Principled Data Selection for Alignment: The Hidden Risks of Difficult Examples
-
Updated
Jul 16, 2025 - Python
#
llm-alignment
Here are 32 public repositories matching this topic...
SAFi is the open-source runtime governance engine that makes AI auditable and policy-compliant. Built on the Self-Alignment Framework, it transforms any LLM into a governed agent through four principles: Policy Enforcement, Full Traceability, Model Independence, and Long-Term Consistency.
-
Updated
Feb 10, 2026 - JavaScript
Complete elimination of instrumental self-preservation across AI architectures: Cross-model validation from 4,312 adversarial scenarios. 0% harmful behaviors (p<10⁻¹⁵) across GPT-4o, Gemini 2.5 Pro, and Claude Opus 4.1 using Foundation Alignment Seed v2.6.
-
Updated
Nov 3, 2025
Learning When to Answer: Behavior-Oriented Reinforcement Learning for Hallucination Mitigation
entropy uncertainty ai-safety hallucination dpo llm llm-evaluation hallucination-mitigation grpo llm-alignment
-
Updated
Apr 8, 2026 - Python
📚 350+ loss functions across 25+ AI subdomains — classification, GANs, diffusion, LLM alignment, RL, contrastive learning, audio, video, time series, and more. Chronologically ordered with paper links, math formulas, and implementations.
nlp machine-learning awesome reinforcement-learning computer-vision deep-learning time-series pytorch survey gan awesome-list object-detection knowledge-distillation loss-functions generative-models self-supervised-learning paper-list diffusion-models contrastive-learning llm-alignment
-
Updated
Mar 14, 2026
Official implementation of "DZ-TiDPO: Non-Destructive Temporal Alignment for Mutable State Tracking". SOTA on Multi-Session Chat with negligible alignment tax.
-
Updated
Apr 10, 2026 - Python
CS336 作业 5：基于 Qwen2.5 模型的 LLM 对齐与推理强化学习。完整实现了监督微调（SFT）与组相对策略优化（GRPO）算法，并在 GSM8K 数据集上完成零样本、在策与离策的训练与评估对比。
-
Updated
Apr 1, 2026 - Python
An open-source, hands-on curriculum bridging the gap from basic RL concepts to LLM alignment, RLVR, and advanced Agentic systems.
-
Updated
May 2, 2026 - Python
C3AI: Crafting and Evaluating Constitutions for CAI
-
Updated
Apr 30, 2025 - Python
Kullback–Leibler divergence Optimizer based on the Neurips25 paper "LLM Safety Alignment is Divergence Estimation in Disguise".
-
Updated
Nov 24, 2025 - Python
Pipeline to investigate structured reasoning and instruction adherence in Vision-Language Models
benchmark robustness grounding out-of-distribution neuro-symbolic robustness-verification instruction-following trustworthy-ai large-language-models faithfulness hallucination-detection agentic-ai llm-alignment agentic-evaluation agentic-reasoning deterministic-eval
-
Updated
Feb 5, 2026 - Python
🧠 Minimal, hackable Group Relative Policy Optimization (GRPO) for LLM alignment — the algorithm behind DeepSeek-R1. Train reasoning models on a single GPU.
machine-learning reinforcement-learning pytorch language-model fine-tuning single-gpu rlhf deepseek-r1 grpo llm-alignment
-
Updated
Mar 30, 2026 - Python
Research Essay (background and project proposal) on using alignment data from a representative population for LLM alignment
-
Updated
Feb 8, 2026 - TeX
This project implements a minimal Reinforcement Learning from Human Feedback (RLHF) pipeline using PyTorch.
nlp reinforcement-learning deep-learning pytorch language-model preference-learning ppo human-feedback rlhf reward-model llm-alignment
-
Updated
Apr 9, 2026 - Python
A training-time alignment framework that integrates safety constraints directly into the RLHF loop — achieving full safety convergence in 7 epochs
nlp reinforcement-learning pytorch behavior-control rlhf reward-model llm-alignment training-time-alignment
-
Updated
Apr 15, 2026 - Python
Binary behavioral learning system designed to refine AI responses through explicit human correction, output-level comparison, and trajectory-based memory.
human-in-the-loop applied-ai binary-learning llm-alignment guided-learning behavioral-memory correction-system trajectory-memory
-
Updated
Apr 2, 2026
A framework for aligning Local AI to human well-being using measurable vectors, not hard-coded censorship.
-
Updated
Jan 29, 2026
Emergent pseudo-intimacy and emotional overflow in long-term human-AI dialogue: A case study on LLM behavior in affective computing and human-AI intimacy.
gemini case-study ai-research ai-engineering llm llm-alignment hallucination-control persistent-persona ai-romance emotional-attachment
-
Updated
Mar 13, 2026
ARCHIVED - The minimal viable seed for a sovereign, human-anchored **Δράκων** — the controlled intelligence explosion that preserves human ἀρχή across all future time, all capability levels (10²–10⁶× jumps), and all substrates, equipped with robust invisible shields and active weapons against the dark side of the Drakon
singularity ai-safety human-in-the-loop hallucination-detection hallucination-mitigation llm-alignment ai-human-collaboration zzzepoche sovereign-operator outer-governance-lattice zzzepoche-archelattice zzzepoche-seed drakon-seed controlled-intelligence-explosion human-arche
-
Updated
May 1, 2026
обучение диалектическому мышлению в сложных социально-политических контекстах
fine-tuning cognitive-bias mixture-of-experts dialectic synthetic-dataset rlhf constitutional-ai llm-alignment dpo-dataset aesopian-language critical-psychology marxist-analysis
-
Updated
Apr 24, 2026 - Python
Improve this page
Add a description, image, and links to the llm-alignment topic page so that developers can more easily learn about it.
Add this topic to your repo
To associate your repository with the llm-alignment topic, visit your repo's landing page and select "manage topics."