Tagged articles
Post-Training
52 articles · Page 1 of 1
The RSI Arena experiment had 10 frontier AI agents autonomously post-train a base model under equal compute constraints, evaluating via benchmarks and human blind voting; results show optimal strategies beat raw compute, human preference correlates with response length not capability, and popularity doesn't guarantee safety, prompting the RSIHF framework for human-in-the-loop recursive improvement.
AI agentsAI safetyHuman Feedback
0 likes · 14 min read

Leaked R&D records reveal Zhipu is conducting end-to-end programming post-training on 32B, 122B, and 256B models for GLM-5.5, targeting 100% recursive self-improvement via coding agents, a 100T token dataset goal, and new scale-correctness validation methods.
100T tokensGLM-5.5HarnessPro
0 likes · 10 min read


AntTechOct 27, 2026 · Artificial Intelligence
Ant Group's Bailing team open-sources Ling-3.1-flash-base (560B MoE, 25B active) and post-trained Ling-3.1-flash-1019 (44 on Artificial Analysis Index), detailing architecture, benchmark results across 47 tests, vision model Ling-3.1-flash-VL, and real-world Personal Agent applications like homework analysis and travel planning.
Ant GroupLLMMoE architecture
0 likes · 12 min read

This article provides a comprehensive analysis of On-Policy Distillation (OPD), covering white-box distillation with token-level probability matching, self-distillation using privileged context, black-box distillation via reward modeling, and industrial applications from MiMo, DeepSeek, and Qwen, while discussing OPD's limitations and generalization capabilities.
Black-Box DistillationGKDMiniLLM
0 likes · 56 min read

Diogo Almeida argues that current RLHF-trained AI excels at human-assisted tasks but fails at true automation because its optimization target is human preference, not calibrated correctness, and proposes a new post-training paradigm optimizing for reliable, automated decision-making in software.
AI assistantsAI automationPost-Training
0 likes · 12 min read

Tsinghua researchers introduce MGFlow and Elastic Forcing, two distributional training methods that achieve state-of-the-art one-step image generation and enable efficient 14B-parameter streaming video model post-training on just 8 GPUs, using sample distribution matching instead of per-sample supervision.
Elastic ForcingMGFlowPost-Training
0 likes · 20 min read

Zhipu AI founder Tang Jie outlines GLM's five-stage evolution toward AGI — Knowledge, Chat, Coding, LHT, and AAS — highlighting three scaling directions: pre-training, post-training with RL, and agentic reasoning, while introducing SAO for long-horizon agent training and emphasizing memory over context length.
AGIGLMMemory Mechanisms
0 likes · 10 min read

Tsinghua researchers open-source VeriLoop E2, a post-trained model using verifiable recurrence to achieve 67.35% zero proportion lower bound for Riemann hypothesis, while setting new benchmarks in code, math, and physics reasoning with external verification.
Post-TrainingQwen 3.8-27BRiemann hypothesis
0 likes · 40 min read

Researchers from Virginia Tech, UW-Madison, and Dartmouth introduce SAGE, a post-training method that uses symbolic closure analysis to diagnose exploration and accumulation biases in long-horizon reasoning, applying algebraic sparsification and hyperbolic structure guidance to improve sampling and provide early feedback, achieving near 8x Lean verification pass rate on Andrews-Curtis tasks and outperforming baselines across 12 benchmarks.
Andrews-CurtisLean verificationLong-Horizon Reasoning
0 likes · 14 min read

Jev, a new System One model from TypeSafe AI, outputs calibrated decisions instead of tokens using RLCD rather than RLHF, offering speed, generalization, and calibrated confidence that could shrink business post-training roles for classification, routing, and filtering tasks, leaving only constrained long-text generation and non-standard schema extraction.
CalibrationDecision ModelJev
0 likes · 20 min read

Naive AI demonstrates AI-driven R&D where agents autonomously redesign model architectures, optimize training systems, and achieve breakthrough benchmarks—Naive-N0.5-Flash surpasses larger models on coding and research tasks, while AI-built NaiveRT delivers 30-50x inference speedups and AutoWM sets new world model records through self-discovered algorithms.
AI for AIAI research agentsInference Optimization
0 likes · 22 min read

Shanghai Jiao Tong University's LIFT method enables Vision-Language-Action models to learn force perception through post-training alone, achieving significant performance gains on contact-rich tasks with only 20-30 force-labeled demonstrations while preserving pretrained knowledge via weight copying and shifted causal attention.
CoRL 2026LIFTPost-Training
0 likes · 8 min read

The article analyzes how enterprise AI is evolving from external harness scaffolding to internalized model weights, forming a closed loop where harness strategies are validated in production, decision traces become private evaluation and post-training data, and stable behaviors are distilled into specialized model weights, as demonstrated by Harvey, Glean, Engram, Strands, NVIDIA×Palantir, and Grok 4.7.
AI LoopDecision TracesHarness Architecture
0 likes · 16 min read


java1234Sep 16, 2026 · Artificial Intelligence
This article analyzes five major paradigm shifts in large model post-training over the past six months, covering expert distillation, online distillation as RL alternative, RLVR refinements, SFT-RL distribution mismatch, and data quality as irreducible constraint, with specific papers and metrics.
Post-TrainingRLVRReinforcement Learning
0 likes · 11 min read

Analysis of AI-era data engineering job descriptions reveals emerging technology stacks including RAG pipelines, model post-training, agent engineering, and AI infrastructure, outlining five key technical directions for data professionals transitioning into AI roles.
AI data engineeringAI infrastructureAgent Engineering
0 likes · 4 min read

This analysis of Harvey, a $15.5B legal AI company, reveals its 'vertical Neo-Lab' model that integrates real legal workflows, legal engineers, expert evaluations, agent runtimes, and synthetic training environments into a closed loop, prioritizing task definition and evaluation before model training, offering five lessons for enterprise AI startups.
HarveyLegal AIModel Evaluation
0 likes · 11 min read

This article analyzes DeepSeek Harness from a post-training perspective, revealing how its architecture bakes enterprise evolution into environment shaping and trajectory sedimentation, validated by a self-evolution POC that identifies interface design, contract shape, and feedback quality as critical bottlenecks.
Agent ArchitectureDeepSeek HarnessPoC validation
0 likes · 37 min read

Microsoft research shows sliding window attention with four attention sinks achieves 99% performance recovery on knowledge tasks without any post-training, matching linear attention methods requiring hundreds of millions of tokens while outperforming them 2-10x on long-context benchmarks.
KV CacheLinear attentionLong Context
0 likes · 9 min read

As AI models and frameworks turn into off‑the‑shelf components, enterprises must convert their unique business knowledge—task standards, evaluations, harnesses, and continual learning loops—into proprietary intelligent assets that persist beyond each model upgrade, creating a sustainable competitive moat.
AI agentsContinual LearningPost-Training
0 likes · 28 min read

The article argues that reinforcement learning is crucial in post‑training because it refines and localizes chain‑of‑thought patterns learned during supervised fine‑tuning, improves model controllability, and can be complemented or substituted by distillation—especially soft distillation—to transfer high‑quality patterns from stronger teachers to weaker models.
LLMPost-TrainingReinforcement Learning
0 likes · 12 min read

As foundation models become more capable, AI application firms initially rely on external APIs, but in high‑stakes domains they must capture task standards, expert rubrics, data, and feedback in a continuous learning loop, prompting a strategic shift from renting to owning core intelligence, illustrated by Harvey’s legal AI practice.
AI strategyLegal AIPost-Training
0 likes · 30 min read

The article surveys five emerging approaches to LLM continual learning—external agent memory, context engineering, post‑training, pre‑training, and self‑modifying models—explaining how each tackles the core obstacle of catastrophic forgetting, citing benchmarks such as TRACE, ACE, and SDFT, and reflecting on Karpathy’s ten‑year timeline.
Agent MemoryCatastrophic ForgettingContinual Learning
0 likes · 17 min read

DeepSeek released the V4‑Flash model with the same architecture as the preview but a revamped post‑training pipeline—SFT, reinforcement learning with GRPO, and distillation—yielding dramatic benchmark jumps and illustrating how post‑training now defines the model's real‑world capabilities.
DeepSeekGRPOLLM training
0 likes · 11 min read

DeepSeek announced the public beta of its V4‑Flash model, highlighting a dramatic agent capability upgrade, performance gains from post‑training that surpass the previous preview and rival Opus 4.8 on DSBench tests, native Responses API support, full Codex compatibility, and easy setup scripts for developers.
AI ModelAgentCodex Integration
0 likes · 6 min read

In a four‑hour investor briefing, DeepSeek founder Liang Wenfeng explains that the real competitive edge for large language models now lies in the ability to continuously produce high‑quality training signals, a capability limited by time rather than capital.
AI industryDeepSeekHigh-Quality Data
0 likes · 10 min read

The article explains how post‑training shifts AI investment from massive pre‑training compute to targeted, low‑cost fine‑tuning that lets enterprises embed proprietary business logic, outlines the ROI formula based on data exclusivity, eval formalizability, and task frequency, and presents real‑world case studies.
AI economicsPost-TrainingReinforcement Learning
0 likes · 24 min read

The article analyzes whether post‑training on GLM‑5.1/5.2 can bridge the gap to the banned Mythos model, citing Musk’s nine‑month claim, Tang’s rebuttal, Mind Lab’s benchmark gains, architectural adaptations, and the high barriers that make post‑training a critical yet scarce capability in China.
AI developmentGLM-5.2IndexCache
0 likes · 9 min read

The article analyzes China’s race to achieve Mythos‑level intelligence, contrasting Musk’s nine‑month claim with Tang’s skepticism, and highlights Mind Lab’s unique post‑training work on GLM‑5.1/5.2 that has already delivered significant benchmark gains, while outlining the technical hurdles and timeline uncertainties.
AI development in ChinaGLM-5.2Mind Lab
0 likes · 8 min read

VibeThinker‑3B, a 3‑billion‑parameter language model built on Qwen2.5‑Coder‑3B, achieves scores within the range of 671 B‑parameter models on benchmarks such as LiveCodeBench, AIME26, IMO‑AnswerBench and GPQA, thanks to a two‑stage SFT, multi‑domain reinforcement learning, offline self‑distillation and a claim‑reliability (CLR) evaluator that together push its reasoning ability to the frontier.
Post-TrainingReinforcement LearningVibeThinker-3B
0 likes · 9 min read

On-Policy Distillation (OPD) trains large language models by letting the student generate its own inference paths while the teacher supplies token‑level guidance, offering denser signals than RL but sometimes failing when teacher and student reasoning diverge, as detailed by THUNLP’s recent study.
Distillation MetricsPost-TrainingToken-level Supervision
0 likes · 8 min read

The article examines the bottleneck of synchronous rollout in large‑model post‑training, proposes an asynchronous design using gradient accumulation and a global micro‑batch count to preserve loss equivalence, and introduces LogitsExpress for efficient top‑K knowledge‑distillation communication, all implemented in the lightweight LiteScale framework.
Post-Trainingasynchronous rolloutdistributed training
0 likes · 16 min read

The CHAI framework introduced by CMU and Harvard defines a structured video‑language annotation scheme, scalable human‑AI oversight, and a post‑training pipeline that enables an 8B open‑source model to outperform closed‑source GPT‑5 and Gemini‑3.1‑Pro on professional cinematic techniques.
Post-TrainingQwen3-VLannotation
0 likes · 11 min read

The article details how Xiaomi‑Robotics‑0 achieves precise earphone‑to‑case insertion after only 20 hours of post‑training, outlines the sub‑millimetre precision challenges, presents a triple‑strategy (asynchronous execution, adaptive loss re‑weighting, Λ‑shape attention mask and random masking) to avoid the "lazy effect", and releases the full pipeline and code as open source for the robotics community.
Embodied AIPost-TrainingXiaomi Robotics
0 likes · 6 min read


ArchitectApr 25, 2026 · Artificial Intelligence
The DeepSeek V4 technical report shows how a 1 million‑token context forces a redesign of attention, KV‑cache, optimizer, quantization and inference budgeting, turning long‑context capability from a costly showcase into a production‑ready feature for agents, search and Chinese professional tasks.
1M contextAgentic SearchAttention optimization
0 likes · 28 min read


AIWalkerApr 20, 2026 · Artificial Intelligence
VA‑π introduces a lightweight post‑training framework that uses variational inference and reinforcement learning to align tokenizers with visual autoregressive generators, achieving dramatic quality gains, extreme training efficiency, and robust pixel‑level reconstruction across diverse image generation tasks.
Pixel AlignmentPost-TrainingReinforcement Learning
0 likes · 14 min read

World Engine introduces a post‑training pipeline that combines high‑fidelity 3DGS simulation, hard‑case mining with diffusion generation, and reinforcement‑learning optimization to give autonomous‑driving models true decision‑making ability, surpassing data‑scaling limits and achieving significant safety gains in both industrial simulations and real‑world tests.
Autonomous DrivingPhysical AIPost-Training
0 likes · 11 min read

This article systematically reviews the core post‑training techniques for large language models—including supervised fine‑tuning, RLHF, PPO, GRPO, DPO, RLVR and Agentic RL—explains their evolution, compares their trade‑offs, and highlights the most promising research directions for 2025‑2026.
AI AlignmentGRPOLLM
0 likes · 20 min read

This article systematically explains the post‑training pipeline for large language models, covering supervised fine‑tuning, RLHF, PPO, GRPO, RLVR, DPO and emerging Agentic RL, while illustrating each method with analogies, detailed workflows, tables, and recent research findings.
DPOGRPOLLM
0 likes · 24 min read

The MIT‑CSAIL paper introduces RandOpt, a single‑step, gradient‑free, fully parallel post‑training algorithm that adds Gaussian noise to pretrained LLM weights and ensembles the results, achieving or surpassing PPO/GRPO performance by exploiting dense "neural thickets" that emerge as model scale grows.
LLMPost-TrainingRandOpt
0 likes · 12 min read

This article shares practical insights and pitfalls from six months of applying reinforcement learning to fine‑tune large language models, covering exploration efficiency, training stability, model selection, and special considerations for thinking‑oriented agents.
AIEfficiencyLLM
0 likes · 12 min read

The SOP (Scalable Online Post‑training) framework redesigns VLA post‑training from offline, single‑machine, sequential processing to a distributed, parallel online system, allowing robot fleets to continuously learn, share experiences, and scale intelligence while maintaining stability and generalization in complex real‑world environments.
Post-TrainingRoboticsSOP
0 likes · 11 min read

This article provides a detailed technical analysis of the Olmo‑Thinking project, covering why a new open‑source LLM was built, the challenges of reinforcement learning at scale, data‑mix optimization, architectural bottlenecks such as missing GQA and QK‑Norm, and the post‑training techniques used to improve reasoning and long‑context capabilities.
Post-TrainingRLVRdata selection
0 likes · 20 min read

This article explains Meta's LLaMA model series, detailing LLaMA 2's architectural innovations—RMSNorm, SwiGLU, rotary positional encoding, and grouped-query attention—and LLaMA 3's post-training pipeline including supervised fine-tuning, reward modeling, rejection sampling, and direct preference optimization.
DPOGrouped-Query AttentionLLM Architecture
0 likes · 10 min read

The article explains how to choose suitable large AI models for product use by focusing on three key aspects—model inference, post‑training optimization techniques such as quantization, pruning and distillation, and the role of GPUs, including their kernels, software stack, and computing platforms.
GPUPost-Traininginference
0 likes · 8 min read

This article explores the importance of post‑training for large language models, explains scaling laws for pre‑ and post‑training, details common fine‑tuning methods (full, PEFT, LoRA), outlines alignment techniques such as RLHF, DPO, PPO, and presents practical workflows using Llama 3 and DeepSeek‑R1, while also discussing test‑time reasoning optimizations.
LLMPost-TrainingRLHF
0 likes · 19 min read

This article explains post‑training concepts, compares SFT, DPO, and GRPO fine‑tuning methods, and provides step‑by‑step guidance for using Alibaba Cloud's PAI platform—including Model Gallery and DSW—to fine‑tune large language models with code examples and practical tips.
DPOGRPOLLM
0 likes · 14 min read

This article introduces Alibaba Cloud's PAI platform and demonstrates how open‑source reinforcement‑learning frameworks such as Cosmos‑RL, Ray, and VeRL accelerate post‑training for large language models, offering higher throughput, fault‑tolerance, and seamless integration for AI developers.
AI PlatformOpen Source FrameworksPost-Training
0 likes · 9 min read

This article provides a comprehensive technical overview of large language model post‑training, covering fine‑tuning methods (full, parameter‑efficient, LoRA families, prompt tuning), domain‑adaptive tuning, reinforcement‑learning reward modeling, process vs. outcome rewards, inference‑enhancement strategies, dynamic compute allocation, verifier‑augmented reasoning, current challenges, and emerging research directions such as meta‑cognition, physical reasoning, and swarm intelligence.
LLMPost-Trainingmeta-cognition
0 likes · 21 min read

The article analyzes how massive compute resources drive AI breakthroughs, highlighting Grok 3's top‑tier performance, DeepSeek's efficient engineering under constraints, and the emerging post‑training paradigm that reshapes competition among major AI players.
AI scalingCompute powerDeepSeek
0 likes · 7 min read

This article compiles and analyzes the post‑training pipelines of Llama 3.1, DeepSeek‑V3, TÜLU 3 and Qwen 2.5, detailing their data compositions, SFT, reward modeling, DPO, GRPO, RLVR methods, hyper‑parameters, and practical tricks for large‑language‑model alignment.
DPODeepSeek-V3Llama3.1
0 likes · 22 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


NewBeeNLPSep 23, 2024 · Artificial Intelligence
This article analyzes recent post‑training trends in large language models, comparing DPO and PPO, examining the scarcity of open‑source preference data, the iterative training process, the rise of synthetic data pipelines, and emerging methods for improving math and reasoning capabilities.
DPOLLMPPO
0 likes · 12 min read
