Getting ready for the role of AI/ML PM

Unlike many other disruptive technologies (like virtual reality), AI is growing too large to ignore.

It appears that all college students are currently studying AI, either for coursework or research. I’ve had conversations with master’s degree holders in chemical engineering and data science, among other fields.

How can I become a forward-thinking tech PM by learning enough technical material about AI now?

I’m thinking of suggestions for free or paid courses, news sources to follow, or both, but anything is helpful. I truly don’t want to return to school. I’ve earned three degrees already. I assumed a six-month program would be sufficient for me.

What do you think guys?

13 Likes

I am also just starting down this path, here are a couple resources I’ve come across so far:

According to what I’ve learned so far, a lot of the AI in product space—like anything in product—depends greatly on the environment in which one works. I hope those with more wisdom than I do can share some with us. Understanding the nuances of the product space and its AI applications requires a deep understanding of the specific industry and market dynamics. Factors such as customer preferences, competition, and technological advancements play a crucial role in shaping AI strategies. By learning from experienced professionals who have navigated these complexities, we can gain valuable insights that will help us make informed decisions in this rapidly evolving field.

11 Likes

Excellent observation. I work for a cloud provider as an AI PM, and part of my job description is to develop services that will either be integrated into SAAS or power other solutions.

I divided the market for AI products into two groups: customer-ready solutions and fundamental services. It will take foundational services for other businesses to develop artificial intelligence solutions. For example, LLMs, NLP/I skills, etc.

Then there are solutions that are ready for the consumer, such document summarization and conversational bots.

In all cases, you must begin with the use-case and the issue it attempts to solve before working your way backward from the end point.

I hope that’s useful.

10 Likes

PM currently, former ML engineer. I would continue to concentrate only on the user if I were you. It’s just as difficult to define the interface for human interaction with these technologies as it is to advance AI research. The success of AI systems heavily depends on how well they can understand and cater to the needs of users. By prioritizing user-centric design, you can ensure that AI technologies are not only advanced but also effectively serve their intended purpose. Additionally, understanding user behavior and preferences can provide valuable insights for further enhancing AI research and development.

Learn a little bit about unsupervised learning, regression, and classification at a high level before considering how this might lead to improved experiences. Even for me, it’s not a simple task, but that’s where time is best spent. By familiarizing yourself with unsupervised learning, regression, and classification, you can gain a deeper understanding of how AI algorithms work and their potential applications. This knowledge will enable you to make informed decisions when it comes to designing AI systems that deliver more accurate and personalized experiences for users. Investing time in learning these concepts will ultimately pay off in creating more effective and efficient AI technologies.

9 Likes

Absolutely agreed. As a general AI PM, I frequently advise other PMs that experience is more important than parameters or tokens. Since everyone will ultimately have what they call the “best” LLM, experience is our opportunity. Experience allows PMs to understand the nuances and complexities of managing AI projects, enabling them to make informed decisions and navigate challenges effectively. It helps in identifying potential pitfalls, anticipating user needs, and ensuring the successful implementation of AI solutions. Additionally, experience equips PMs with valuable insights that cannot be solely derived from parameters or tokens, ultimately contributing to the overall success of their projects.

9 Likes

Yes. We can create an experience that is both incredibly intricate and subtle, which should allow us years to dream about design and what the real demands are. Although the user is unconcerned with the algorithm’s workings, it does occasionally inspire us to consider how best to tailor our experiences to the specific ML problem at hand. Unsupervised issues can vary greatly.

I would love to speak with you one-on-one to hear your opinions about the market, where you are at, and what interests you. I’m more interested in distributed systems and software and would like to go back to my ML roots.

9 Likes

The exclusion of art of the conceivable is because we are unsure of exactly what to make; else, we would already be selling it to you. I would even go so far as to claim that we are able to specify the connections between models and information retrieval, which means that the technological delivery backend is more well understood than the user experience. Your expertise puts you closer to cognitive science than machine learning.

Regarding LLMs, In other words, what do you hope to accomplish or learn from your prompt? enlightening or doable actions that the software additionally offers? How would you want to view it? Even with ML in general, should we include you into every choice or can we filter based on urgency and importance and make certain low-cost decisions on your behalf? This kind of stuff keeps me awake at night. Just my viewpoint

Nice I’ll give you a call; I work at a big company too.

9 Likes

We’re using a public cloud here, that’s the whole idea. Although there are 500 possible approaches to satisfy the consumer, we must comprehend the technological and business context. a slightly different strategy than that of most software firms. We can construct the functionality to the specifications provided by the customer because we own the infrastructure. I chat to consumers the majority of the time to ascertain whether my solution is a good fit and to better understand their use cases.

However, you make several excellent arguments, with which I generally agree.

8 Likes

If you’re more interested in high-level information than in actually creating and deploying the models, Stratechery and Lenny have some excellent articles about the present state of the LLM explosion. These articles provide valuable insights into the current landscape of machine learning and its impact on various industries. They offer a broader perspective that can complement the technical knowledge gained from the Google Crash Course. Additionally, staying informed about the latest developments in the field can help you make more informed decisions regarding your solution’s fit within the rapidly evolving machine learning ecosystem.

A course on statistics introduction would be very beneficial. Understanding statistics is crucial for effectively analyzing and interpreting data in the field of machine learning. It provides a solid foundation for making accurate predictions and understanding the significance of results. By incorporating a course on statistics, you can enhance your understanding of the underlying principles and improve your ability to make informed decisions when developing machine learning solutions.

7 Likes

I’m a PM who has worked in AI/ML for the past ~9 years in various roles. When I’m learning something new either in CS or ML, I like to balance theory with practical coding.

If you’re looking for a course that does the balance well, check out the DeepLearning.AI Coursera ML specialization. Overall, it does a good, however the practical part isn’t sufficient to make you confident in self-implementation due to the “fill in the blanks” nature of the coding assignments.

If you’re looking for a book, there is no better starting point than https://www.statlearning.com/Theory/math isn’t overly complex, and there are coding labs and exercises at the end of each chapter. Coding is in R, but a Python version is coming Sumer 2023.

If you’re looking for quick books that skip the practical coding part, I really like these two books by Andriy Burkov. https://themlbook.com/, start [Machine Learning Engineering] However, his explanations might be too concise if you’re starting from zero.

6 Likes

@FergusXavier, How has acquiring coding skills benefited you as a PM? I take it that you are in charge of an engineering team that dissects and completes the task. Is the goal solely to improve project estimation skills, or does it aid in product vision or another area?

6 Likes

Gaining knowledge in coding benefits you in several ways:

Improved project estimation Given your current codebase, available resources, and project deadline, you’ll comprehend this better. It’s crucial to remember that this should truly be led by your engineers. I never commit to tasks or dates without first consulting the engineering team.

Increase your knowledge of the product’s technical aspects. What can be done, what cannot be done, and where is technology going? This is crucial in a field that is evolving quickly, like machine learning.

It strengthens your reputation with technical stakeholders and engineers and facilitates communication with them.

When you are aware of the complexities and difficulties involved in coding, you develop empathy for engineers. When coding at a production level, increase this by a factor of >ten.

It lets you take a look at the source and see what’s really occurring (there are often differences between what the documentation says, what the code does, and what the team believes is happening).

It gives you the ability to create prototypes or demos without the assistance of bug engineers.

5 Likes

Acquiring coding skills has greatly benefited me as a PM in multiple ways. Firstly, it allows me to have more effective communication and collaboration with my engineering team. Understanding the technical aspects of the project enables me to provide better guidance, address any challenges, and ensure smooth execution. Additionally, having coding skills helps me evaluate the feasibility of different product ideas and make informed decisions based on technical constraints. It also enhances my ability to prioritize features and align them with the overall product vision, ultimately leading to more efficient and effective product development. By being able to understand the intricacies of the engineering process, I can communicate more effectively with my team, allowing for a seamless exchange of ideas and a better understanding of what is possible within our time and resource constraints. This also allows for better collaboration, as I can actively contribute to code reviews and provide valuable insights to my team members. Ultimately, my technical knowledge and skills empower me to make well-informed decisions and lead my engineering team towards successful product outcomes.

4 Likes

I’ve got tons of textbooks. I read quite a few long form ”intro to” posts too. But the single best way I found was by reading the scikit-learn user guide. The scikit-learn user guide provides comprehensive and in-depth explanations of the library’s functionalities, making it an excellent resource for learning. Additionally, it includes practical examples and code snippets that help reinforce the concepts and improve understanding.

It’s a Python library with a soup-to-nuts of many ML techniques. The UG also describes many trade offs, appropriate use cases, and easily replicated examples. The scikit-learn user guide not only provides a comprehensive overview of various machine learning techniques but also offers insights into the trade-offs involved and suggests appropriate use cases. Additionally, it includes easily replicated examples that help in understanding and implementing the concepts effectively. This resource serves as an invaluable tool for learning machine learning techniques and their practical applications.

3 Likes

Having spent some time in the field, I can say that a lot depends on the issue you’re trying to address and the results you want to achieve. Although understanding ML and AI principles can be helpful, it can also be more difficult if you’re not actively employing them.

However, by using this resource, you can gain a solid foundation in machine learning and AI principles, which can enhance your problem-solving skills and enable you to make informed decisions when implementing these techniques. Additionally, the easily replicated examples provided in this resource allow for hands-on practice, making it easier to grasp the concepts and apply them effectively in real-world scenarios.

The other is that you don’t really need to know how anything works if you’re using ChatGPT, for instance. You must know how to use the technology to its fullest. Studying machine learning for its own sake—that is, without any practical application—might not be the best use of time.

Nevertheless, understanding the underlying principles of machine learning can greatly enhance your ability to utilize ChatGPT and other similar technologies effectively. By studying machine learning, you can gain insights into the limitations and potential biases of these models, enabling you to make informed decisions and mitigate any potential risks. Therefore, while practical application is crucial, a foundational understanding of machine learning can provide valuable context and empower you to make the most of these powerful tools.

2 Likes

Since I’m supposed to deliver an ML-based product this year, I’m also trying to figure this out. My college major was math and statistics, and I spent a lot of time learning on my own about the mathematics underlying deep learning and its practical applications. However, when I tried to use this knowledge to build a product, I found very little guidance on practical matters such as how to legally collect data, how much work it takes to manually annotate and train the models from the data, best practices for deploying to an edge-based architecture, etc. Navigating the practical aspects of implementing deep learning models can be challenging without proper guidance. Despite having a strong foundation in math and statistics, I realized that real-world applications require knowledge beyond just theoretical understanding. Factors like data collection legality, manual annotation efforts, and deploying to an edge-based architecture are crucial considerations that demand additional insights and expertise. It became evident that bridging the gap between theoretical knowledge and practical implementation is essential for successfully building deep learning products.

1 Like

You are on the cutting edge I guess! Probably good questions to ask other PMs in the space or the ML vendors. Something that seems consistent in being good at PM, regardless of ML factor, is learning to make decisions with limited data.

Still - I’m curious what else is different and what is the same about managing ML products vs say other B2B software.

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.