Troubleshooting Large Language Models with Amber Roberts ODSC's Ai X Podcast
-
- Technology
Despite the skill and effort that goes into creating LLMs, every data scientist will run into issues and problems that they will have to troubleshoot. But, what is the best way to choose the correct metrics, fix errors, and deal with hallucinations? In this episode of ODSC’s Ai X Podcast, the talented Amber Roberts of Arize AI, will help us answer these questions and take us on an exploration of interpretability tools like Phoenix.
Sponsored by: https://odsc.com/
Find more ODSC lightning interviews, webinars, live trainings, certifications, bootcamps here – https://aiplus.training/
Topics
1. Amber, can you introduce yourself and give tell us about your journey in Data Science and AI
2. Tell us about Arize AI and its mission.
3. What are the biggest challenges with large language models?
4. Why is troubleshooting large language models so difficult?
5. Describe your experience troubleshooting issues in large language models (LLMs), including open-source and proprietary models. What were the types of issues you encountered, and how did you approach troubleshooting them, especially in respect to the 9 parts of the LLMs?
6. How would you choose the right metrics for a specific LLM task to evaluate its performance?
7. Once we've identified an error in an LLM, how do we go about fixing it such as updating the model with new or corrected information?
8. Tell us about what constitutes a hallucination, what Causes Hallucinations and are there specific techniques or tools that can identify when a model is producing (unobvious) hallucinations?
9. The development of interpretability tools for LLMs is still in the early stages, but tell us about Phoenix?
10. How does Phoenix help the evaluation, troubleshooting, and fine-tuning of large language models (LLMs)?
11. Tell us about some of the tools Phoenix integrates with such as "llama-index" and LangChain?
12. Can you explain the key features of Phoenix and how they contribute to the effective management of machine learning models in production?
13. What types of challenges in AI model monitoring and observability does Phoenix address, and what solutions does it offer?
14. Looking ahead, what are the future development plans for Phoenix?
15. This is an incredibly exciting time to be in AI. Do you feel the path to a career in AI has changed and what advice would you give?
Some useful links:
Phoenix: AI Observability & Evaluation - docs.arize.com/phoenix/
Community Paper Reading: arize.com/resource/community-papers-reading/
Latest Arize Workshop: arize.com/resource/rag-time/
Arize docs: docs.arize.com/arize/
Vector DB Comparison - vdbs.superlinked.com/
Connect with Amber over Linkedin - www.linkedin.com/in/amber-roberts42/
Despite the skill and effort that goes into creating LLMs, every data scientist will run into issues and problems that they will have to troubleshoot. But, what is the best way to choose the correct metrics, fix errors, and deal with hallucinations? In this episode of ODSC’s Ai X Podcast, the talented Amber Roberts of Arize AI, will help us answer these questions and take us on an exploration of interpretability tools like Phoenix.
Sponsored by: https://odsc.com/
Find more ODSC lightning interviews, webinars, live trainings, certifications, bootcamps here – https://aiplus.training/
Topics
1. Amber, can you introduce yourself and give tell us about your journey in Data Science and AI
2. Tell us about Arize AI and its mission.
3. What are the biggest challenges with large language models?
4. Why is troubleshooting large language models so difficult?
5. Describe your experience troubleshooting issues in large language models (LLMs), including open-source and proprietary models. What were the types of issues you encountered, and how did you approach troubleshooting them, especially in respect to the 9 parts of the LLMs?
6. How would you choose the right metrics for a specific LLM task to evaluate its performance?
7. Once we've identified an error in an LLM, how do we go about fixing it such as updating the model with new or corrected information?
8. Tell us about what constitutes a hallucination, what Causes Hallucinations and are there specific techniques or tools that can identify when a model is producing (unobvious) hallucinations?
9. The development of interpretability tools for LLMs is still in the early stages, but tell us about Phoenix?
10. How does Phoenix help the evaluation, troubleshooting, and fine-tuning of large language models (LLMs)?
11. Tell us about some of the tools Phoenix integrates with such as "llama-index" and LangChain?
12. Can you explain the key features of Phoenix and how they contribute to the effective management of machine learning models in production?
13. What types of challenges in AI model monitoring and observability does Phoenix address, and what solutions does it offer?
14. Looking ahead, what are the future development plans for Phoenix?
15. This is an incredibly exciting time to be in AI. Do you feel the path to a career in AI has changed and what advice would you give?
Some useful links:
Phoenix: AI Observability & Evaluation - docs.arize.com/phoenix/
Community Paper Reading: arize.com/resource/community-papers-reading/
Latest Arize Workshop: arize.com/resource/rag-time/
Arize docs: docs.arize.com/arize/
Vector DB Comparison - vdbs.superlinked.com/
Connect with Amber over Linkedin - www.linkedin.com/in/amber-roberts42/