Uploaded April 2023 | Updated November 2026, 3 hours ago
Large language models have taken the world by storm, but we're still learning what they do and how they work. In this conversation, UC Berkeley Professor & Aqueduct co-founder explains what LLMs do, how they work, and how the space has evolved over recent years.
Aqueduct: github.com/aqueducthq/aqueduct
Aqueduct on Twitter: twitter.com/aqueducthq
Joey's Twitter: twitter.com/profjoeyg
Vikram's Twitter: twitter.com/vsreekanti
0:00 Intro
0:57 What are LLMs?
3:08 How do LLMs predict what comes next?
4:58 What does the number of parameters in a model mean?
7:50 What do LLM parameters represent?
9:24 How much randomness is in an LLM generation?
11:30 Do embedding spaces matter?
13:24 Is the era of large models over?
15:17 Can we run LLMs on the edge?
17:22 Conclusion
Large language models have taken the world by storm, but we're still learning what they do and how they work. In this conversation, UC Berkeley Professor & Aqueduct co-founder explains what LLMs do, how they work, and how the space has evolved over recent years.
Aqueduct: github.com/aqueducthq/aqueduct
Aqueduct on Twitter: twitter.com/aqueducthq
Joey's Twitter: twitter.com/profjoeyg
Vikram's Twitter: twitter.com/vsreekanti
0:00 Intro
0:57 What are LLMs?
3:08 How do LLMs predict what comes next?
4:58 What does the number of parameters in a model mean?
7:50 What do LLM parameters represent?
9:24 How much randomness is in an LLM generation?
11:30 Do embedding spaces matter?
13:24 Is the era of large models over?
15:17 Can we run LLMs on the edge?
17:22 Conclusion










