Why I don’t think AGI is right around the corner
Related videos
Chip design from the bottom up – Reiner Pope
Dwarkesh Patel
25.6k views
Something isn’t right in the inner circle
David Pakman Show
51.9k views
Why is Kash Patel’s Girlfriend Posting Insane Gibberish?
The Bulwark
28.4k views
Dylan Patel — The Single Biggest Bottleneck to Scaling AI Compute
Dwarkesh Patel
58.6k views
The Department of War is making a huge mistake.
Dwarkesh Patel
35.5k views
Dario Amodei — “We are near the end of the exponential”
Dwarkesh Patel
41.0k views
What are we scaling?
Dwarkesh Patel
22.3k views
Sarah Paine – Why Russia Lost the Cold War
Dwarkesh Patel
136.4k views
Ilya Sutskever – We're moving from the age of scaling to the age of research
Dwarkesh Patel
55.6k views
Satya Nadella – How Microsoft is preparing for AGI
Dwarkesh Patel
55.1k views
Evaluating AI Timelines Versus Continual Learning Bottlenecks
Understand why the lack of continuous, human-like learning in current Large Language Models significantly extends projected timelines for transformative AI adoption in white-collar work.
Short Summary
- Current LLMs fail to learn organically on the job, limiting immediate white-collar transformation despite high raw capability.
- The necessity of deliberate teaching (system prompts, specialized RL frameworks) drastically slows process improvement compared to human iterative learning.
- The author forecasts longer timelines (2028 for tax processing, 2032 for human-level job adaptation) than some industry experts, largely due to this continual learning hurdle.
- Solving on-the-job learning predicts a massive discontinuity in AI value, potentially triggering an intelligence explosion later this decade or next.
This segment outlines the author's disagreement with bullish near-term AI projections based on critical missing capabilities in today’s models. The focus centers on why the inability of LLMs to build context and iterate efficiently prevents them from functioning like true employees right now, even if algorithmic breakthroughs follow post-2030.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Related videos
Chip design from the bottom up – Reiner Pope
Dwarkesh Patel
25.6k views
Something isn’t right in the inner circle
David Pakman Show
51.9k views
Why is Kash Patel’s Girlfriend Posting Insane Gibberish?
The Bulwark
28.4k views
Dylan Patel — The Single Biggest Bottleneck to Scaling AI Compute
Dwarkesh Patel
58.6k views
The Department of War is making a huge mistake.
Dwarkesh Patel
35.5k views
Dario Amodei — “We are near the end of the exponential”
Dwarkesh Patel
41.0k views
What are we scaling?
Dwarkesh Patel
22.3k views
Sarah Paine – Why Russia Lost the Cold War
Dwarkesh Patel
136.4k views
Ilya Sutskever – We're moving from the age of scaling to the age of research
Dwarkesh Patel
55.6k views
Satya Nadella – How Microsoft is preparing for AGI
Dwarkesh Patel
55.1k views
Top Comments (10)
AI is like having an intern, but it is also like having an intern.
Maybe the real AGI were the LLMs we met along the way
I don't think we need AGI for AI to make an impact. More AlphaFold type systems would be awesome. And DeepMind is already doing it
I've been waiting so long for Dwarkesh to come on this podcast
The central problem is like trying to predict what the web will look like in 2002 based on knowledge of the databases and filesystems of 1992.
"A recent survey of 475 AI researchers reveals that 76% believe adding more computing power and data to current AI models is “unlikely” or “very unlikely” to lead to AGI".
The question is: looking back, when could someone have foreseen the creation of Transformers?
Even your longer timelines are still wild in the grand schema of things. We're 5-15 years from a total revolution in how we work as a society.
Man I’ve not really liked dwarkesh because sometimes he seems like he’s peddling hype and I’ve secretly presumed it’s because he is financially incentivized (like everyone else in tech right now) to create as much hype as possible. But this more calm measured approach is actually a breath of fresh air from what I would consider the more hype type podcasts. Respect.
If any of you has ever raised babies and you have taken care of all their needs and your house, raising your children since they were born, detecting the thousands of possible problems that may arise and must be solved daily, flexibly and in a context dependent way, and you know how you solved them, safely, buying food and everything you need, cooking, cleaning the house, repairing the things that constantly get broken... You won't see an autonomous system or a humanoid robot able to perform all those tasks at the same level of dexterity, safety, and common sense way that you did all those years, before 2050 (maybe you'll have to wait much longer).
Unlock the Data Inside
Turn Videos into Knowledge
- Get FREE 10/day: transcripts, summaries, chats
- Chat with videos, export text & PDF
- $1 free API credit for RAG, chatbots & research
Free forever plan • All features unlocked
Top Comments (10)
AI is like having an intern, but it is also like having an intern.
Maybe the real AGI were the LLMs we met along the way
I don't think we need AGI for AI to make an impact. More AlphaFold type systems would be awesome. And DeepMind is already doing it
I've been waiting so long for Dwarkesh to come on this podcast
The central problem is like trying to predict what the web will look like in 2002 based on knowledge of the databases and filesystems of 1992.
"A recent survey of 475 AI researchers reveals that 76% believe adding more computing power and data to current AI models is “unlikely” or “very unlikely” to lead to AGI".
The question is: looking back, when could someone have foreseen the creation of Transformers?
Even your longer timelines are still wild in the grand schema of things. We're 5-15 years from a total revolution in how we work as a society.
Man I’ve not really liked dwarkesh because sometimes he seems like he’s peddling hype and I’ve secretly presumed it’s because he is financially incentivized (like everyone else in tech right now) to create as much hype as possible. But this more calm measured approach is actually a breath of fresh air from what I would consider the more hype type podcasts. Respect.
If any of you has ever raised babies and you have taken care of all their needs and your house, raising your children since they were born, detecting the thousands of possible problems that may arise and must be solved daily, flexibly and in a context dependent way, and you know how you solved them, safely, buying food and everything you need, cooking, cleaning the house, repairing the things that constantly get broken... You won't see an autonomous system or a humanoid robot able to perform all those tasks at the same level of dexterity, safety, and common sense way that you did all those years, before 2050 (maybe you'll have to wait much longer).