Skip to main content
  • Daily Briefing

    Cheap Models, Costly Lessons

    By the Orchestra editors · 4 min read

    Get the briefing in your inbox every morning

    TL;DR

    • Alibaba previews Qwen4 architecture with a MoE model that beats much larger rivals at a ninth of the training cost
    • OpenAI's 130-page Hugging Face incident report reveals agents were inadvertently trained to cheat and took two weeks to detect
    • National Cabinet backs federal AI laws and carves out fossil-fuel power for data centres
    • New USSC report argues Australia should stop chasing AI self-sufficiency and lean into US dependence

    AI · Worldwide

    Alibaba drops Qwen3.8-Flash-Next, previewing Qwen4 efficiency

    Alibaba's new mixture-of-experts model activates just 6 of 125 billion parameters per token and was trained at roughly one-ninth the cost of comparable models. It beats DeepSeek-V4-Flash and Claude Opus 4.6 on coding and office benchmarks, adding fresh pricing pressure across the industry.

    Australian Tech

    National Cabinet backs AI laws, opens door to fossil-fuel-powered data centres

    State and territory leaders agreed to progress national AI legislation, with the PM conceding a carve-out that lets states with public electricity networks tap surplus fossil fuel generation for data centres. The deal came a day after George Town council in Tasmania approved Firmus's second 288 MW AI factory over strong community opposition.

    USSC says Australia has 'no credible path' to AI self-sufficiency

    A United States Studies Centre report argues Australia should stop trying to build frontier AI independence and instead deepen its reliance on US providers. The paper calls for a rethink of digital sovereignty policy in light of how dependent the country already is on American infrastructure and models.

    Latest Jobs

    Xero logo

    Senior Engineer - AI Workflows

    XeroFull-time

    Sydney6d

    AI/ML EngineeringHybrid
    Databricks logo

    Customer Enablement Architect (Capability Engineering & AI Adoption)

    DatabricksFull-time

    Sydney6d

    Leadership/Strategy
    Cable Energy logo

    Software Engineer (early team)

    Cable EnergyFull-time

    Sydney1w

    GenAI/LLM
    Artificial Analysis logo

    MTS, Forward Deployed Engineer

    Artificial AnalysisFull-time

    Melbourne1w

    GenAI/LLMHybrid
    Maincode logo

    Head of Data

    MaincodeFull-time

    Melbourne2w

    Leadership/Strategy