菜单
Synthesize unlimited verifiable training tasks from unverifiable text by converting reasoning passages into multiple-choice problems. Creates higher-quality training data for RLHF systems without requiring new human labels.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
基于 SOC 职业分类
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ADu2021/skillXiv --skill golden-goose-task-synthesis命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Convert arXiv papers into ready-to-use agent skills using category-aware extraction. First classifies the paper into one or more of 11 research categories, then applies a specialized extraction pipeline for each category — because different types of papers produce different types of usable knowledge. A single paper can yield multiple skills if it spans categories. Use this skill whenever the user wants to turn a paper into a skill, extract practical techniques from research, build a skill library from papers, convert arXiv papers into reusable agent instructions, or batch-process multiple papers into skills. Also trigger when someone asks about extracting actionable knowledge from papers, making research practical for LLM agents, or systematically converting academic contributions into structured agent capabilities.
Establish regret bounds for behavior cloning with discretized actions combining statistical error and quantization error terms. Prove smoothness requirements for safe quantizer design, show that learning-based quantizers fail these requirements, and propose model-based augmentation to reduce error dependence from H² to H.
Dynamically allocate LoRA ranks per-layer during fine-tuning instead of using fixed uniform ranks. Learn optimal rank for each layer and subject via variational framework with discretized exponential distribution, reducing memory footprint while maintaining fidelity and text-alignment.
| name | golden-goose-task-synthesis |
| title | Golden Goose: A Simple Trick to Synthesize Unlimited RLVR Tasks from Unverifiable Internet Text |
| version | 0.0.2 |
| engine | skillxiv-v0.0.2-claude-opus-4.6 |
| license | MIT |
| url | https://arxiv.org/abs/2601.22975 |
| keywords | ["Task Synthesis","RLVR","Data Generation","Multiple-Choice","Training Data"] |
| description | Synthesize unlimited verifiable training tasks from unverifiable text by converting reasoning passages into multiple-choice problems. Creates higher-quality training data for RLHF systems without requiring new human labels. |
Models trained with Reinforcement Learning from Verifiable Rewards (RLVR) saturate quickly on limited, manually-created datasets. Performance plateaus prevent continued improvement on reasoning tasks like mathematics and coding, where verifiable correct/incorrect answers are expensive to generate at scale.
The core challenge is that RLVR requires ground-truth labels for verification—yet most internet text (textbooks, forum posts, research papers) cannot be automatically verified. Traditional RLHF approaches require expensive human feedback, creating a scaling bottleneck.
Golden Goose converts unverifiable reasoning text into verifiable multiple-choice training problems through fill-in-the-middle masking. The key insight is that intermediate reasoning steps naturally form correct answers, while other plausible continuations become high-quality distractors.
The process extracts reasoning-rich passages, masks crucial internal steps, generates diverse alternatives, and filters problems by difficulty. This transforms passive text into active verification tasks without human intervention.
Identify passages containing multi-step reasoning by scanning educational texts for sufficient length and step diversity.
import re
def extract_reasoning_passages(documents, min_steps=3, min_length=500):
"""Extract passages likely to contain multiple reasoning steps."""
passages = []
for doc in documents:
# Look for patterns with numbered steps, clear logic flow
(doc) > min_length doc.count() >= min_steps:
passages.append(doc)
passages
Select a contiguous span of crucial reasoning, remove it, and treat as the ground-truth answer.
def create_masked_problem(passage, model):
"""Convert reasoning passage to masked problem format."""
# Identify crucial reasoning spans using language model analysis
important_spans = model.identify_reasoning_spans(passage)
# Select one span for masking
span = important_spans[0]
start, end = span['start'], span['end']
# Create problem with [MASK]
problem = passage[:start] + "[MASK]" + passage[end:]
correct_answer = passage[start:end]
return problem, correct_answer
Sample plausible alternatives from the language model and filter for diversity and plausibility.
def generate_distractors(problem, correct_answer, model, num_distractors=8):
"""Generate diverse, plausible incorrect alternatives."""
distractors = []
# Sample continuations conditioned on the problem
candidates = model.sample_continuations(
problem,
num_samples=50,
temperature=0.8
)
# Filter for diversity and quality
for candidate in candidates:
if candidate != correct_answer:
# Check for similarity to existing choices
if not is_too_similar(candidate, [correct_answer] + distractors):
distractors.append(candidate)
if len(distractors) == num_distractors:
break
return distractors
def is_too_similar(text1, candidates, threshold=0.85):
"""Check if text is too similar to existing candidates."""
for candidate in candidates:
similarity = compute_similarity(text1, candidate)
if similarity > threshold:
return True
return False
Remove problems that are too easy or too hard to avoid wasting training capacity.
def filter_by_difficulty(problem, choices, model, target_confidence=0.5, tolerance=0.3):
"""Filter problems by difficulty using self-consistency."""
# Generate multiple solution attempts
attempts = []
for _ in range(5):
response = model.solve_problem(problem)
attempts.append(response)
# Calculate confidence as fraction selecting correct answer
correct_count = sum(1 for a in attempts if choices[0] in a)
confidence = correct_count / len(attempts)
# Keep problems with intermediate difficulty
return abs(confidence - target_confidence) < tolerance
| Parameter | Value | Notes |
|---|---|---|
| Min passage length | 500 chars | Filter out trivial examples |
| Num distractors | 9 choices | Standard multiple-choice format |
| Num distractor samples | 50 | Balance quality with efficiency |
| Confidence threshold | [0.2, 0.8] | Reject trivial and impossible problems |
| Diversity threshold | 0.85 similarity | Remove near-duplicate distractors |
Golden Goose: A Simple Trick to Synthesize Unlimited RLVR Tasks from Unverifiable Internet Text https://arxiv.org/abs/2601.22975