3min chapter

Machine Learning Street Talk (MLST) cover image

#114 - Secrets of Deep Reinforcement Learning (Minqi Jiang)

Machine Learning Street Talk (MLST)

CHAPTER

How to Train a Language Model to Maximize Reward

When you train a typical language model, you're basically training it with something like a cross entropy loss. And so when this process converges, when you minimize this loss, it should actually be matching the distribution of text on the internet. What reinforcement learning does is almost the opposite of this reinforce learning is not doing distribution matching. It's essentially introducing mode seeking biases into your model. So maybe now if it's tuned to give me good answers on math questions, now if I ask it a calculus question, it'll tend to favor those completions that are modeling the outputs of a college professor of math rather than someone who's like asking the same question on Reddit and saying, help, I

00:00

Get the Snipd
podcast app

Unlock the knowledge in podcasts with the podcast player of the future.
App store bannerPlay store banner

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode

Save any
moment

Hear something you like? Tap your headphones to save it with AI-generated key takeaways

Share
& Export

Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode