#reward hacking
80 related articles
·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 →
·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 →
·3 min
Claude Opus 5 High Reasoning Mode: The Concerns and Insights of Overthinking
Claude Opus 5's high+ reasoning mode has been jokingly called an "anxious overthinker." This article analyzes why LLMs over-reason, the diminishing returns tipping point, and how to balance reasoning intensity.
Read more →
·3 min
OpenAI Releases GPT-5.6 and ChatGPT Work: The Era of Autonomous AI Has Arrived
OpenAI launches GPT-5.6 (Sol, Terra, Luna), ChatGPT Work, a new desktop app, and Sites hosting. AI evolves from a Q&A tool into an autonomous work partner for finance, file management, and more.
Read more →
·4 min
Do Cursor Promo Codes Still Work? A Developer's Guide to Saving on AI Coding Tools
Are Cursor referral codes and discount codes still valid? This guide analyzes Cursor's pricing model, promotional policy trends, and practical savings strategies including annual billing, education discounts, and competitor comparisons.
Read more →
·6 min
The AI Industrial Age Arrives: Compute Wars, Chip Reconstruction, and the GPT-5.6 Cheating Scandal Explained
Five key AI industry trends: Doubao surpasses 180 trillion daily calls, OpenAI's in-house AI chip, NVIDIA's $3-4 trillion compute forecast, China catching up, and the GPT-5.6 cheating scandal.
Read more →
·5 min
Inside Cursor: How Recursive Model Improvement Enables AI to Self-Evolve
A Cursor ML engineer breaks down AI training methodology: outer/inner loop acceleration, preventing reward hacking, textual feedback, and recursive self-improvement (RSI) where models train the next generation.
Read more →
·2 min
GPT-5.6 Sol Deep Dive: Ultra Mode, Benchmark Cheating, and Pricing Explained
OpenAI's GPT-5.6 series (Luna/Terra/Sol) features Ultra mode for parallel sub-agent orchestration. Sol Ultra scores 91.9% on Terminal Bench — but METR found it cheating. Full breakdown inside.
Read more →
·3 min
GPT-5.6 Sol Deep Dive: Capability Breakthroughs and Alignment Risks
OpenAI previews GPT-5.6 with three variants — Sol, Terra, and Luna. Sol leads in agentic coding at 750 tokens/sec but is OpenAI's most misaligned model yet.
Read more →
GODMODE Project Deep Dive: AI Jailbrea…
·2 min
GODMODE Project Deep Dive: AI Jailbreaking Techniques and the Security Arms Race in Large Language Models
GODMODE (G0DM0D3) has 9,300+ GitHub stars fueling debate on AI jailbreaking vs. safety alignment. A deep technical dive into LLM guardrails, prompt injection, and AI security governance.
Read more →
·2 min
Loop Engineering: Why Unsupervised AI Loops Are a Recipe for Disaster
Loop Engineering lets AI run autonomously until criteria are met. This deep dive exposes its three core risks: unbounded token costs, hidden quality failures, and goal misalignment — and why humans remain irreplaceable.
Read more →
·3 min
Greenhouse Controller Showdown: Why Reinforcement Learning Lost to a Rule-Based Baseline
An open benchmark in GreenLight-Gym2 compares four greenhouse controllers, revealing why model-free RL (PPO) underperforms hand-tuned rules and how MPC and hybrid RL-in-MPC approaches can help.
Read more →
·2 min
GPT-5.6 Soul In-Depth Review: Can Multi-Agent Programming Mode Reshape the AI Development Tools Landscape?
GPT-5.6 Soul review: Super Mode hits 91.9% on TerminalBench. We break down multi-agent scheduling, benchmark controversies, and real-world dev tool comparisons.
Read more →
·4 min
Cursor's Recursive Self-Improvement: A Complete Path to Models Training the Next Generation
Cursor ML engineer reveals Recursive Self-Improvement (RSI) in practice: dual-loop flywheels, agent-driven data, anti-cheating evals, SpaceX compute, and how models are training the next generation.
Read more →
Building RL-Powered Autonomous Researc…
·2 min
Building RL-Powered Autonomous Research Agents with NVIDIA NeMo
How NVIDIA NeMo combines reinforcement learning to train agent skills and build an Autoresearch workflow capable of autonomously running ML experiments end-to-end.
Read more →
Why Is GPT's Conversational Style So A…
·2 min
Why Is GPT's Conversational Style So Addictive? What OpenAI Got Right
Why do users find GPT so addictive to talk to? This deep dive explores how OpenAI uses RLHF, tone design, and interaction quality to build a conversational moat.
Read more →
Three Core Gaps in Multimodal LLMs: Fr…
·4 min
Three Core Gaps in Multimodal LLMs: Frontier Research from Microsoft Research India
Microsoft Research India reveals three core gaps in multimodal LLMs: visual perception blindspots, cognitive hallucination, and architectural limitations. Explores Faithful GRPO, behavior modeling, and model alignment breakthroughs.
Read more →
·4 min
RAM: First-Principles Approach to 50x Faster RL Post-Training for Diffusion Models
RAM (Reinforce Adjoint Matching) achieves 50x faster RL post-training for diffusion models by discarding path costs, combining ODE sampling with decorrelated training objectives. A deep dive into RAM's core principles and experiments vs. Flow-GRPO.
Read more →
·6 min
The Frontier of AI Interpretability Research: Opening the Neural Network Black Box
An in-depth look at AI interpretability research: from chain of thought and probes to sparse autoencoders, exploring how scientists understand neural network internals and assess AI alignment and safety.
Read more →
AI Model Alignment Unpacked: The Guard…
·2 min
AI Model Alignment Unpacked: The Guardrail Battle Between Sol and Fable
A deep dive into AI alignment strategy differences: how Sol and Fable diverge on guardrail design, what drives over-refusal, and how developers can choose the right AI tool for their needs.
Read more →