Exploring ChatGPT for Improved Observability
summary
Claims ChatGPT (GPT-3.5) is useful for observability but has limits: it’s fine-tuned with about 175 billion parameters, uses reinforcement learning from human feedback, and often returns nondeterministic or hallucinatory answers. A practical pipeline is shown: agent pulls JSON from an observability tool, generates embeddings, searches a vector DB, then queries an LLM to produce summaries or SQL. Vendors (AWS Bedrock, Amazon Titan, Google) invest heavily; training can cost tens of millions and prompt engineering remains essential.
AI-written summary · about
chapters (19)
- 0:00Introduction and welcome to the talk
- 1:11Why observability matters in modern systems
- 3:19Cost of downtime and observability challenges
- 6:08Macroeconomic factors driving AI adoption
- 7:56Customer expectations from observability tools
- 9:21Evolution of large language models
- 13:19How large language models learn and improve
- 16:11Breakthroughs behind ChatGPT and generative AI
- 17:37Key features and use cases of ChatGPT
- 19:28Limitations and challenges of using ChatGPT
- 22:25The art of prompt engineering for better results
- 25:38Zero-shot, one-shot, and few-shot learning techniques
- 27:25Rules of thumb for effective prompt design
- 28:51Building a realistic solution with AI in observability
- 31:01Industry investments in generative AI
- 33:09Integrating predictive, causal, and generative AI
- 34:13Challenges and considerations for AI adoption
- 35:59Cost and environmental impact of large language models
- 37:02Summary and closing remarks
related talks
44mExploring ChatGPT for Improved Observability
23mExploring ChatGPT for Improved Observability
29mThe Hater's Guide to Dealing with Generative AI
25mThe Story of OpenAI GPT3 - Improve Observability with Tracing
24mHow OpenTelemetry Helps Generative AI
33mWhere Are We Going With Observability?
30mBuilding Reliable AI Agents: An Open Source Approach To Evaluation and Observability
41mGenerative AI: bliss or miss?