Uploaded June 2022 | Updated September 2026, 4 hours ago

---

Artificial General Intelligence (AGI) is popularly defined as AI that can do anything a human can do. More specifically in this video I talk about my thoughts on when we will reach human level AI. Given recent successes in Machine Learning, and especially with scaling Deep Learning, there has been a lot of talk about AGI recently. I have heard a lot of wild opinions, so with the video I try to take a research-centric approach, talking about results from recent papers and models like Gopher, Chinchilla, Gato, "Scaling Laws for Neural Language Models," and more. I talk about this in the context of other recent works like GPT-3 and Dall-E 2.

Outline:
0:00 - Intro & AGI Buzz
1:41 - What is AGI?
2:13 - Pro AGI Argument
4:09 - Scaling Isn't Enough
6:51 - "Level 2 Intelligence"
9:14 - Counterarguments
12:12 - Supervised Learning Isn't Enough
13:40 - An Optimistic Perspective

Sources (other than the ones cited in the video)
Refrenced AI/AGI articles:
venturebeat.com/2021/06/26/deepmind-agi-paper-adds-urgency-to-ethical-ai
venturebeat.com/2022/06/04/is-deepminds-gato-the-worlds-first-agi
thenextweb.com/news/deepminds-astounding-new-gato-ai-makes-fear-humans-will-never-achieve-agi
technologyreview.com/2022/05/23/1052627/deepmind-gato-ai-model-hype

Image segmentation example: towardsdatascience.com/image-segmentation-with-six-lines-0f-code-acb870a462e8
Trackmania footage: youtube.com/watch?v=dLKAlWHn2vo