Skip to content

Add 18 new paper summaries across 5 categories + Ollama generation scripts + NeurIPS 2025 KB builder - #1

Open
odilitime wants to merge 4 commits into
cognitivetech:mainfrom
odilitime:main
Open

Add 18 new paper summaries across 5 categories + Ollama generation scripts + NeurIPS 2025 KB builder#1
odilitime wants to merge 4 commits into
cognitivetech:mainfrom
odilitime:main

Conversation

@odilitime

Copy link
Copy Markdown

Summary

Adds 18 new LLM-generated paper summaries and two Python scripts for AI/ML research paper knowledge management.

New Paper Summaries (18)

Training (4):

  • Q-SFT: Q-Learning for Language Models via Supervised Fine-Tuning (2411.05193)
  • MeRF: Motivation-enhanced Reinforcement Finetuning of Reasoning Models (2506.18485)
  • CHORD: Harmonizing SFT and RL via Dynamic Weighting (2508.11408)
  • Unextractable Protocol Models: Collaborative Training without Weight Materialization, NeurIPS 2025 (2605.23464)

AI Psychology (2):

  • BIG5-CHAT: Shaping LLM Personalities Through Training (2410.16491)
  • TRAIT: Personality Testset for LLMs with Psychometrics (2406.14703)

Mental Health (3):

  • PsyLLM: Integrating Diagnostic and Therapeutic Reasoning (2505.15715)
  • SoulSpeak: LLM-enabled Psychotherapy Chatbot (2412.02987)
  • CounselBench: Expert Evaluation of LLMs in Mental Health QA (2506.08584)

Social Science (4):

  • LLM Social Simulations Are a Promising Research Method (2504.02234)
  • Automated Social Science: LLMs as Scientist and Subjects (2404.11794)
  • Identifying Non-Replicable Social Science Studies with LLMs (2503.10671)
  • Exploring LLM Agents for Piloting Social Experiments (2508.08678)

Interactive Agents (5):

  • DPT-Agent: Dual Process Theory for Real-time Human-AI Collaboration (2502.11882)
  • Mutual Theory of Mind in Human-AI Collaboration (2409.08811)
  • Scaling LLM-based Multi-Agent Collaboration (2406.07155)
  • ChatCollab: Human-AI Collaborative Software Teams (2412.01992)
  • ReHAC: LLM-based Human-Agent Collaboration (2402.12914)

New Scripts

  • scripts/generate_summaries.py — CLI tool that uses Ollama (qwen3.6) to generate structured paper summaries from arxiv metadata. Supports batch generation and single-paper mode (--arxiv). Produces clean markdown matching the repo's style with section-by-section bullet-pointed summaries.
  • scripts/build_papers_kb.py — Knowledge base builder that aggregates papers from 5 sources (The AI Timeline, DAIR.AI, llm-research-summaries, NeurIPS 2025, arxiv-daily) into a deduplicated JSONL corpus with auto-categorization.
  • .gitignore — Added to keep .DS_Store out of the repo.

All summaries generated with qwen3.6:latest (36B Q4_K_M) via local Ollama.

… social-science, interactive-agents

Training:
- Q-SFT: Q-Learning for Language Models via Supervised Fine-Tuning (2411.05193)
- MeRF: Motivation-enhanced Reinforcement Finetuning (2506.18485)
- CHORD: Harmonizing SFT and RL via Dynamic Weighting (2508.11408)
- Unextractable Protocol Models: Collaborative Training without Weight Materialization (2605.23464)

AI Psychology:
- BIG5-CHAT: Shaping LLM Personalities Through Training (2410.16491)
- TRAIT: Personality Testset for LLMs with Psychometrics (2406.14703)

Mental Health:
- PsyLLM: Integrating Diagnostic and Therapeutic Reasoning (2505.15715)
- SoulSpeak: LLM-enabled Psychotherapy Chatbot (2412.02987)
- CounselBench: Benchmark for LLMs in Mental Health (2506.08584)

Social Science:
- LLM Social Simulations Are a Promising Research Method (2504.02234)
- Automated Social Science: LLMs as Scientist and Subjects (2404.11794)
- Identifying Non-Replicable Social Science Studies with LLMs (2503.10671)
- LLM Agents for Piloting Social Experiments (2508.08678)

Interactive Agents:
- DPT-Agent: Dual Process Theory for Human-AI Collaboration (2502.11882)
- Mutual Theory of Mind in Human-AI Collaboration (2409.08811)
- Scaling LLM-based Multi-Agent Collaboration (2406.07155)
- ChatCollab: Human-AI Collaborative Software Teams (2412.01992)
- ReHAC: RL-based Human-Agent Collaboration (2402.12914)
generate_summaries.py: CLI tool using qwen3.6 to generate structured
paper summaries from arxiv. Supports batch and single-paper modes.
build_papers_kb.py: Knowledge base builder aggregating 5 sources
(The AI Timeline, DAIR.AI, llm-research-summaries, NeurIPS 2025,
arxiv-daily) into a searchable JSONL corpus.
Copilot AI review requested due to automatic review settings June 15, 2026 04:20

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds a large batch of new arXiv paper summary markdown files across training/, models-review/, social-science/, and interactive-agents/, plus a helper script to generate summaries via Ollama and a .gitignore update.

Changes:

  • Added many new structured paper summaries (mostly new files) across multiple topic folders
  • Standardized a few existing summaries to use **Source**: [arXiv:...] link formatting
  • Added scripts/generate_summaries.py to automate summary generation + added .DS_Store to .gitignore

Reviewed changes

Copilot reviewed 31 out of 32 changed files in this pull request and generated 8 comments.

Show a summary per file
File Description
training/unextractable-protocol-models-collaborative-training-and-inference-without-weight-materialization_2605.23464.md Add paper summary
training/self-adapting-language-models_2506.10943.md Add paper summary
training/q-sft-q-learning-for-language-models-via-supervised-fine-tuning_2411.05193.md Add paper summary
training/on-policy-rl-meets-off-policy-experts-harmonizing-supervised-fine-tuning-and-reinforcement-learning-via-dynamic-weightin_2508.11408.md Add paper summary
training/inheritune-Smaller-Yet-More-Attentive-Language-Models-2404.08634.md Standardize Source link format
training/inference-time-scaling-for-generalist-reward-modeling_2504.02495.md Add paper summary
training/absolute-zero-reinforced-self-play-reasoning-with-zero-data_2505.03335.md Add paper summary
training/a-simple-motivation-can-enhance-reinforcement-finetuning-of-large-reasoning-models_2506.18485.md Add paper summary
training/InfiniPot-Efficiently-Handling-Long-Input-Contexts_LLM_Memory-Constrained-Devices-2410.01518.md Standardize Source link format
social-science/mental-health/counselbench-a-large-scale-expert-evaluation-and-adversarial-benchmarking-of-large-language-models-in-mental-health-ques_2506.08584.md Add paper summary
social-science/mental-health/beyond-empathy-integrating-diagnostic-and-therapeutic-reasoning-with-large-language-models-for-mental-health-counseling_2505.15715.md Add paper summary
social-science/mental-health/advancing-conversational-psychotherapy-integrating-privacy-dual-memory-and-domain-expertise-with-large-language-models_2412.02987.md Add paper summary
social-science/llm-social-simulations-are-a-promising-research-method_2504.02234.md Add paper summary
social-science/identifying-non-replicable-social-science-studies-with-language-models_2503.10671.md Add paper summary
social-science/exploring-large-language-model-agents-for-piloting-social-experiments_2508.08678.md Add paper summary
social-science/automated-social-science-language-models-as-scientist-and-subjects_2404.11794.md Add paper summary
social-science/ai-psychology/do-llms-have-distinct-and-consistent-personality-trait-personality-testset-designed-for-llms-with-psychometrics_2406.14703.md Add paper summary
social-science/ai-psychology/big5-chat-shaping-llm-personalities-through-training-on-human-grounded-data_2410.16491.md Add paper summary
scripts/generate_summaries.py Add Ollama-based generator script for summaries
models-review/transformers-without-normalization_2503.10622.md Add paper summary
models-review/q-sparse-all-large-language-models-can-be-fully-sparsely-activated_2407.10969.md Add paper summary
models-review/energy-based-transformers-are-scalable-learners-and-thinkers_2507.02092.md Add paper summary
models-review/demystifying-long-chain-of-thought-reasoning-in-llms_2502.03373.md Add paper summary
models-review/bitnet-b1-58-2b4t-technical-report_2504.12285.md Add paper summary
models-review/MLLM-Landscape_Review-Current-Surveys.md Standardize Source link format
interactive-agents/scaling-large-language-model-based-multi-agent-collaboration_2406.07155.md Add paper summary
interactive-agents/mutual-theory-of-mind-in-human-ai-collaboration-an-empirical-study-with-llm-driven-ai-agents-in-a-real-time-shared-works_2409.08811.md Add paper summary
interactive-agents/leveraging-dual-process-theory-in-language-agent-framework-for-real-time-simultaneous-human-ai-collaboration_2502.11882.md Add paper summary
interactive-agents/large-language-model-based-human-agent-collaboration-for-complex-task-solving_2402.12914.md Add paper summary
interactive-agents/chatcollab-exploring-collaboration-between-humans-and-ai-agents-in-software-teams_2412.01992.md Add paper summary
.gitignore Ignore macOS .DS_Store files

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

return s[:120]


def generate_one(arxiv_id, target_dir, model=MODEL, force=False):
Comment on lines +259 to +261
fname = f"{slugify(meta['title'])}_{arxiv_id}.md"
repo_path = REPO_BASE / target_dir / fname
repo_path.write_text(content, encoding="utf-8")
Comment thread scripts/generate_summaries.py Outdated
Comment thread scripts/generate_summaries.py Outdated
Comment thread scripts/generate_summaries.py Outdated
published = entry.find("a:published", NS).text[:10]
sections = []
try:
html_url = f"https://arxiv.org/html/{arxiv_id}"
Comment on lines +23 to +24
REPO_BASE = Path.home() / "Sites" / "llm-research-summaries"
BACKUP_DIR = Path.home() / "Sites" / "hermes-lean" / "papers_kb" / "generated_summaries"
@@ -0,0 +1,384 @@
# On-Policy RL Meets Off-Policy Experts: Harmonizing Supervised Fine-Tuning and Global Weighting via Dynamic Weighting
- Add --force flag existence check (skip existing files)
- Fix BACKUP_DIR.mkdir with parents=True
- Use HTTPS for arxiv API (http→https)
- Capture all arxiv categories (a:category) + primary category
- Add --repo-base and --backup-dir CLI args for portability
- Fix CHORD paper filename (weightin→weighting) and title
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants