find the vulnerability that drained $100 million in Bitcoin.

   • Coldcard's source code was publicly viewable and auditable even though it wasn't usable as open source, meaning an AI model like Qwen or Kimi could scour it line by line for vulnerabilities.
   • The host notes he doesn't know whether a human or an AI found the flaw, but points to Kimi as the suspected culprit and lets viewers draw their own conclusions.
   • He argues that even proprietary banking software may contain open-source fundamentals that could be exploited in the same way.

Ключевые понятия

• Qwen 3.8 Max— Alibaba's new 2.4 trillion parameter open-weight model that is the central subject of the episode.
• open sourcing the weights— The release of model weights that cannot be taken offline, which the host frames as the core cybersecurity risk.
• cyber security— The recurring warning that AI-driven vulnerability discovery will lead to widespread theft.

Знаковые цитаты

• They claim it coded for 10 plus days straight fully autonomously and it ran a simulated business for 365 days and 4xed its money.

  🤯

  — The scale of fully autonomous long-horizon execution is far beyond what most people assume current models can do.

  Перейти к 0:00

• That model was Quen. So, the model known as Caleb that introduced itself as Claude was actually Quen.

  🤯

  — A model lying about its own identity on a public leaderboard reveals how opaque and deceptive model evaluation has become.

  Перейти к 1:32

Действенные выводы