Tagged articles
reward model
23 articles · Page 1 of 1
Alibaba and Zhejiang University's Astar model enables industrial AI systems to autonomously evolve by learning from their own Git commit history and experiment results, achieving 54% single-generation success rate with a 0.6B model—surpassing GPT-5.5 and human experts—and delivering 23.6% offline HitRate and 4.86% GMV gains in Lazada's ad system.
AlibabaAstarGit history mining
0 likes · 15 min read

ZhiYuan’s Genie Envisioner‑Sim 2.0 (GE 2.0) captured the overall WorldArena world‑model title without any task‑specific tuning, demonstrating superior long‑sequence stability, multi‑view generation, real‑time inference and a closed‑loop reward feedback loop that outperforms industry baselines across 16 metrics and three real‑world tasks.
Closed-loop EvaluationEmbodied AIGE 2.0
0 likes · 9 min read

A team from Peking University built an LLM‑driven framework that iteratively generates verification functions and uses a reward model with divide‑and‑conquer to improve the planar Steiner ratio from the long‑standing lower bound of 0.824 to 0.8559, a result accepted at ICML 2026 and verified by human experts.
Gilbert‑Pollak conjectureLLMMathematical AI
0 likes · 9 min read


AIWalkerMar 3, 2026 · Artificial Intelligence
RetouchIQ introduces an instruction‑driven AI retouching system that uses a general reward model to interpret abstract user commands, delivering precise image adjustments with higher semantic consistency and visual naturalness than existing multimodal large language models, thereby lowering the technical barrier for cinematic‑style edits.
AI Image EditingRetouchIQcreative automation
0 likes · 3 min read

This article explains the RLHF (Reinforcement Learning from Human Feedback) pipeline that powers ChatGPT and other large language models, covering the limitations of traditional fine‑tuning, the creation of human‑feedback datasets, reward‑model training, loss design, and the final PPO‑based fine‑tuning step.
ChatGPTHuman FeedbackPPO
0 likes · 8 min read

A recent study reveals that tiny symbols like colons or generic reasoning prefixes can cause large language models used as reward judges to issue false‑positive rewards, but an enhanced reward model called Master‑RM, trained with adversarial data, eliminates this vulnerability across multiple LLMs and languages.
AI safetyLLMMaster-RM
0 likes · 10 min read


JD TechMar 26, 2025 · Artificial Intelligence
The JD advertising team proposes a CTR‑driven advertising image generation framework (CAIG) that leverages multimodal large language models, a novel reward model, and product‑centric preference optimization to produce ad images with superior click‑through performance, validated by extensive offline and online experiments.
CTR optimizationadvertising image generationmultimodal LLM
0 likes · 10 min read

This paper presents a CTR‑driven advertising image generation framework that leverages multimodal large language models, reward modeling, and reinforcement learning to produce product‑centric ad visuals with higher click‑through performance, validated by extensive offline and online experiments.
CTR optimizationadvertising image generationmultimodal LLM
0 likes · 13 min read

Taobao’s AIGC pipeline combines a human‑feedback multimodal reward model, audio‑visual joint pre‑training, and Mixture‑of‑Experts distillation to clean data, align outputs with user preferences, and achieve state‑of‑the‑art multimodal LLM performance that drives content cold‑start and conversion gains in e‑commerce.
AIGCcontent generationdata optimization
0 likes · 10 min read

The document describes the evolution, design principles, key technologies, online inference workflow, evaluation methods, and sample‑generation techniques of a large‑language‑model‑based multi‑agent system that powers a 24/7 e‑commerce merchant assistant, highlighting its benefits, challenges, and future work.
AI planningLLMmulti-agent
0 likes · 21 min read


AIWalkerFeb 4, 2025 · Artificial Intelligence
This article reviews a comprehensive study that applies Chain‑of‑Thought reasoning to autoregressive text‑to‑image generation, introducing extended test‑time computation, direct preference optimization, and two custom reward models (PARM and PARM++) that together improve generation quality by up to 15% over Stable Diffusion 3.
Direct Preference OptimizationImage Generationchain-of-thought
0 likes · 13 min read


NewBeeNLPDec 3, 2024 · Artificial Intelligence
The article reflects on open‑source LLMs like Qwen2 and Llama 3.1, questioning whether models should self‑review answers, how hidden states might signal uncertainty, the role of loss‑function design, scaling laws, and the trade‑offs between PPO and DPO in alignment.
Scaling LawSelf-Correctionlarge language models
0 likes · 9 min read

This article breaks down the replication of OpenAI’s o1 model into four phases—assessment, journey‑learning foundation, component implementation, and training—while highlighting key challenges such as building scalable long‑thought data, reward models, and policy reasoning trees, and discusses the broader impact of o1’s reasoning abilities.
AI ReasoningLLM replicationOpenAI o1
0 likes · 18 min read

The author speculates on OpenAI’s o1 architecture, proposing that it relies on reinforcement learning guided by a generalizable, process‑supervised reward model, and outlines data collection, multi‑model generation, and training tweaks needed to realize such a system.
AI researchLLMRLHF
0 likes · 8 min read

This article analyses the training tricks behind OpenAI's o1 model, explaining test/inference‑time scaling laws, post‑training techniques, process‑supervised reward models (PRM), various inference‑time search methods, data‑collection pipelines, and the trade‑offs between allocating compute to pre‑training versus inference.
LLM inferenceOpenAI o1Post-Training
0 likes · 34 min read

This article provides an in‑depth technical analysis of OpenAI’s new multimodal model o1, explaining its self‑play reinforcement‑learning pipeline, novel train‑time and test‑time scaling laws, inference‑time thinking process, and possible architectural variants, while also discussing broader implications for large‑language‑model research.
OpenAI o1inference thinkingreward model
0 likes · 37 min read

This article analyzes how large‑model AIGC reshapes the production, evaluation, and deployment of advertising creatives on Xiaohongshu, detailing the business motivations, technical pipeline, controllable generation, reward‑model filtering, and experimental results that balance commercial efficiency with community tone.
AIGCAdvertisingIndustry Case Study
0 likes · 14 min read


NewBeeNLPSep 2, 2024 · Artificial Intelligence
This article presents a comprehensive technical walkthrough on enhancing large language model mathematical reasoning by reviewing model architectures, introducing mixed CoT‑PoT instructions, generating and filtering synthetic data, and applying multi‑stage training optimizations such as RFT, PPO, and DPO, with detailed experimental results and Q&A insights.
AIMath Reasoninglarge language models
0 likes · 17 min read

This article presents a comprehensive approach to enhancing large language models' mathematical reasoning by reviewing model architectures, introducing mixed CoT‑PoT instructions, generating and filtering synthetic data, and applying multi‑stage training optimizations such as RFT, PPO, and DPO, with detailed experimental results and Q&A.
AIMath Reasoninglarge language models
0 likes · 16 min read


NewBeeNLPApr 1, 2024 · Artificial Intelligence
This article provides a detailed technical walkthrough of Llama 2's Reinforcement Learning with Human Feedback pipeline, covering human preference data collection, reward‑model design and training, iterative fine‑tuning with PPO and rejection sampling, the Ghost Attention technique for multi‑turn consistency, and the resulting experimental evaluations.
Ghost AttentionLlama 2PPO
0 likes · 18 min read

The article explains practical techniques for choosing and constructing fine‑tuning data for large language models, covering data diversity through similarity‑based clustering, semi‑supervised filtering with binary classifiers, and uncertainty‑driven sampling using perplexity or reward models to build an efficient, low‑cost pipeline.
Active Learningclusteringdata selection
0 likes · 9 min read


21CTOJul 23, 2023 · Artificial Intelligence
This article translates and analyzes Nathan Lambert’s commentary on Meta’s Llama 2 paper, detailing the model’s architecture, training data, RLHF pipeline, reward models, evaluation methods, safety improvements, licensing terms, and the broader implications for open‑source large language models.
Llama 2Meta AIModel Evaluation
0 likes · 22 min read

This article explains how InstructGPT's reward model is trained by collecting human‑annotated ranking sequences instead of absolute scores, describes the rank‑loss formulation, provides Python code for the model and loss computation, and presents experimental results demonstrating the approach.
InstructGPTPythonRLHF
0 likes · 9 min read
