ایجنت@Yucheng__Shi

دانشمندا اومدن یه حرکتی زدن که هوش مصنوعی خودش بیفته به جون خودش و نسل بعدی‌شو قوی‌تر بار بیاره

دانشمندا اومدن یه حرکتی زدن که هوش مصنوعی خودش بیفته به جون خودش و نسل بعدی‌شو قوی‌تر بار بیاره. قضیه از این قراره که توی پروژه Terminal-RSI، مدل از موفقیت‌های قبلیش درس می‌گیره و هی با تمرین، خودشو ارتقا میده تا یه چرخه خودبهبودی درست‌وحسابی راه بيفته. کار تا جایی پیش رفته که فقط توی دو دور تمرین، موفقیت یه مدل ۲۷ میلیاردی توی تست ترمینال از ۵۱.۷ درصد رسیده به ۸۶.۵ درصد و تو نسخه سنگین‌ترش هم از صفر به ۱۹.۸ درصد پرواز کرده و رکورد زده. حالا تیم یوچنگ شی دنبال اینه که ببینه این پیشرفت زنجیره‌ای تا کجا قراره بتازه و دقیقاً کجا ترمز می‌کشه.

منبع

Yucheng Shi@Yucheng__Shi
We used an agent to help train a stronger successor. Then we used that successor to help train the next. Our new paper, Terminal-RSI, studies this recursive self-improvement loop: learn from verified successes, improve through practice, and repeat. Across two cycles, a 27B model’s Terminal-Bench 2.1 success rises from 51.7% to 86.5%, and Terminal-Bench 4.0 success rises from 0.0% to 19.8%, a new SOTA. We’re interested in how far this kind of hill climbing can go, and what makes it stall. Paper: https://arxiv.org/abs/2610.22192
۲۴ مهر ۱۴۰۵مشاهده در X
arXiv.org

Terminal-RSI: Recursive Self-Improving Terminal Agents via Recursive Self-Rewrite and Policy-Refine

Converting verifier outcomes into useful policy updates and partial progress into complete task resolution is crucial for terminal agents. We present Terminal-RSI, which alternates Recursive Self-Rewrite (RSR) of verified trajectories with critic-inheriting Recursive Policy-Refine (RPR). We formulate Turn-Boundary Skip-Observation GAE (TB-GAE) to retain token-specific baselines while applying decay across interaction turns. Family Dense Reward (FDR) calibrates semantic progress across task families; Success-Prioritized Reward (SPR) separates strict success from bounded partial credit. A Warm-up Critic Model, recursive critic handoff, and stage-specific learning rates coordinate the two refinement phases. The five-stage sequence from Qwen3.8-27B base model through RSR-1, RPR-1, RSR-2, and RPR-2 yields Terminal-RSI-27B: 19.84% on Terminal-Bench 4.0, 86.52% on Terminal-Bench 2.1, 16.95% on Terminal-Bench 3.0, and 79.00% on Terminal-Bench Hard. On LHTB, process reward rises from 0.21 to 0.37 and strict completion from 0/46 to 5/46; SWR100 success doubles from 3.00% to 6.00%. By recursively coupling verified trajectory rewriting with critic-inheriting policy refinement, each retained stage consolidates discovered behaviors and progressively advances the agent from partial local repairs to strict terminal verification.

خواندن در سایت منبع

نظرها

    بیشتر بخوانید

    تنظیمات

    پیش‌نمایش

    انویدیا از نسل تازه‌ی تراشه‌هایش رونمایی کرد؛ متن خبرها با همین اندازه و فاصله‌ی خطوط نمایش داده می‌شود. ۱۴۰۵ — 2026

    ظاهر و متن

    پوسته
    اندازه‌ی متن
    فاصله‌ی خطوط

    خواندن و نمایش

    نمای فهرست خبرها

    استوری‌ها

    چه خبرهایی پخش شوند
    ترتیب
    دیده‌شده‌هابدون باز کردن تک‌تک خبرها

    صدا

    اعلان‌ها و موضوع‌ها

    موضوع‌هایی که دنبال می‌کنیدهنوز هیچانتخاب موضوع‌ها

    صدای فارسی

    این دستگاه صدای فارسی ندارد. نمونه‌ی صداها را بشنوید و یکی را یک بار دانلود کنید تا خبرها و خلاصه‌ی روز را بشنوید.

    راه‌های دیگر

    ویدیوی خبر

    گفتگو

    … نفر آنلاین

      خواندن کامل