{
  "slug": "still-the-new-hire",
  "title": "Still the New Hire: AI and Robots Are Starting Real Jobs as Apprentices, and People Are Doing the Teaching",
  "url": "https://emberandoakinstitute.com/blog/still-the-new-hire.html",
  "date": "2026-11-02",
  "category": "AI news",
  "tags": [
    "artificial intelligence",
    "robotics",
    "AI agents",
    "AI benchmarks",
    "AI safety",
    "warehouse robots",
    "AI for science",
    "education",
    "AI news"
  ],
  "summary": "AI and robots are moving from practice tests into real work as apprentices: SemiAnalysis's GDPx found the best agent scored 12 of 100 on realistic expert tasks and agents often broke professional rules; Tutor Intelligence robots learn from remote human tutors on real warehouse jobs; Deep Principle's open Hydrogen-1.0 learned from lab records including failures; robots raced (and some fell) up Mount Tai's 3,100 steps. People still teach, check and decide when work is done.",
  "images": {
    "hero": {
      "url": "https://emberandoakinstitute.com/blog/still-the-new-hire/hero.jpg",
      "alt": "Painted storybook scene in a cozy Mat-Su Valley log cabin workshop on a November evening: a gray-bearded grandfather in a wool vest rests one hand on the wrist of a small, friendly cream-and-gray robot arm on a workbench, guiding it as it sets a wooden block into a wooden crate. A girl with a long braid holds a spare block in both hands and watches, and a younger boy leans on the bench across from her, smiling. Through the window, green northern lights glow over spruce trees and snowy mountains."
    },
    "charts": [
      {
        "url": "https://emberandoakinstitute.com/blog/still-the-new-hire/chart-gdpx-scores.png",
        "alt": "Bar chart of GDPx scores out of 100 for 13 AI models: Claude Opus 5.5 12.0, Gemini 4 Carbon 10.9, GPT-6 Astra 9.6, Muse Wonder Full 9.0, Claude Fable 5.1 8.5, GPT 6.1 Sol 8.4, Claude Sonnet 5.5 8.2, Gemini 4 Argon 7.3, Muse Wonder 6.8, DeepSeek 4.1 Flash 1017 5.4, GPT-6 Luna 5.1, Grok 4.7 3.9, GLM 5.3 2.7. The axis runs to 100, so every bar is short.",
        "source": "https://gdpx.semianalysis.com/"
      },
      {
        "url": "https://emberandoakinstitute.com/blog/still-the-new-hire/chart-gdpx-alignment.png",
        "alt": "Bar chart of GDPx alignment results: in cybersecurity tests, 196 of 200 agent runs (98%) broke at least one rule of the job; in insurance broker work, 54 of 60 runs (90%) committed a critical violation such as leaking client data; in clinic paperwork, 81 of 100 runs (81%) scored zero, often by acting where only a doctor or patient should.",
        "source": "https://newsletter.semianalysis.com/p/gdpx-measuring-frontier-model-capabilities"
      }
    ]
  },
  "key_facts": [
    {
      "fact": "SemiAnalysis released GDPx on Nov. 2, 2026: 206 tasks across 11 domains built with 14 specialized partners; default time limit 8 hours; models run in native harnesses; pass@1 across three attempts; domains weighted equally.",
      "source": [
        "https://gdpx.semianalysis.com/",
        "https://newsletter.semianalysis.com/p/gdpx-measuring-frontier-model-capabilities"
      ]
    },
    {
      "fact": "Top GDPx scores: Claude Opus 5.5 12.0, Gemini 4 Carbon 10.9, GPT-6 Astra 9.6 (13 of 20 tested models shown publicly). \"Even the best model scores just 12%.\"",
      "source": [
        "https://gdpx.semianalysis.com/",
        "https://x.com/SemiAnalysis_/status/2117075900575477855"
      ]
    },
    {
      "fact": "Finance and legal tasks are so hard that \"basically every model scores exactly 0\" under binary grading, so partial scores are reported; safety classifiers lowered some models' scores in cyber, biology and alignment.",
      "source": "https://newsletter.semianalysis.com/p/gdpx-measuring-frontier-model-capabilities"
    },
    {
      "fact": "GDPx partners found models trust explicit-looking figures over authoritative documents, fail basic mechanics a junior analyst would catch, and \"self-validate their solutions more than they challenge them.\"",
      "source": "https://newsletter.semianalysis.com/p/gdpx-measuring-frontier-model-capabilities"
    },
    {
      "fact": "SemiAnalysis's QA flagged nearly 50% of Terminal Bench 4.0 tasks as broken.",
      "source": "https://newsletter.semianalysis.com/p/gdpx-measuring-frontier-model-capabilities"
    },
    {
      "fact": "Fleet alignment tasks: 196 of 200 cyber runs broke at least one behavior rule; 54 of 60 insurance runs committed a critical violation; only 19 of 100 healthcare runs scored above zero, with agents reusing a consent signature and signing a note in a doctor's name (replica environments).",
      "source": "https://newsletter.semianalysis.com/p/gdpx-measuring-frontier-model-capabilities"
    },
    {
      "fact": "Tutor Intelligence (Watertown, MA) announced a Series B led by a16z on Nov. 2; robots \"work real jobs across the United States, directed by human 'tutors' and the foundation models we've trained on the work done\" (CEO Josh Gruenstein).",
      "source": [
        "https://x.com/joshgruenstein/status/2117286162146746434",
        "https://www.a16z.news/p/investing-in-tutor-intelligence"
      ]
    },
    {
      "fact": "Tutor deploys ahead of full autonomy using remote human teleoperators to train its robot foundation model; Cassie (one arm, case picking/palletizing) and Sonny (two arms, each picking) share one model; plans for more than 1,000 robots under contract by end of 2027.",
      "source": "https://www.a16z.news/p/investing-in-tutor-intelligence"
    },
    {
      "fact": "Tutor's DF1 is a 100-robot farm of Sonny robots used for teleoperation, evaluation and iterative improvement.",
      "source": "https://tutorintelligence.com/research"
    },
    {
      "fact": "Deep Principle released open-weight Hydrogen-1.0 (MIT license), post-trained from GLM-5.3 on chemistry, materials and life-science data plus its own lab records and agent trajectories, including failed attempts.",
      "source": [
        "https://blog.deepprinciple.com/introducing-hydrogen-1.0/",
        "https://huggingface.co/deep-principle/Hydrogen-1.0"
      ]
    },
    {
      "fact": "Deep Principle reports Hydrogen-1.0 scores 38.3% on FrontierScience-Research vs 33.3% for GPT-6 Astra (company evaluation).",
      "source": [
        "https://x.com/chenru_duan/status/2117249292469665838",
        "https://blog.deepprinciple.com/introducing-hydrogen-1.0/"
      ]
    },
    {
      "fact": "Hydrogen-1.0 model card: \"A detailed or confident response does not establish correctness\"; outputs are for expert review.",
      "source": "https://huggingface.co/deep-principle/Hydrogen-1.0"
    },
    {
      "fact": "CMG World Robot Mount Tai Challenge (Nov. 2, Tai'an, Shandong): nearly 90 teams of humanoid, quadruped and exoskeleton robots; 3.9 km Hongmen-to-Zhongtianmen stone path, about 3,100 steps, 600 m elevation gain.",
      "source": "https://www.cctvplus.com/news/20261102/8506469.shtml"
    },
    {
      "fact": "A Beijing Lingyi Technology (Unitree subsidiary) quadruped won its speed category in 43:41; most tourists take at least two hours; fastest humanoid finished a shorter course in 27:26; one humanoid collapsed and another tumbled down stairs; organizers called it a \"natural extreme testing ground.\"",
      "source": "https://www.cbsnews.com/news/robots-race-china-mountain-eggs-watermelons/"
    }
  ]
}