Tools

Search

Explore

Videos Channels Figures

Atmrix

About

Tools

Search

Explore

Videos Channels Figures

Atmrix

About

O

OpenAI Cast

04/11/25

@ OpenAI

Pre-training is inherently broad and diverse, making it challenging to achieve the same breadth when using reinforcement learning in specific environments.

Video

O

Pre-Training GPT-4.5

@ OpenAI

04/11/25

Related Takeaways

O

OpenAI Cast

04/11/25

@ OpenAI

Pre-training and unsupervised learning enhance the model's broad-based intelligence and improve its generalization capabilities, complementing reasoning abilities that can be more variable.

LF

Lex Fridman Cast

02/03/25

@ Lex Fridman

No matter how a model is designed to behave, the techniques used for reinforcement learning and preference tuning also enhance performance across various tasks.

MS

Murray Shanahan

06/29/17

@ Y Combinator

Deep reinforcement learning, which DeepMind is known for, allows systems to learn from scratch by trial and error, achieving superhuman performance in some tasks.

LF

Lex Fridman Cast

02/03/25

@ Lex Fridman

The idea behind reinforcement learning is that you can learn from very sparse rewards, which applies to both language generation and tasks for robots.

IF

Isa Fulford

05/08/25

@ Sequoia Capital

To develop Deep Research, we created reinforcement learning tasks to teach the model browsing and data analysis capabilities, and provided it access to tools for training.

MJ

Max Jaderberg

04/29/25

@ Sequoia Capital

Reinforcement learning allows us to tackle problems where we don't know the answers, focusing on whether the model's outputs are good or not, rather than needing predefined solutions.

SC

Sequoia Capital Cast

02/25/25

@ Sequoia Capital

Reinforcement learning is experiencing a resurgence due to advancements in language models, making it ripe for tuning to specific use cases with defined reward functions.

DS

David Silver

04/11/25

@ Google DeepMind

Human data might provide a head start for AI, but sustainable progress will depend on reinforcement learning as a continuous source of experience.

LF

Lex Fridman Cast

02/03/25

@ Lex Fridman

Even small models can boost their performance in tasks like math with a small amount of reinforcement learning training.