Long-Horizon State Tracking in LLMs: Executing MD5 through a Deep Sequence of Dependent Tool Calls

arXiv نُشر في تم التحديث AI & Machine Learning
سجّل الدخول للحفظ

الأصول والمواضيع المتأثرة

LLM GPT REPORT

لماذا يهم

التحليل معروض بالإنجليزية · الترجمة العربية قيد الإعداد

The arXiv pre‑print demonstrates that the open‑source 120‑billion‑parameter LLM (gpt‑oss‑120b) can retain exact intermediate state across 196 dependent tool calls to compute an MD5 hash, achieving correct results in most runs. The authors attribute success to keeping the model’s reasoning in context and using a voting mechanism with a secondary LLM worker.

  • article reports gpt‑oss‑120b maintains full state across 196 dependent tool calls and returns the correct MD5 digest in most runs
  • success is linked to keeping reasoning in the model's context each turn
  • voting over a thinking‑enabled worker reduces arithmetic errors

نبرة المقال

محايد كيف كُتب المقال، وفق ما ذكره المصدر.

التأثير المتوقع على السوق

محايد الثقة 42% كيف تُقرأ نسبة الثقة الأفق الزمني: المدى القصير الأثر: الأدنى

The paper provides evidence of improved long‑horizon state tracking in large language models, which could, in principle, lower engineering effort for complex AI workflows. However, the article does not reference any commercial products, companies, or market transactions, so any effect on public equities is uncertain and indirect.

المخاطر

  • the work is a research pre‑print with no disclosed commercial deployment
  • no public companies or products are named, making exposure to listed equities speculative

مسار الأدلة

الأدلة
المصدر arXiv
الادّعاء Long-Horizon State Tracking in LLMs: Executing MD5 through a Deep Sequence of Dependent Tool Calls
استنتاج الذكاء الاصطناعي Neutral · 42%
أُنشئ في 2026-09-02 04:00

مصدر التحليل بالذكاء الاصطناعي

حُلِّل بواسطة Openai/gpt Oss 120B (Groq) المنهجية v1.0 أُنشئ في
المعرّفات التقنية
وسم المزوّد
groq-openai/gpt-oss-120b
إصدار التحليل
groq-openai/gpt-oss-120b
معرّف المقال
125690

المصدر الأصلي

arXiv:2609.00012v1 Announce Type: new Abstract: Long-horizon tasks remain uncommon in large language model (LLM) evaluation, and for a reason: when each step depends on the last, per-step accuracy that looks excellent in isolation decays catastrophically, as errors cascade and the end-to-end failure probability grows sharply with length. Existing agentic benchmarks report end-to-end success but confound this state-tracking difficulty with instruction interpretation, give no control group that isolates it, and are vulnerable to shortcuts such as a hallucinated final answer, so they cannot say why a long run fails. Whether an LLM can carry exact intermediate state across many tool calls at all is itself not well established. We test this cleanly by having the model compute a cryptographic hash, MD5, step by step: a sequence of $196$ dependent tool calls over $64$ rounds while it carries four $32$-bit words $(a,b,c,d)$ in its own context from one call to the next. Interpretation is trivial and, because we implement MD5 from scratch (RFC~1321), we align every call to the ground-truth trace and check the digest to the bit, so any failure is pure bookkeeping. gpt-oss-120b, a mixture-of-experts model with only $\sim$5.5B active parameters per token, at temperature $0$ with a short fixed prompt, carries the full state across all $196$ calls and returns the correct digest on a majority of completed runs. In the strongest setting we replace every primitive tool with a second LLM, so a driver and a worker compute the whole hash from scratch with no exact-arithmetic oracle in the loop. Two ingredients decide success and neither changes the weights: keeping the model's own reasoning in its context each turn, and voting over a thinking-enabled worker to remove its modular-arithmetic slips. We localize the residual failures by origin, separating state-carrying from arithmetic and from serving.

اقرأ المقال كاملاً على arXiv

المقال الأصلي منشور بواسطة arXiv في سبتمبر 2, 2026. التحليل والرؤى المقدمة من AnalystMarkets AI.

تغطية ذات صلة