Suggestions
Sasha Friedman
Principal Product Manager - Technical at AWS
Sasha Friedman is a Principal Product Manager - Technical at Amazon Web Services (AWS).1 He is a product and technology leader with extensive experience in software product management.1
Professional Background
Sasha's current role at AWS involves focusing on problems of scale. As a hands-on product leader, he brings a wealth of experience to his position.2 His passion lies in making complex technical challenges more manageable, which aligns well with his role at AWS.
Skills and Expertise
Sasha's background is diverse, combining both technical and managerial skills:
- Product Management: He has significant experience in software product management.1
- Mechanical Engineering: Sasha has mentioned being both a product manager and a mechanical engineer, suggesting a strong technical foundation.2
- Problem-Solving: His focus on problems of scale indicates an ability to tackle complex technical challenges.
Education and Career Path
While specific details about his education are limited in the provided information, it's mentioned that Sasha attended Portland State University.1 His career path has led him to a senior position at one of the world's leading cloud computing platforms, indicating a successful trajectory in the tech industry.
Online Presence
Sasha maintains a professional presence on various platforms:
- His LinkedIn profile can be found under the username "sashadavidfriedman".1
- He occasionally shares insights about product management on LinkedIn.2
- Sasha has participated in discussions about product management, including a video on "Top Qualities of a Great Product Manager".34
Sasha Friedman's role as a Principal Product Manager at AWS, combined with his diverse background and focus on solving complex technical challenges, positions him as a notable figure in the field of product management within the cloud computing industry.
Highlights
The danger of recirculating AI slop is both the degradation and the homogenization of the response. Furthermore, and a problem that I spend a lot of time working on, is that without a strong human created and verified data set, your AI is stuck at a point in time. If you don't continuously update your models your AI will be:
-
Stuck in its knowledge. RAG can help, but are humans providing or verifying the retrieved data? For example, if a RAG system is pointed unknowingly at a source that is, under the hood, the source model. It turns into a snake eating its tail of data.
-
Stuck in its language. Language evolves over time, and there will be a new dynamic for how people use language and the model response. For example, think of language code switching, where regional, pop, or other culture is reflected in the accents and word choices of the speakers.
-
Stuck in its bias. When net new knowledge comes out that contradicts existing knowledge, how will the model correctly predict the next words if the corpus is suddenly false? For example, think of the back and forth of if butter is good for you.
https://lnkd.in/gYQQuVrj
#genai #rag #nytimes
I've been putting some cycles into prompt engineering, and the concept of Context continues to be a combination of interesting and vexing. In some cases I find myself wishing for infinite context when I've established a good working partnership with an agent, but in others I'm finding that I can use the inherently limited context that an agent has to my advantage via (painfully long) prompts.
