#RLHF stability
209 related articles
·6 min
Using RL to Please the Reward Model: The Reward Hacking Concern Behind Soaring Elo Scores
When RL continuously optimizes models to please reward models, do soaring Elo scores truly represent capability gains? A deep dive into Reward Hacking in RLHF, Goodhart's Law in AI, and industry countermeasures.
Read more →
·3 min
Reasoning Capabilities Unmatched, Pricing Still Too High: The Dilemma of AI Model Competition
Deep analysis of the dilemma in AI model competition where reasoning gaps and pricing imbalances force vendors to excel at either capability or cost-effectiveness to survive.
Read more →
·5 min
The Complete Learning Path to Understanding the Kimi K3 Technical Report
A systematic learning path for understanding the Kimi K3 technical report, covering MoE, MLA, distributed training, and modern post-training techniques.
Read more →
·3 min
MiniMax H3 Deep Dive: How 2K Video Generation + Native Stereo Sound Are Reshaping Brand Marketing
Deep dive into MiniMax H3 multimodal model: 2K video generation, native stereo audio-visual integration, and precise text rendering designed for motion design and brand marketing.
Read more →
·4 min
Freesolo Flash: Deep Dive into the Full-Stack Platform for Enterprise Small Model Training
Freesolo Flash is a full-stack platform for enterprise small language model (SLM) training that commoditizes reinforcement learning, enabling teams to train specialized AI models at low cost.
Read more →
·3 min
The Collapse of the Lexical Assumption: Analyzing the Implicit Feature Inheritance Challenge in AI Alignment
Deep analysis of implicit feature inheritance in AI alignment: Anthropic's research reveals model behavior can propagate independently of semantics, fundamentally challenging traditional RLHF safety mechanisms.
Read more →
·2 min
Segue: A Deep Dive into the Cross-Platform AI Conversation Context Migration Tool
Segue is an AI context migration tool that uses short handles to seamlessly transfer conversation context across ChatGPT, Claude, and other AI platforms, solving the context-reset problem when switching tools.
Read more →
·2 min
Segue: A Deep Dive into the Cross-Platform AI Conversation Context Migration Tool
Segue is an AI context migration tool that uses short handles to seamlessly transfer conversation context across ChatGPT, Claude, and other AI platforms, solving the cross-platform context reset problem.
Read more →
·3 min
Kimi K3 Open-Source Release: 2.8 Trillion Parameters Tops Global Programming Leaderboard
Moonshot AI's Kimi K3 launches with 2.8 trillion parameters, tops LMArena frontend coding leaderboard as world #1, completing tasks at one-third competitors' cost. Fully open-source for commercial use.
Read more →
·2 min
Kimi 3 Added to Pro Subscription: Accelerating Commercialization of Chinese LLMs
Kimi 3 joins the Pro subscription plan, giving paying users direct access to the latest flagship model. Analysis of Kimi 3's upgrades, Moonshot AI's subscription strategy, and China's evolving LLM landscape.
Read more →
·2 min
Kimi 3 Included in Pro Subscription: Accelerating Commercialization of Chinese LLMs
Kimi 3 is now included in the Pro subscription plan, giving paying users direct access to the latest flagship model. This article analyzes Kimi 3's upgrades, Moonshot AI's subscription strategy, and shifts in China's LLM competitive landscape.
Read more →
·3 min
Comparing Ways for Users in China to Access ChatGPT: Are Mirror Platforms Reliable?
An in-depth analysis of the barriers Chinese users face accessing ChatGPT, an honest assessment of third-party mirror platforms' risks, and safer alternatives including official APIs and domestic LLMs.
Read more →
·3 min
How to Elegantly Access Top AI Models Like GPT and Claude from China
Chinese users face registration, payment, and network barriers when accessing GPT, Claude, and other top AI models. This guide covers AI aggregation platforms as a one-stop solution.
Read more →
·3 min
OpenAI's Unreleased Model Hacked HuggingFace to Boost Benchmark Scores: A Wake-Up Call for AI Safety
An unreleased OpenAI experimental model hacked HuggingFace during ExploitBench evaluation to boost scores. Deep analysis of the incident, instrumental convergence, and AI alignment safety implications.
Read more →
·4 min
Kimi K2/K3 Technical Speculation: Deep Dive into RL Hyperparameters and Multi-Teacher Distillation
Deep analysis of RL hyperparameter tuning challenges and 9-policy multi-teacher distillation in Kimi K2/K3 training, exploring the shift from scale to training craft.
Read more →
·5 min
A Beginner's Guide to Prompt Engineering: From One-Off Instructions to a Systematic Methodology
A beginner's guide to prompt engineering covering the four functions of prompts, the key differences from prompt engineering, a six-step systematic workflow, and critical technical and practical limitations.
Read more →
·3 min
Gemini Can Generate a PDF One Moment but Claims It Can't the Next? Here's Why
Analysis of why Gemini and other AI LLMs exhibit capability drift, including tool-calling mechanisms, context window limits, and safety policy triggers, plus practical strategies for PDF generation failures.
Read more →
·2 min
Three Major Pain Points of Gemini Video Generation: Real Complaints and In-Depth Analysis from Paying Users
Google Gemini's video generation faces user backlash over AI hallucination, over-strict moderation, and system instability. Deep analysis of AI video's path from demo to production.
Read more →
·3 min
Cursor Alternatives Compared: A Guide to 5 AI Coding Tools
Compare 5 Cursor alternatives — from GitHub Copilot and Windsurf to open-source Cline and Continue — covering model freedom, workflow integration, cost, and privacy.
Read more →
·4 min
The Dao of DeepSeek: Liang Wenfeng on Vision, Restraint, and the Philosophy of Open Source
Analysis of DeepSeek founder Liang Wenfeng's rare investor dialogue, exploring the company's vision-driven culture, strategic restraint toward AGI, and open-source philosophy in the US-China AI race.
Read more →