#rlhf

Hello, I'm Rijul, and I'm building LiveReview — a blast-radius aware AI code review built for your...

What data annotation jobs and AI response review are: the tasks involved, how they feed model training (RLHF), and how annotator quality is measured.

从 RLHF 到 Constitutional AI，从辩论方案到递归奖励建模，系统梳理 AI 安全对齐的技术路线与开放挑战。

Large language models don't become aligned simply because they become larger. They become useful when...

Reward models power RLHF-style training and are genuinely complex to build well. A plain-language explanation, and when a student project actually nee

RLHF Burned $50K Before We Admitted SFT Would've Worked Reinforcement Learning from Human...

MTurk is closing to new customers: 6 human-data alternatives compared for 2026 Summary....

The largest audit of AI language model judges to date — 21 judges, over half a million grading decisions — finds that standard reliability metrics are inflated by roughly a third, that the same judge can score differently on different benchmarks, and

Several research groups landed on the same idea at once - improve a model by learning from its own attempts instead of expensive human labels - and the field is debating whether it really removes the labeling burden or just hides it.

How RLHF-trained language models may develop instrumental goals, and the information-theoretic limits on detecting them.

The DPO Hype Promised to Kill RLHF — It Didn't Everyone said Direct Preference...

The $12,000 Surprise RLHF training for a 7B parameter model ran us $12,400 on AWS for...