EP. 149 · 14.07.2026 · 59:48
A PRIMER ON LARGE LANGUAGE MODELS
with Andrej Karpathy
Auto-generated brief
EPISODE BRIEF — EP. 149 · A PRIMER ON LARGE LANGUAGE MODELS Guest: Andrej Karpathy · Published: 14.07.2026 · Duration: 59:48 【OVERVIEW】 The clearest one-hour explanation of how large language models actually work: pretraining and finetuning, what the weights really are, scaling laws, tool use, and the emerging security surface of models that read untrusted text. 【KEY TAKEAWAYS】 1. What a language model actually is 2. Pretraining, finetuning and the assistant 3. Scaling laws and the LLM operating system 4. Jailbreaks, prompt injection and model security 【CHAPTERS】 00:00 — What a language model actually is 10:00 — Pretraining, finetuning and the assistant 25:00 — Scaling laws and the LLM operating system 42:00 — Jailbreaks, prompt injection and model security 【TOPICS】LLM · Fundamentals · Education · Pretraining · Alignment 【WATCH】https://thedaopod.com/episodes/ep-149-intro-to-large-language-models
SUBSCRIBE
Subscribe to The DAO
Get every dialogue, roundtable and observatory briefing in your podcast app — or copy the RSS feed and add it manually.
Subscription status unknown
We can't read your YouTube subscription directly — confirm it manually, and redo it anytime.
Follow us
https://thedaopod.com/rss.xmlABOUT THIS EPISODE
THE CONVERSATION
The clearest one-hour explanation of how large language models actually work: pretraining and finetuning, what the weights really are, scaling laws, tool use, and the emerging security surface of models that read untrusted text.
SHARE
Copy link & social preview
thedaopod.com
EP. 149 · A PRIMER ON LARGE LANGUAGE MODELS
The clearest one-hour explanation of how large language models actually work: pretraining and finetuning, what the weights really are, scaling laws, tool use, and the emerging security surface of models that read untrusted text.
Pasting this link on X, WeChat or LinkedIn renders the preview card above.
CHAPTERS
TIMELINE
Chapters derived from show notes
GUESTS
ON THE ROUNDTABLE
Founding member, OpenAI · Former Director of AI, Tesla
A founding member of OpenAI and former Director of AI at Tesla. His long-form lectures are widely regarded as the best available introduction to how modern language models are trained and deployed.
No subtitle track for this video — the timeline below is generated from chapters and follows playback
FULL TRANSCRIPT
Searchable Transcript
Chapter-derived timeline
0:00 · What a language model actually is
Key Moments
4 lines
Related videos
Matched by host & keywords
EpisodeEP. 152 · BUILDING WITH LLMS: CRAFT, TOOLING AND THE NEW STACK
with Andrej Karpathy
EpisodeEP. 171 · USING AI TO INCREASE INTELLIGENCE AND ENRICH HUMANITY
with Fei-Fei Li
EpisodeEP. 168 · JOBS, ROBOTS AND WHY WORLD MODELS ARE NEXT
with Fei-Fei Li
EpisodeEP. 147 · SCALING, SEARCH AND WHAT COMES AFTER THE TRANSFORMER
with Demis Hassabis
EpisodeEP. 146 · A DECADE OF DEEP LEARNING, AND WHAT COMES NEXT
with Ilya Sutskever
EpisodeEP. 125 · SOFTWARE 2.0, LLMS, AND THE FUTURE OF CODING
with Andrej Karpathy