Skip to main content

AGI Breakthroughs Are Already Here: OpenAI's Bob McGrew on AI's Future

[HPP] Stephanie ZhanJune 17, 202548 min
35 connections·40 entities in this video→
Capture as you watch

Save any video to veridive in one click.

The free veridive Chrome extension pulls the transcript from any YouTube video or podcast you're watching — ready to ask, cite, and connect.

Add to Chrome

The AI Trifecta and AGI's Foundations

  • 💡 Bob McGrew identifies three core components of AI: pre-training, post-training, and reasoning.
  • 🔑 He controversially argues that the fundamental concepts for AGI (Transformers, scaled pre-training, reasoning, and multimodality) have already been discovered.
  • 🎯 2025 is predicted to be the "year of reasoning" due to rapid progress and low-hanging fruit, such as the development of tool use in models.

Evolution of AI Training

  • 📈 Pre-training is experiencing diminishing returns due to the log-linear relationship between intelligence and compute, but remains vital for improving architectures like efficiency and context handling.
  • 🧠 Post-training focuses on "model personality", which is a
Ask, don't scrub

Discover the spoken web.

veridive answers questions with exact timestamps and citations — across every podcast, video, and article you've saved.

Open Chat
Knowledge graph40 entities · 35 connections

How they connect

An interactive map of every person, idea, and reference from this conversation. Hover to trace connections, click to explore.

Hover · drag to explore
40 entities
Chapters19 moments

Key Moments

Transcript179 segments

Full Transcript

Follow the thread

Find every place these ideas show up.

veridive maps the same people, claims, and topics across thousands of sources — so you can trace an idea from one conversation to the next.

Explore more
Topics15 themes

What’s Discussed

Artificial General Intelligence (AGI)TransformersScaled Pre-trainingReasoning ModelsMultimodal CapabilitiesModel PersonalityAI AgentsCompute CostsRoboticsProprietary DataCodingSoftware EngineeringCybersecurityLanguage Models (LLMs)Vision Encoders
Smart Objects40 · 35 links
Companies· 3
Concepts· 23
People· 4
Products· 9
Event· 1
veridive

Hours of content, seconds to the answer.

Save what you listen to. Ask it anything. Watch the threads between sources surface on their own.

Get started free