Uploaded May 2023 | Updated November 2026, 3 hours ago
LLMs are incredibly powerful, but they fail in shocking and hilarious ways from time to time. What's going on?
In this episode, we dive into the typical failure modes (you've probably heard of hallucinations), the types of things LLMs and foundation models can't quite grasp, and areas researchers are actively working on to improve LLMs.
Let us know what you think, and let us know what else you'd like us to cover! Interested in using LLMs? See how Aqueduct can help: aqueducthq.com/post/the-easiest-way-to-run-open-llms
Aqueduct: github.com/aqueducthq/aqueduct
Twitter: twitter.com/aqueducthq
Joey's Twitter: twitter.com/profjoeyg
Vikram's Twitter: twitter.com/vsreekanti
0:00 Intro
0:35 How and why do LLMs fail?
2:49 How do computer vision models fail?
3:55 How does bias creep into foundation models?
4:31 Why can't LLMs handle structured data?
5:54 Do we need vector DBs to make LLMs useful?
8:20 Is human feedback required for LLMs?
8:54 Conclusion
LLMs are incredibly powerful, but they fail in shocking and hilarious ways from time to time. What's going on?
In this episode, we dive into the typical failure modes (you've probably heard of hallucinations), the types of things LLMs and foundation models can't quite grasp, and areas researchers are actively working on to improve LLMs.
Let us know what you think, and let us know what else you'd like us to cover! Interested in using LLMs? See how Aqueduct can help: aqueducthq.com/post/the-easiest-way-to-run-open-llms
Aqueduct: github.com/aqueducthq/aqueduct
Twitter: twitter.com/aqueducthq
Joey's Twitter: twitter.com/profjoeyg
Vikram's Twitter: twitter.com/vsreekanti
0:00 Intro
0:35 How and why do LLMs fail?
2:49 How do computer vision models fail?
3:55 How does bias creep into foundation models?
4:31 Why can't LLMs handle structured data?
5:54 Do we need vector DBs to make LLMs useful?
8:20 Is human feedback required for LLMs?
8:54 Conclusion










