The evolution of training AI models

25 Aug 2025 · 1 min
1:04

AI training has two stages: pre-training on internet data, then post-training where labs use domain experts in fields like medicine, law, and math to refine models. As models got better, the bottleneck shifted from gathering more data to getting specialized knowledge from experts.

Chapters

More from Lenny

0:50
youtube.com

The internet is going to agents

3d ago
1:02:12
youtube.com

Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)

4d ago
0:51
youtube.com

Autonomy Only Works If You Own the Mistakes

4d ago
0:35
youtube.com

AEO Won't Get Your Plugin Recommended

5d ago
0:44
youtube.com

AI Loops Actually Aren't The Future

6d ago
0:42
youtube.com

OpenAI has more than 120 ex-YC founders

1w ago
0:43
youtube.com

The AI Work Transformation Isn't Over

1w ago
0:40
youtube.com

His Agent Caught the Outage Before the Demo

1w ago
0:36
youtube.com

The Best AI Gets Out of the Way

1w ago
37:23
youtube.com

Why most actions on the internet will soon be taken by AI (OpenAI’s head of ChatGPT, Tibo Sottiaux)

1w ago
0:47
youtube.com

Shipping More Isn't Ambition Anymore

1w ago
See all content