Tools

Explore

Videos Channels Figures

Atmrix

Tools

Explore

Videos Channels Figures

Atmrix

O

OpenAI Cast

04/11/25

@ OpenAI

The GPT architecture excels at efficiently absorbing and generalizing data, but as compute resources grow, data becomes the bottleneck, necessitating algorithmic innovations to learn more from the same amount of data.

Video

Pre-Training GPT-4.5

@ OpenAI

Pre-Training GPT-4.5

04/11/25

Related Takeaways

Geoffrey Hinton

01/18/25

@ Curt Jaimungal

GPT-4's extensive knowledge comes from multiple copies running on different hardware, allowing them to share learning efficiently, unlike human brains.

Y Combinator Cast

03/11/25

@ Y Combinator

GPT-4.5 is potentially more than 10 times the size of GPT-4, representing a step forward in both pre-training and post-training processes.

OpenAI Cast

04/11/25

@ OpenAI

The defining characteristics of the GPT paradigm are that you can predict the test loss, and it scales predictably, leading to greater intelligence in various intangible ways.

Y Combinator Cast

01/24/25

@ Y Combinator

The longer models like GPT-3 are able to think through complex problems, the better they perform.

Y Combinator Cast

11/16/23

@ Y Combinator

GPT technology is underhyped; its capabilities in reading, understanding, and processing legal documents can revolutionize the industry.

Y Combinator Cast

01/24/25

@ Y Combinator

GPT-3 represents a huge leap forward, not just a small improvement over its predecessors, and researchers believe this trajectory will continue.

OpenAI Cast

04/11/25

@ OpenAI

The sheer volume of work required for GPT-4.5 was a byproduct of the model's specifications, necessitating changes that couldn't be achieved with the same stack used for GPT-4.

Y Combinator Cast

03/11/25

@ Y Combinator

GPT-4.5 shows that scaling unsupervised learning continues to yield valuable improvements in accuracy, emotional intelligence, and creativity, even if these gains are now more incremental than before.

Sam Altman

06/18/25

@ OpenAI

The reasoning models push the capabilities of GPT models further, allowing for better performance by thinking step by step.