Here's something most people don't realize: every time you use ChatGPT, you're using…

reinforcement learning.

RLHF — Reinforcement Learning from Human Feedback — is what makes LLMs actually helpful instead of just statistically completing text. It's RL that transforms a base model into something that follows instructions, refuses harmful requests, and gives useful answers.

And RL is having a moment beyond LLMs too. Robotics companies are using it for dexterous manipulation. Trading firms use it for strategy optimization. Drug discovery teams use it to explore molecular spaces.

The field went through a "trough of disillusionment" after the initial AlphaGo hype. But now it's quietly powering some of the most important systems in AI.

The catch: RL is genuinely hard. The concepts (PPO, policy gradients, reward shaping) are less intuitive than supervised learning. Training is unstable. Debugging is painful.

But that difficulty is exactly why RL engineers are rare and valuable. If you have the appetite for a steep learning curve with a big payoff, RL is wide open.