- Hacker News thread on difference between and rationale for using embeddings vs. fine-tuning an LLM. Takeaway: embeddings are good for retrieval of specific information (e.g. RAG) and fine-tuning is good for synthesizing responses (though subject to confabulation) and tone & voice (e.g. “style”)
- Glossary entry about AI PM. Takeaway: An AI PM is a PM that has a deep enough knowledge of AI to drive product strategy and delivery for AI-enabled products. Often folks with this knowledge have background in data science or ML. (I have lots of questions about this category fwiw)
- How to Use Your Product Management Experience to Ride the AI Wave. Takeaways: AI PMs bring together “strategic”, “tactical”, and “technical” factors to effectively deliver products. Strategic factors focus on understanding the actual value added by AI (rather than hype) and what it takes to implement before beginning. Tactical factors focus on breaking down design and development into milestones and tasks, and being able to effectively communicate to leadership. Technical factors focus on “the ability to discuss model trade-offs, experimentation approaches, infrastructure choices, and technology stacks, for example, is integral to the role”. In short, AI PMs have a deep enough knowledge to be effective. An important consideration–and why subject matter expertise matters–is an AI PM needs to be aware of the skills necessary for the multidisciplinary teams needed for delivering AI.
Some reading from this week on AI
Thomas Lodato
@deptofthomas