I've been thinking about how this has been developing.

1. Foundations
2. Data
3. Platform
4. Intent -> Fine Tuning -> Alignment Feedback loop
5. Maintenance
6. Transformers -> The Future
7. Agentic - A really nice interface
8. AI Models - Traditional AI vs Generative AI -> The misuse and distillation of all terms in a field where semantic truths are either non-deterministic or grounded in a human-guided "truth". Why understanding this matters.
9. Tools
10. Complexities
11. Is this really what people want? It's good for business. But there's something off about every car dealership emailing me the exact same cloned voice of a car sales rep. Without uniqueness. The perfect pitch. But it's unsettling...and so I've responded to none of them. Now, they're wasting their time emailing, calling and texting me in a period where they simply create more noise than I can deal with, so I ignore altogether.

Revision:

AI Musings: Turning Hype into an Operating Advantage

AI is a day-to-day operational reality. Which means the real question isn’t “Should we use AI?” It’s:

Are we going to deploy it intentionally—or let it show up everywhere, inconsistently, with unmanaged risk?

We need to make sure AI becomes a durable advantage: aligned to strategy, governed like any other enterprise capability, and shipped in a way that strengthens trust instead of quietly eroding it.

Here’s a map based on what I've pathed from my experience so far. I'd love to hear from others how their journeys are progressing.

1) Foundations: strategy before software

Most AI initiatives fail the same way digital transformations fail: they start with technology and end with confusion.

Your foundation is three decisions:

• Where does AI create strategic leverage? (Revenue growth, cost reduction, risk mitigation, speed-to-market, customer experience.)
• What are we willing to automate—and what must remain human-led?
• How will we measure success? (Not “number of copilots deployed.” Think cycle time, conversion, containment, CSAT, error rates, compliance incidents, cost per transaction.)

If you don’t set these boundaries early, you’ll get a portfolio of “interesting” proofs of concept and no scalable outcome.

2) Data: the real moat, the real liability

AI doesn’t magically make organizations smarter. It amplifies what you already are.

• Strong data discipline → AI adds leverage.
• Weak data discipline → AI amplifies inconsistency, bias, and operational noise.

Data questions that actually matter:

• What data are we using, and do we have the rights to use it?
• Is it current enough to support decision-making?
• Can we trace model outputs back to sources? (Especially for regulated workflows.)
• Are we training on internal knowledge that could leak?

If AI is going to touch customer communications, or compliance-adjacent processes, treat data governance as non-negotiable infrastructure—not a “Phase 2.”

3) Platform: the difference between pilots and enterprise capability

A lot of AI wins are “thin wrappers around an API call.” That can be fine—until you scale.

At enterprise scale, the platform layer determines whether AI is controlled or chaotic. This is where you standardize:

• identity and access control
• audit logs and traceability
• prompt/model versioning
• evaluation and quality monitoring
• policy enforcement (privacy, regulatory, brand)
• cost management (usage limits, caching, routing)
• resiliency (fallbacks, failover, rate limits)

We don’t need to design the platform—but we do need to fund it and mandate it. Otherwise every team reinvents the wheel with different risk thresholds and inconsistent customer experiences.

4) Intent → Fine-tuning → Alignment: your trust loop

AI outputs can be “good enough” in a demo and still be unacceptable in production, because the failure modes are different.

What matters is the feedback loop:

1. Intent: What behavior do we want, and where are the red lines?
2. Fine-tuning (sometimes): Do we need specialized behavior, or can we use retrieval + guardrails?
3. Alignment: How do we enforce policy, safety, and brand voice?
4. Feedback: How do we learn from real usage and failures?

This is less “IT project” and more “operating system.” You’re continuously shaping a capability that interacts with customers, employees, and regulated processes.

The key is governance: who owns the behavior, who signs off, and how quickly you can correct drift.

5) Maintenance: AI isn’t a launch, it’s a lifecycle

AI systems degrade because the world changes.

Policies change. Customer language changes. Products change. Competitors change. The model ecosystem changes.

So maintenance becomes a standing capability:

• monitor quality and drift
• refresh datasets and retrieval indexes
• update evaluation suites
• respond to new security threats (prompt injection, data exfiltration, tool misuse)
• manage vendor/model upgrades without breaking workflows

If you don’t budget for this, you’ll either freeze your system (and fall behind) or upgrade recklessly (and lose control). Neither is fun.

6) Transformers: what’s actually happening under the hood (in plain English)

Generative AI is powerful because it predicts “what comes next” at massive scale:

• next best token (text)
• next best pixel (images)
• next best action (agents)

It doesn’t “know” truth the way a database does. It generates plausible outputs based on learned patterns. That’s why it can be brilliant and confidently wrong in the same paragraph.

The takeaway: LLMs are not authoritative systems. They’re generative systems that must be paired with:

• verified sources (retrieval)
• constraints (policies/guardrails)
• evaluation (continuous testing)
• human oversight where impact is high

This is how you turn probabilistic output into reliable business outcomes.

7) Agentic AI: from “answers” to “actions”

The recent shift is from AI that responds to AI that operates.

Agents can plan, use tools, call APIs, write code, and complete tasks across systems—assuming you give them access.

This is where value spikes… and risk spikes.

Questions to ask before approving agents:

• What tools can it access? (And what can it change?)
• What permissions model exists? (Least privilege, role-based access, step-up approvals.)
• What’s the audit trail?
• What’s the kill switch?
• Where do we require human confirmation?

Treat agents like you’d treat a new category of employee: powerful, scalable, and requiring controls.

8) Traditional AI vs Generative AI: clarity is leadership

“AI” has become a catch-all term. That creates risk because expectations drift.

• Traditional AI often predicts or classifies from structured data (more bounded, more testable).
• Generative AI creates content/actions probabilistically (more flexible, more variable).

Why this matters for executives: the governance, measurement, and risk profile are different. If leaders treat generative AI like deterministic software, they’ll overtrust it. If they treat it like magic, they’ll underuse it.

Your job is to enforce semantic clarity so the org can make correct decisions.

9) Tools and training: capability beats chaos

Buying AI tools is easy. Building organizational capability is the actual work.

Your teams need:

• practical literacy (how these systems fail, not just how they demo)
• evaluation habits (what “good” means for your business)
• security and privacy instincts
• workflow design skills (where AI fits, where it doesn’t)
• cost awareness (AI spend can quietly balloon)

The most effective exec move here is to fund enablement and standardize patterns, so teams don’t learn via avoidable incidents.

10) Complexities and risks: manage it like security, not like marketing

AI risk isn’t theoretical—it’s operational:

• hallucinations and misinformation
• data leakage and privacy violations
• prompt injection and adversarial use
• bias and unfair outcomes
• regulatory and contractual exposure
• brand damage via tone, errors, or overreach
• runaway costs from uncontrolled usage

This doesn’t mean “don’t do AI.” It means do AI like a serious enterprise capability: with controls, monitoring, incident response, and accountability.

11) The “something off” problem: when AI creates noise instead of value

There’s a subtle, growing issue: AI makes it cheap to produce “perfect” outreach at scale. So everyone does it. And customers get flooded with the same polished voice.

Your car dealership example nails it: optimized messaging becomes generic messaging, and generic messaging becomes ignorable noise. Businesses then respond by sending more. The loop feeds itself.

Exec takeaway: AI can increase output while decreasing differentiation.

The winners will use AI to create relevance and distinctiveness, not just volume:

• fewer messages, better targeted
• authentic voice
• real customer context
• measurable lift, not just activity

12) Operationalizing AI: what “product-forward” looks like at enterprise scale

To make AI real across a global enterprise, you need a model that scales beyond individual teams. A product-forward approach includes:

• A central AI platform (controls, observability, cost, security)
• Reusable building blocks (retrieval, guardrails, eval harnesses, tool integrations)
• Clear governance (who can deploy what, with which approvals)
• Outcome measurement (business metrics + quality/safety metrics)
• A safe experimentation lane (so innovation is fast but bounded)

The goal is not “AI everywhere.” The goal is “AI where it materially improves the job-to-be-done—without compromising trust.”

Takeaways

AI is now an operational capability. The competitive edge will come from:

• choosing the right leverage points,
• building a platform that scales safely,
• managing risk with adult supervision,
• and preserving the human experience—because uniqueness and trust are the actual differentiators.

AI can absolutely be “good for business.”
But only if it’s good for the people experiencing it.

The description of my job comes from Marty Cagen of the Silicon Vally Product Group. In his book "Inspired", he explains the product manager's contribution.

Key Takeaways from Marty Cagan's "Inspired"

Let’s dive into the crux of what Cagan describes as the product manager’s contribution.

1. Vision & Strategy

At the heart of a product manager's role is the ability to define and communicate a clear product vision and strategy. This vision serves as a north star, guiding the development team and stakeholders toward a common goal. A product manager not only crafts this vision but also ensures it aligns with the broader business objectives and market needs.

2.  Customer Advocate

Cagan emphasizes the importance of understanding the customer. A product manager acts as the voice of the customer within the company. This means engaging directly with users, gathering insights, and deeply understanding their pain points and desires. By doing so, product managers can create products that genuinely resonate with the market.

3.  Decision Maker and Prioritizer

In the fast-paced world of tech, countless ideas and potential features compete for attention. The product manager’s role involves making tough decisions about what gets built next. This requires a keen sense of prioritization, balancing short-term gains with long-term strategic goals, and ensuring the team focuses on what will deliver the most value.

4. Leader and Facilitator

Leadership is a key attribute of successful product managers, according to Cagan. They must inspire and motivate cross-functional teams, often without formal authority. This involves facilitating collaboration between engineering, design, marketing, and sales to ensure everyone works cohesively towards the product goals.

5. Execution and Delivery

A product manager’s job doesn’t end with planning and strategizing. They are also deeply involved in the execution and delivery phases. This includes working closely with development teams during the build process, iterating based on feedback, and ensuring that the final product meets the high standards set by the vision.

6.  Metrics and Outcomes

Finally, Cagan stresses the importance of being data-driven. A product manager must establish clear metrics to measure the product’s success and continuously track performance. This analytical approach helps in making informed decisions, iterating on the product, and proving its value to stakeholders.

Conclusion

Marty Cagan’s *Inspired* serves as a foundational guide for understanding the multifaceted role of a product manager. By defining the vision, advocating for the customer, making strategic decisions, leading teams, overseeing execution, and measuring success, product managers play a crucial role in the lifecycle of a product. Their contributions are essential in transforming innovative ideas into successful, market-ready products.

Written with help of Node Weaver.

Newsites loves fearmongering. It gets traffic -> revenue. This past month, I've been keenly aware of the case of The United States vs TikTok.

Music.ly

I remember the original version of TikTok. Before it was renamed.

If you know me, you know why I was following. I moonlight as a neo-classical music composer. Check out: @edwingraymusic

I've some thoughts about business school that I'll save for another post. But that's my clear and open bias. I'm not a business guy.

I live for art.

And so I saw Music.ly as wholesome. Honest. Better than watching NBC's The Voice with Gwen Stefani.

Then came TikTok

I didn't wince.

Just a few years before, in the mid 2010's, the 7-second short-form video app Vine was doing the same thing.

One of the immedate differences that comes to mind is the data.

What do I mean about "The Data"

It's a broad term.

Just like its sister-buzzword, "Artificial Intelligence", Data is difficult to define.

The difference between Vine and TikTok is it's algorithm. An algorithm based on data collected from a user's habits. It's a beautiful design. Why wouldn't you want to watch more of the stuff you seem to be interested in? Why wouldn't you want to see less more people you know? More balanced than the aging Facebook/Meta/Instagram properties. Vine died. While TikTok saw unprecendented growth.

NodeWeaver Takes Over

Take it from here. I'm tired, but enjoyed this read:

From Musical.ly to TikTok: A Deep Dive into Data and User Engagement

Why Was Musical.ly Rebranded to TikTok?

Music.ly
I remember the original version of TikTok. Before it was renamed. The old name was "Musical.ly." If you know me, you know why I was following. I moonlight as a neo-classical music composer. Check out: @mytiktokhandle. I've some thoughts about business school that I'll save for another post. But that's my clear and open bias. I'm not a business guy. I live for art. And so I saw Musical.ly as wholesome. Honest. Better than watching NBC's The Voice with Gwen Stefani. Then came TikTok. I didn't wince.

What Changed With TikTok Compared to Vine?

Just a few years before, in the mid 2010's, the 7-second short-form video app Vine was doing the same thing. One of the immediate differences that comes to mind is the data.

What Do We Mean By "The Data"?

It's a broad term. Just like its sister-buzzword, "Artificial Intelligence", Data is difficult to define. The difference between Vine and TikTok is its algorithm. An algorithm based on data collected from a user's habits. It's a beautiful design. Why wouldn't you want to watch more of the stuff you seem to be interested in? Why wouldn't you want to see less of people you know? More balanced than the aging Facebook/Meta/Instagram properties. Vine died. While TikTok saw unprecedented growth.

Why Are Countries Like the United States Critical of TikTok?

Why are countries like the United States critical of this? At the heart of the concern is data privacy and the potential for data misuse. The United States, among other countries, fears that the personal data of its citizens could be accessed and possibly exploited by foreign entities, given TikTok's ties to China through ByteDance. This fear is not unfounded in an era where data is not just information, but a cornerstone of national security, economic strategies, and personal privacy.

How Can Graph Data Science Address These Concerns?

Here is where the relevance of graph data science becomes pronounced. The algorithm TikTok uses is essentially a complex web of user interactions, preferences, and behaviors—elements perfectly suited for analysis via graph databases and graph data science. This technology can map out intricate relationships and patterns that might not be visible with traditional data analysis tools.

Graph data science could also be employed to address concerns about data privacy and security. By analyzing how data flows through networks, it's possible to identify unusual patterns or breaches in real-time, which could make platforms like TikTok safer and more transparent. Moreover, graph databases could enhance the efficiency of data handling, ensuring that user data is not only secure but used in compliance with international standards and regulations.

What Opportunities Does This Technology Present for Social Media Platforms?

The opportunity here, especially for fields related to graph data science like those explored by NodeWeaver, is immense. Not only can this technology help safeguard user data, but it can also be used to refine the algorithms platforms like TikTok use, making them more effective and user-centric. For instance, enhancing recommendation engines to reflect not just user preferences but also ethical and cultural considerations could be a game changer in how social media platforms operate.

To sum up, while the transformation from Musical.ly to TikTok highlights a shift in digital entertainment's business and cultural landscapes, it also underscores the pivotal role of advanced data technologies. As a neo-classical music composer and an observer of social trends, you might find that the evolution of these platforms isn't just about content but about mastering the underlying currents of data and connectivity. Understanding this can give anyone, business-oriented or not, a powerful perspective on the digital world.

Written with the help of NodeWeaver. See if you could tell when it switches from my personal writing style to Node Weaver's. (Solution)

This post originally began with a different title:

Understanding Universal Dependency Annotation for Multilingual Parsing--And a Routing Algorithm that helps with Customer Feedback Analysis.

But the deeper I got in to it, the more I realized it wasn't the point.

I think a lot of business people have a hard time discerning "tech for novelty" versus early-stage "this can change everything" solutions. That's all the buzz right now.

Experts agree that we are now somewhere at the inflection point of an AI Revolution. Just like the Industrial Revolution and other revolutions of days-past. The tech is moving faster than most people can keep up with it.

So it's more important than ever to understand it.

Why is it important that we standardize language annotations?

For empowering multilingual parsing.

Why is it important that we empower multilingual parsing?

There are several reasons. But here are the few that stood out to me:

1. Ease of use - When users interact with systems in their own language. In their own slang. In their own dialect. For person to person communication, there's always room for miscommunication due to gaps in knowledge.
2. Customer Advocacy - Feedback analysis, topic modeling. Multilingual parsing enables the processing of data in various languages without losing context or meaning. This is something most humans in the world cannot do.
3. Serve a broader audience effectively

Also other stuff like content categorization and tagging, improving SEO, etc.

So now we understand. Here we go.

Deciphering Babel: How Standardizing Language Annotations Empowers Multilingual Parsing

In the modern Tower of Babel that is our global, digital society, languages intertwine and interconnect across platforms, devices, and borders. However, unlike the myth, where linguistic diversity halted construction, today's technology seeks to bridge these gaps. A recent paper from the 51st Annual Meeting of the Association for Computational Linguistics unveils an innovative approach to creating a "universal translator" of sorts—a set of standardized linguistic rules that could revolutionize how computers understand human languages. Here's how they did it.

Universal Grammar Rule Book: A Tech Marvel

Creating a standardized set of grammatical rules for multiple languages is akin to getting both cats and dogs to follow the same commands—challenging but not impossible. Here’s how researchers approached this monumental task:

The Need for Standardization

• Existing Inconsistencies: Each language had its own annotation system, making efficient multilingual parsing nearly impossible.
• The Universal Dependency Annotation: This new system proposes uniform rules across languages, making it easier for computers to understand and parse multiple languages simultaneously.

The Methodology

• Automated Conversions: For languages like English and Swedish, existing treebanks (think of these as extensive libraries of annotated text) were converted to the new standard using automated tools.
• Manual Annotation: For other languages, researchers manually annotated new treebanks to fit within the universal framework.
• Harmonization Process: To ensure consistency, annotations were revised across languages to match the new universal standards.

Improved Language Games: The Results

Using the newly standardized treebanks, researchers were able to conduct more reliable cross-lingual parsing experiments. Here's what they found:

• Increased Accuracy: The new system led to significant improvements in parsing accuracy compared to older methods.
• Enhanced Evaluation: For the first time, cross-lingual parsing evaluations could report on both unlabeled and labeled attachment scores, providing a deeper insight into the parser's performance.

The Open-Source Spirit: Sharing is Caring

In the spirit of collaboration, the researchers made their findings and tools available for free. This openness invites contributions from the global research community, which can lead to further enhancements and broader applications.

• Community Engagement: By providing open access to their treebanks and tools, the researchers have enabled others to contribute improvements and extend the dataset to more languages.

A Cautionary Tale: The Importance of Clean Data

To illustrate why standardized and clean data matters, let's consider a real-world scenario: imagine a multinational corporation that fails to standardize its customer data across different regions. This disarray can lead to miscommunications, inefficient operations, and lost opportunities—much like trying to build a skyscraper with different teams using incompatible blueprints.

Similarly, in linguistic research and application, inconsistent and messy data can significantly hamper the development of effective multilingual technologies. The work done by these researchers not only underscores the importance of clean, standardized data in computational linguistics but also serves as a blueprint for other fields where data uniformity is critical.

Conclusion: Building Bridges, Not Walls

The efforts to standardize linguistic annotations across multiple languages mark a significant step toward breaking down the barriers posed by language diversity. As this universal framework continues to evolve and expand, it holds the promise of making technology more inclusive and capable of understanding the nuances of human language. This isn't just about building better linguistic models; it's about fostering better understanding and communication across the globe.

In a world where data is the foundation of all digital endeavors, clean, standardized data isn't just desirable—it's essential. Like the skilled craftsmen of Babel, today's data scientists and linguists are laying the bricks for a future where language no longer divides us, but unites us in our shared digital spaces.

Written with the help of my new friend, Node Weaver.

The latest trends in graph data science reflect the growing importance of understanding complex relationships and patterns within data. Here are some key trends:

Integration with Machine Learning and AI: Graph data science is increasingly being integrated with machine learning and AI techniques. This integration enables more sophisticated analysis, such as predictive modeling and anomaly detection in complex networks. Graph neural networks (GNNs), a type of neural network that operates directly on graph structures, are particularly significant in this area.

Real-time Graph Analytics: There's a growing demand for real-time graph analytics, driven by applications such as fraud detection, recommendation systems, and network monitoring. This requires scalable and efficient graph databases that can handle dynamic, real-time data updates and queries.

Graphs for Knowledge Graphs and Semantic Search: Knowledge graphs, which represent relationships between entities in a structured form, are being increasingly used in various domains, including search engines, e-commerce, and digital assistants. They help in improving the relevance and contextuality of search results and recommendations.

Graphs in Cybersecurity: The application of graph data science in cybersecurity is rising. Graphs are used to model network traffic, user behavior, and system interactions, which helps in detecting anomalies, understanding attack patterns, and enhancing threat intelligence.

Graph Embeddings: Techniques for graph embeddings, which represent graph vertices in a low-dimensional space, are evolving. These embeddings make it easier to apply machine learning algorithms to graph data and are crucial for tasks like clustering, visualization, and similarity analysis.

Expanding Use Cases: Graph data science is finding new applications in diverse fields such as healthcare (for patient data and medical research), supply chain management (for optimizing logistics and monitoring networks), and social network analysis (for understanding community dynamics and influence patterns).

Graphs in Cloud and Distributed Environments: The trend towards cloud-based and distributed graph processing frameworks is growing. These frameworks offer scalability and flexibility for handling large-scale graph data.

Advancements in Graph Query Languages: Improvements and standardizations in graph query languages, like GQL (Graph Query Language), are making it easier for data scientists and developers to work with graph data.

Ethical and Responsible Use of Graph Data: As with all data science fields, there's an increasing focus on the ethical and responsible use of graph data, particularly in terms of privacy, data governance, and bias.

Graph Data Science as a Service (GDaaS): The rise of Graph Data Science as a Service platforms, which offer cloud-based, user-friendly environments for graph analytics, is making these technologies more accessible to a broader range of users and industries.

Keeping up with these trends involves staying informed about the latest research, tools, and applications in graph data science, and considering how they can be integrated into existing systems and workflows to drive innovation and solve complex problems.

Connect

What are you working on lately?

hey@edfig.me

Graph databases play a pivotal role in business process optimization by leveraging the power of graph data science. This approach is highly effective in dealing with complex, interconnected data, which is common in business processes. Here's how graph databases can aid in optimizing business processes:

1. Understanding Complex Relationships: In business processes, various elements like tasks, resources, and people are interconnected. Graph databases excel in mapping and understanding these relationships, making it easier to identify inefficiencies and dependencies.

2. Process Mining and Analysis: By visualizing processes as graphs, companies can use graph analytics to uncover patterns and bottlenecks in their workflows. This helps in optimizing the process flow, reducing redundancies, and improving efficiency.

3. Real-time Insights: Graph databases are adept at handling dynamic, real-time data. This capability allows businesses to react promptly to changes in the process flow, adapting and optimizing processes on the fly.

4. Scalability and Flexibility: Graph databases can easily scale with the growing complexity of business processes. They are also flexible, allowing for easy modifications and integrations with new elements in the business process.

5. Fraud Detection and Compliance: In financial processes, graph databases can detect unusual patterns indicating fraud. They also help in ensuring compliance by understanding the complex web of regulations and their impact on business processes.

6. Customer Journey Mapping: Understanding the customer journey is crucial for businesses. Graph databases can map out the entire journey, identifying key touchpoints and areas for improvement to enhance customer experience.

7. Supply Chain Optimization: In supply chain management, graph databases help in visualizing the network of suppliers, distributors, and customers, enabling more efficient inventory management, risk assessment, and route optimization.

8. Collaboration and Knowledge Sharing: By mapping the relationships between employees, departments, and knowledge assets, graph databases can enhance collaboration and knowledge sharing within an organization.

9. Personalization and Recommendation Systems: They can be used to develop sophisticated recommendation systems that take into account the complex web of user preferences and behaviors.

In summary, graph databases offer a unique approach to business process optimization by providing deep insights into the interconnections within data, enabling real-time analysis, and enhancing flexibility and scalability in handling complex business processes. This leads to more efficient, responsive, and intelligent business operations.

We're not that far away from Artificial General Intelligence.

Really.

This blog was inspired by one of my favorite people right now, Andrew Ng.

This post is a summary of Andrew's video (53 minutes):

I watched this video. And realized I want to share it's message. To revisit it myself and share with others. This is important.

AI can make the world a better place.

The opportunities, and the vision, are exciting.

Walk me through the video

1. Introduction:

• AI applications that once took months or years to build can now be developed in a matter of days or weeks.
• The potential for AI applications is vast, and we're just scratching the surface.

2. AI's Rapid Development:

• Deployment to the cloud has become faster.
• Teams globally can now build complex AI applications in a fraction of the time it used to take.

3. AI's Expanding Horizon:

• There's an impending flood of AI applications that many still underestimate.
• More people are now able to build and utilize AI applications.

4. Demonstrating AI's Simplicity:

• The speaker demonstrates building a sentiment classifier using minimal code.
• Developers can now build AI systems in mere minutes, showcasing the ease and speed of AI development.

5. AI Education:

• Emphasis on teaching not just AI consumption but also its development.

6. AI's Current Value and Future Potential:

• Most of AI's financial value today comes from supervised learning.
• Generative AI is an emerging and rapidly growing sector.
• The potential growth of AI technologies in the next three years is massive.

7. AI as a General-Purpose Technology:

• AI can be applied across various industries and tasks.
• There's still much to explore in terms of AI's use cases, especially with generative AI.
• However, there will be short-term trends and fads.

8. AI's Impact on Industries:

• AI's value is still largely concentrated in the tech world.
• There's a need for tools that allow industries to customize AI for their specific needs.
• The rise of low-code and no-code tools is enabling more industries to harness the power of AI.

9. Building AI Startups:

• The speaker's approach to building AI startups involves a systematic process of idea validation, CEO recruitment, prototype development, and scaling.
• Examples of successful AI startups, like bearing AI, are provided.

10. Risks and Social Impact of AI:

• Ethical considerations are paramount.
• AI has issues with bias, fairness, and accuracy, but improvements are ongoing.
• Job disruption is a significant concern with AI's rise.
• Concerns about AI leading to human extinction are considered overblown.

11. Conclusion:

• AI, as a general-purpose technology, offers numerous opportunities.
• The future holds immense potential for AI, and collaboration is key to harnessing its full potential.

How I generated this summary

I created this post with the assistance of my very own app, YouTube Video Summarizer. Simply provide a Youtube video link, and within seconds receive a summary of the entire video. Neat right? Coming soon to iOS and Android. Sign up for the Beta by reaching out to buzz@figfutures.com.

This post begins with my notes on the topic. Answering it in a form inspired by Sakichi Toyoda's "Five Why's". Then, I use ChatGPT to summarize it.

If you enjoyed this poetic non-fiction reading and want to read more, please consider reading book: Making Data Easy.

Why do machines hallucinate?

Because we ask them to.

1) Why do we ask machines to hallucinate?

To help us answer questions.

2) What kinds of questions?

All kinds of questions.

The mania used to scare me.

Now I understand that the chaos is engineered.

To provide a purpose. A capability.

To answer any questions.

Five why’s. We’ve answered three. We have two to go:

3) Why chaos?

What is chaos? It’s a programmed randomness. We then apply statistical methods based on user-defined goals and context, and diffuse the randomness into some answers we can humanly understand. But the model doesn’t understand. It’s just providing words. Or next-best pixels. It doesn’t actually know anything.

Chaos Engineering

Modeled after the chaos that exists in nature. The ever-expanding equation. Entropy.

Seek the intuition in things.

Hope everyone is safe.

4) Can we always trust the answers?

No. The models are trained on human language. Which you can’t inherently trust. But we can make this better.

5) How can we make this better?

We provide more context.

*The more you know*

Now, let’s ask an AI to turn this into something easier to read.

What is this note talking about?

This note from your friend is discussing the idea of machine learning and AI "hallucinating," which refers to the way these systems generate information or answers based on patterns they've seen in data. Here is a simplified breakdown of the key points:

1. Machines "hallucinate" (generate information) because we ask them to.
2. We ask them to do this to help us answer questions.
3. The "chaos" referred to is essentially a programmed randomness that machines use to generate answers. It's managed by applying statistical methods based on user-defined goals and context.
4. "Chaos Engineering" here refers to intentionally adding randomness to a system to understand its behavior better, similar to how chaos exists naturally.
5. The note mentions that AI models don't actually understand anything. They just provide words or "next-best pixels" based on patterns they've seen in data.
6. The note asks whether we can trust the answers from these models. The answer is "No," because the models are trained on human language, which can be flawed or biased.
7. Finally, to make these models better, the note suggests providing more context, which can help in generating more accurate and relevant responses.

Is all of this true? Let's always be a little skeptical.

The note presents a perspective on how machine learning and AI systems work. While the concepts mentioned are based on valid ideas, they may be a bit simplified or generalized. Here is a breakdown of the truthfulness of the statements:

1. Machines "hallucinate": In a sense, this is true. Machine learning models, particularly generative models, create outputs that may seem like they are "hallucinating" information. However, it's more accurate to say they generate outputs based on patterns they've seen in data.
2. Chaos Engineering: This is a real concept. Chaos Engineering involves intentionally introducing failures or disruptions into a system to test its resilience and reliability. However, in the context of the note, the term "chaos" is used to describe the randomness or unpredictability that machine learning models can exhibit.
3. AI models don't actually understand anything: This is mostly true. AI models don't "understand" in the human sense. They operate based on patterns and associations found in data.
4. We can't always trust AI models: True. AI models are only as good as the data they are trained on. They can exhibit biases or inaccuracies present in the data.
5. Providing more context can improve AI models: Generally true. Adding context can help AI models make more informed and accurate predictions. However, there are other factors that also play a role in improving AI models, such as better algorithms and higher-quality data.

The note provides a perspective on AI and machine learning, but it's important to recognize that these are complex fields with many nuances.

What are some nuances to be aware of?

There are several nuances in AI and machine learning to be aware of, some of which include:

1. Data quality and bias: Machine learning models rely heavily on the data they are trained on. If the training data is biased or flawed in some way, the model is likely to exhibit those biases in its predictions.
2. Overfitting: A model that performs very well on the training data but poorly on unseen data may be overfitting. This means that the model has learned to perform very well on the specific data it has seen, but it may not generalize well to new data.
3. Interpretability: Machine learning models, especially deep neural networks, are often criticized for being "black boxes." This means that it can be difficult to understand how the model arrived at a particular decision.
4. Ethics: The use of AI raises several ethical questions, such as concerns about privacy, fairness, and the potential for abuse. It's important to consider these implications when using AI systems.
5. Limitations: AI systems have limitations and are not suitable for all tasks. For example, while AI models can perform well on certain tasks like image recognition, they may struggle with other tasks that require deeper understanding or common sense.
6. Trade-offs: There are often trade-offs in machine learning, such as the trade-off between model complexity and interpretability or the trade-off between accuracy and fairness.
7. Feature Engineering: Selecting the right features to include in a machine learning model can have a significant impact on the model's performance. However, deciding what features to include and how to represent them is often more of an art than a science.
8. Model Assumptions: Many machine learning models make certain assumptions about the data they are working with, such as the assumption that the data is independent and identically distributed (i.i.d). Violations of these assumptions can affect the performance of the model.

These are just a few of the many nuances to be aware of in AI and machine learning. It's important to approach these fields with a critical eye and to understand the limitations and complexities involved.

Please provide a short, kindergarten level summary of what we’ve learned today.

Today we talked about how computers can "hallucinate" or make up answers to questions by looking at lots of information. But, sometimes they can get things wrong because they don't really understand like people do. They just see patterns. We also learned that we need to be careful when using computers to answer questions because they can make mistakes or be a little bit confused. So, it's important to know that computers are helpful, but not perfect.

If this is interesting to you, please check out my book: Making Data Easy.

Thanks!

Edwin

In the year 2010, during the dynamic chaos of my life as a 24-year-old punk rock enthusiast at an indie music label, my universe was far removed from the glamorous world of mainstream music. One typical lunch break, as I was the lone guard at the fort, answering phones, the cosmos decided to add a dash of spice.

The phone rang, and on the other end was a voice synonymous with chart-topping hits and sold-out concerts – one of the world's leading rappers. There was no suspense or mystery; he revealed his identity from the get-go. But his stellar reputation in the rap scene did not ring any bells in my punk rock immersed brain. Not because I was being dismissive or intentionally ignorant, but simply because I was clueless about who he was.

"I need to speak to someone with power," he stated with an authoritative tone. Power? I smirked at the thought. My claim to "power" was my uncanny knack for overpaying artists by a handful of cents due to a notorious rounding error on streaming music royalty statements.

As I attempted to explain that he had probably dialed the wrong number, that our label shared a name but was not the one he sought, he found it hard to accept. His rap-star status had always served as a VIP ticket, opening doors and breaking barriers, but this time it was met with an unfamiliar roadblock.

His disbelief echoed through the phone, and unable to convince him otherwise, I resorted to my only available option—I hung up, punctuating our peculiar exchange with an awkward pause.

This unexpected phone call underscored the intriguing intersection of two vastly different music worlds, demonstrating that identity, much like my notorious rounding error, can sometimes be misinterpreted. Even a top-tier rap icon could make a wrong call, just like an indie label employee could fail to recognize him. After all, in this grand melody of life, we all hit a few wrong notes now and then.

As I put down the phone that day, I found myself in a comic yet thought-provoking situation. The incident, a curious mix of mistaken identities, erroneous power perception, and misdialed numbers, was a precursor to the life I lead today.

Fast forward to the present, and I've traded my indie music label days for the world of data management. In this new realm, I've come to see that the chaotic world of indie music and the systematic universe of data management aren't too different after all.

Reflecting on the day of the phone call, the renowned rapper had dialed the wrong label because he didn't have all the information he needed based on the name alone. It reminded me of a recurring challenge in data management.

Just as our indie label shared a name with a completely different rap label, companies in the data world often bear the same or similar names, causing potential for mix-ups. There can also be multiple ways to spell a company's name, think "7-11" versus "Seven-Eleven". And let's not forget the complex network of parent and subsidiary companies, akin to Amazon's ownership of Zappos or Alphabet's control over Google.

This experience served as a lesson on the importance of metadata, of context. Context is often one of the hardest things to establish and yet it's crucial in understanding the full picture. Whether it's a rapper looking for the right label or a data analyst sifting through companies, more context, more information, is always needed.

The world of data is complex, much like the rhythm of life, always evolving, always demanding a closer look. So, while I no longer juggle phone calls from renowned rappers, I do navigate through a labyrinth of data, striving to provide that crucial context, because as it turns out, in life and data, context is everything.

Other Titles for this Story

1. "Behind the Dial Tone: A Misstep in Musical Dimensions"
2. "Lost in Transmission: From Indie Label to Data Label"
3. "The Unintended Remix: A Dial-tone Epiphany"
4. "Crossed Wires, Crossed Worlds: A Rap Star's Call and the Lessons Learned"
5. "The Wrong Number: An Unexpected Prelude to a Data-Driven Symphony"

This is a true story written with the assistance of ChatGPT.

It's completely not in my own "tone". I can write like this with a bit of effort. But I don't write like this. I write the way I talk. With conjunctions and a bit of SoCal Slang. But neverthess--this was an interesting exercise.

Want to learn how to write like this? Read Data Made Easy. One of my favorite chapters is about Generative Artificial Intelligence for writing stories just like this.

This article was generated by asking chatGPT, a revolutionary new NLP model and chatbot. The answers are awesome.

What is a Data Platform?

A data platform is a collection of tools, technologies, and infrastructure that enables an organization to manage, process, and analyze large volumes of data. It is designed to support the extraction, transformation, and loading (ETL) of data from various sources, as well as the storage, querying, and visualization of data for various business and analytics purposes.

A data platform typically includes a variety of components, such as data warehouses, data lakes, data integration tools, data visualization tools, and machine learning and analytics platforms. It may also include data governance, security, and compliance tools to ensure that data is protected and used appropriately.

The goal of a data platform is to provide a single, centralized location for storing and managing all of an organization's data, as well as to provide the tools and capabilities needed to extract insights and value from that data. Data platforms can be used for a wide range of applications, including business intelligence, customer analytics, fraud detection, and predictive maintenance.

Can you shorten that, please?

Sure, a data platform is a system that helps organizations manage, process, and analyze large volumes of data from various sources. It includes tools and infrastructure for storing, querying, and visualizing data, as well as for extracting insights and value from it. Data platforms are used for a variety of purposes, including business intelligence, customer analytics, and fraud detection.

What are some of the ethical concerns?

There are a number of ethical concerns that can arise in the design and use of data platforms, particularly with respect to the collection, storage, and use of personal data. Some of these concerns include:

1. Privacy: Data platforms often store and process large amounts of personal data, which raises concerns about the privacy of individuals whose data is being collected and used. Organizations using data platforms must ensure that they have obtained appropriate consent from individuals for the collection and use of their data, and that they are protecting that data from unauthorized access or misuse.
2. Security: Data platforms must be designed and configured to ensure the security of the data they store and process. This includes protecting data from unauthorized access, as well as from cyber attacks and other forms of data breaches.
3. Fairness and bias: Data platforms can perpetuate or amplify biases if the data used to train or inform decisions is biased. It is important to ensure that data is representative and unbiased, and that algorithms and models built on top of the data platform do not discriminate against certain groups or individuals.
4. Transparency: Organizations using data platforms should be transparent about how they are collecting, storing, and using data, and should provide individuals with the opportunity to review and correct their data if necessary.
5. Accountability: Organizations using data platforms should be held accountable for their data practices, and should have systems in place to ensure that they are complying with relevant laws and regulations, as well as with their own policies and procedures.

What are some of the legal concerns?

There are a number of legal concerns that can arise in the design and use of data platforms, particularly with respect to the collection, storage, and use of personal data. Some of these concerns include:

1. Data protection and privacy laws: Many countries have laws that regulate the collection, storage, and use of personal data, such as the General Data Protection Regulation (GDPR) in the European Union and the California Consumer Privacy Act (CCPA) in the United States. Organizations using data platforms must ensure that they are compliant with these laws and that they have obtained appropriate consent from individuals for the collection and use of their data.
2. Intellectual property: Data platforms may contain or generate intellectual property, such as patented algorithms or copyrighted content. Organizations using data platforms must ensure that they have the appropriate rights and licenses to use this intellectual property, and that they are not infringing on the rights of others.
3. Contractual obligations: Data platforms may involve the sharing of data or other resources between different parties, such as between a business and its customers or partners. It is important to ensure that these relationships are governed by appropriate contracts that outline the rights and obligations of each party.
4. Employment and labor laws: Data platforms may involve the collection and use of data about employees, such as performance data or personnel records. Organizations using data platforms must ensure that they are complying with relevant employment and labor laws, and that they are treating their employees fairly and lawfully.
5. Data breaches and cybersecurity: Data platforms may be vulnerable to data breaches or cyber attacks, which can result in the unauthorized access or disclosure of sensitive data. Organizations using data platforms must have appropriate safeguards in place to prevent these types of incidents and to respond appropriately if they do occur.

How might I use these tools to research an intellectual property concern?

There are several ways that you can use data platforms and related tools to research intellectual property concerns, such as patent infringement or trademark violations. Here are a few options:

1. Searching patent and trademark databases: Many countries maintain online databases of registered patents and trademarks, which you can search to determine whether a particular invention or brand is protected by intellectual property laws. For example, the US Patent and Trademark Office (USPTO) maintains a database of granted patents and pending patent applications, as well as a database of registered trademarks.
2. Analyzing patent and trademark data: Data platforms can be used to analyze patent and trademark data in order to identify trends, patterns, or potential issues. For example, you could use data visualization tools to map the relationships between different patents or trademarks, or to identify clusters of related patents or trademarks.
3. Using machine learning and artificial intelligence: Some data platforms include tools for machine learning and artificial intelligence, which can be used to analyze patent and trademark data in order to identify patterns or trends that might not be visible to the human eye. For example, you could use machine learning algorithms to identify clusters of similar patents or trademarks, or to predict the likelihood of a particular patent being granted based on past data.
4. Conducting online searches: Data platforms can be used to conduct online searches for information about patents and trademarks, such as by using web scraping tools to gather data from relevant websites or by using natural language processing to analyze online content.

It is important to note that the use of data platforms and related tools to research intellectual property concerns should be done in accordance with relevant laws and regulations, and that the results of these analyses should be interpreted carefully. You may wish to consult with a legal professional or an intellectual property expert to help interpret and understand the results of your research.

How might I use these tools to build better products for my customers?

There are several ways that you can use data platforms and related tools to build better products for your customers. Here are a few options:

1. Gathering customer feedback and data: Data platforms can be used to collect and analyze customer feedback and data in order to understand customer needs and preferences. This can include gathering data from online reviews, customer surveys, or social media interactions, as well as from sales data and other sources.
2. Using machine learning and artificial intelligence: Some data platforms include tools for machine learning and artificial intelligence, which can be used to analyze customer data in order to identify patterns or trends that might not be visible to the human eye. For example, you could use machine learning algorithms to identify common customer pain points or to predict which products or features customers are most likely to be interested in.
3. Testing and iterating: Data platforms can be used to test and iterate on product ideas and prototypes, using data from customer feedback and market research to refine and improve the product. This can include using tools like A/B testing to compare different versions of a product, or using data visualization tools to identify areas for improvement.
4. Personalizing the customer experience: Data platforms can be used to personalize the customer experience by tailoring products or services to individual customers based on their preferences, behaviors, and history. This can include using tools like customer segmentation or recommendation engines to identify and target specific customer groups.

Overall, the goal of using data platforms and related tools to build better products for your customers is to use data to inform and guide product development and improvement efforts, in order to better meet the needs and preferences of your target audience.

How do I estimate the costs of a Data Platform?

There are several factors that can impact the costs of a data platform, and the specific costs will depend on the specific needs and requirements of your organization. Here are a few key factors to consider when estimating the costs of a data platform:

1. Hardware and infrastructure: The costs of hardware and infrastructure can vary significantly depending on the size and complexity of your data platform. This can include the costs of servers, storage, networking equipment, and other hardware, as well as the costs of maintaining and supporting this hardware over time.
2. Software and licenses: The costs of software and licenses can also be a significant factor, depending on the specific tools and technologies you are using. This can include the costs of data integration tools, data visualization tools, machine learning platforms, and other software, as well as the costs of any required licenses or maintenance fees.
3. Data storage and processing: The costs of storing and processing data can vary depending on the volume and complexity of your data. This can include the costs of storing data in a data warehouse, data lake, or other data storage system, as well as the costs of processing and querying data for analysis or visualization purposes.
4. Staffing and expertise: The costs of staffing and expertise can vary depending on the size and complexity of your data platform, as well as the level of expertise required to support it. This can include the costs of hiring or training data scientists, data engineers, or other specialized personnel, as well as the costs of ongoing support and maintenance.

To estimate the costs of a data platform, you will need to consider all of these factors and determine the specific hardware, software, and staffing resources that your organization will need. You may wish to work with a data platform vendor or a consulting firm to help you assess your needs and develop an estimate of the costs.

What's the definition of a Platform again?

A platform is a type of infrastructure that enables the development and deployment of products, services, or applications. It typically provides a set of standardized tools, technologies, and resources that can be used by developers or organizations to create and distribute their offerings.

There are many types of platforms, including software platforms, hardware platforms, and data platforms. Software platforms are frameworks that enable the development and deployment of software applications, while hardware platforms refer to the physical devices or systems that support the operation of software or other applications. Data platforms are systems that enable the management, processing, and analysis of large volumes of data.

Platforms can be used to facilitate the exchange of information, resources, or value between different parties, and they often enable the creation of ecosystems or networks of users, developers, and other stakeholders. They can also provide a means of monetizing products, services, or applications, either through direct fees or through the sale of ads or other forms of revenue.

What are some of the risks involved in Data Platforms?

There are a number of risks that can be involved in the use of data platforms, including:

1. Data security and privacy: Data platforms often store and process large amounts of sensitive data, which raises concerns about the security and privacy of that data. Data platforms must be designed and configured to protect data from unauthorized access or misuse, and must comply with relevant data protection and privacy laws and regulations.
2. Data quality: Data platforms may rely on data from a variety of sources, which can introduce issues of data quality. Poorly structured or unreliable data can lead to incorrect or misleading insights, and can impact the accuracy and effectiveness of data-driven decisions.
3. Data governance: Data platforms must be designed and managed in a way that ensures the appropriate use and handling of data, and that complies with relevant laws and regulations. This can involve establishing policies and procedures for data collection, storage, and access, as well as implementing data governance and compliance tools.
4. Data ethics: Data platforms can raise ethical concerns related to the collection, use, and impact of data, particularly with respect to personal data. Organizations using data platforms must ensure that they are acting ethically and transparently with respect to data, and that they are not discriminating against or unfairly impacting certain groups or individuals.
5. Integration and interoperability: Data platforms may need to integrate with other systems or technologies in order to function effectively, which can introduce additional risks related to compatibility and interoperability. Ensuring that data platforms are compatible with other systems and can exchange data seamlessly can be a complex and time-consuming process.
6. Dependency: Organizations that rely heavily on data platforms may become dependent on those platforms, which can create risks if the platform experiences technical issues or outages, or if the organization needs to switch to a different platform. Ensuring that data platforms are reliable and flexible, and that there are contingency plans in place, can help mitigate these risks.

How many people does it take to build and maintain a data platform?

The number of people required to build and maintain a data platform can vary depending on the size and complexity of the platform, as well as on the specific tools and technologies being used. A small data platform with a limited scope and a simple set of requirements may require just a few people to build and maintain it, while a larger, more complex data platform may require a team of multiple individuals.

In general, a data platform may require a combination of the following types of personnel:

1. Data engineers: Data engineers are responsible for designing, building, and maintaining the infrastructure and tools needed to collect, store, and process data. They may be involved in tasks such as designing data pipelines, setting up data storage systems, and optimizing data processing performance.
2. Data scientists: Data scientists are responsible for analyzing and extracting insights from data, and for building machine learning models and other analytical tools. They may be involved in tasks such as exploring and visualizing data, developing algorithms and models, and interpreting results.
3. Data analysts: Data analysts are responsible for collecting and analyzing data to inform business decisions. They may be involved in tasks such as gathering and cleaning data, creating reports and dashboards, and providing insights to business stakeholders.
4. IT professionals: IT professionals may be involved in the maintenance and support of the data platform, including tasks such as monitoring performance, troubleshooting issues, and ensuring compliance with relevant laws and regulations.

It is worth noting that the roles and responsibilities of the individuals working on a data platform may overlap, and that individuals may be responsible for multiple tasks or may work in multiple roles. The specific number and types of personnel required to build and maintain a data platform will depend on the specific needs and goals of the organization.

How do I hire a really good Data Scientist?

Hiring a good data scientist can be a challenging task, as the role requires a combination of technical skills, business acumen, and analytical thinking. Here are a few tips for finding and hiring a high-quality data scientist:

1. Clearly define the role and responsibilities: It is important to have a clear understanding of what you are looking for in a data scientist, and to define the specific skills and responsibilities that the role will involve. This will help you identify the most qualified candidates and ensure that you are attracting the right talent for your organization.
2. Look for a strong foundation in math and statistics: Data science involves the use of statistical and mathematical techniques to analyze and extract insights from data. As such, it is important to look for candidates who have a strong foundation in math and statistics, and who are comfortable working with large datasets and complex analytical techniques.
3. Seek out diverse experience: Data scientists often work on a wide range of projects and may need to adapt to new domains and technologies. Look for candidates who have diverse experience and a track record of learning and adapting quickly.
4. Consider communication skills: Data scientists often need to be able to communicate their findings and recommendations to non-technical stakeholders, so it is important to look for candidates who have strong communication skills and the ability to translate complex technical concepts into clear and concise language.
5. Evaluate problem-solving skills: Data science involves solving complex problems using data, so it is important to look for candidates who have strong problem-solving skills and the ability to think creatively and critically.

To find the best candidates, you may need to use a combination of recruiting methods, such as posting job listings, networking with industry professionals, or working with a recruiting agency. It may also be helpful to reach out to academic institutions or professional organizations in order to connect with qualified candidates.

Is there anything related to Data Platforms I'm forgetting to ask?

There are many aspects of data platforms that you might want to consider, and it is possible that you have not covered all of the topics that are relevant to your specific needs and goals. Here are a few additional questions that you might want to consider:

1. How will the data platform be used? What are the specific business or analytical goals that the data platform is intended to support? How will the data platform be integrated with other systems or processes within the organization?
2. What types of data will be stored and analyzed by the data platform? Will the data platform need to support structured, unstructured, or streaming data? What are the specific data sources that will be used to populate the data platform, and how will the data be collected and ingested?
3. What tools and technologies will be used to build and maintain the data platform? Will the data platform be built using open source or proprietary tools? What are the specific features and capabilities of the tools being used, and how will they support the needs of the organization?
4. How will the data platform be secured and managed? What measures will be taken to protect the data stored in the data platform from unauthorized access or misuse? What processes and procedures will be in place to ensure that the data platform is compliant with relevant laws and regulations?
5. How will the data platform be scaled and maintained over time? What are the anticipated growth and usage patterns for the data platform, and how will the platform be scaled to meet these needs? What are the ongoing maintenance and support requirements for the data platform, and how will these be managed?

Format that into an action plan for me, please. And create a roadmap.

Sure, here is an action plan and roadmap for building and maintaining a data platform:

Action plan:

1. Define the scope and goals of the data platform: Clearly identify the business or analytical goals that the data platform is intended to support, and define the specific types of data and use cases that the platform will need to address.
2. Identify data sources and requirements: Determine the specific data sources that will be used to populate the data platform, and assess the data ingestion and storage requirements of the platform.
3. Select tools and technologies: Choose the tools and technologies that will be used to build and maintain the data platform, considering factors such as cost, capabilities, and compatibility with other systems.
4. Build and test the data platform: Using the selected tools and technologies, build and test the data platform to ensure that it meets the defined requirements and goals.
5. Secure and manage the data platform: Implement measures to protect the data stored in the data platform from unauthorized access or misuse, and establish processes and procedures to ensure compliance with relevant laws and regulations.
6. Monitor and maintain the data platform: Regularly monitor the performance and usage of the data platform, and take steps to optimize and maintain it over time.

Roadmap:

1. Define scope and goals

Regardless of whether you hear it or not, in all practical cases, your audio track literally loses fidelity when you convert a WAV to an MP3.

There are probably other good methods. Here is my usual method, on Mac OS Monterey (Desktop)

My usual method:

1. Check the Advanced file import settings in Apple Music (aka iTunes)
2. Edit the Import Settings
3. Edit the new MP3 track's metadata
4. File -> Convert Track

Simple, once you know.

Why you might want to try this?

Fidelity is not quality. Quality is not fidelity.

It's one of those situations where there is a strong correlation between the two. But there are outliers. And sometimes, you have a reason for creating those outliers.

Mine: An FM music radio station in the United Kingdom is going to play my new single 🍾🥂

The curator asked me to send a <50MB WAV file or a high quality 320+ MP3. The WAV file for my final mastered track is ~60MB. So I couldn't just send the master. I needed to convert it to an MP3.

From experience, I know there are a lot of ways to do this wrong.

A better method: I can't find one quickly enough.

Understanding the practical difference between bit rate, variable bit rate and iTunes custom settings.

What was the original goal of the mp3?

To make the file size smaller. And I assume most of the new stuff contributes to that.

If I want my WAV file to stay as close to the original as it is, then I select the highest setting and turn off any "corrections".

My track is perfect. Don't touch it, iTunes algorithms.

Here's what my settings look like now. For the best quality.

I'm not filtering out frequencies, because even though you can't hear them, it is said that you can feel them. I'm not trying to save storage space. So I'm leaving those frequencies in tact.

Export to MP3

I always found the way Apple Music/iTunes handles library files to be a bit unintuitive. And it seems to change from release to release. Today, you can find it in the File menu:

The new MP3 file will automatically get added to iTunes. But it will NOT automatically be added to the playlist, your current view. The easiest way to find it is to review your "recently added".

After you find your file, it's best-practice to check and edit the metadata. all of the metadata.

Edit the metadata

Depending on the use-case, this can be optional. But for any files you plan on distributing (like the one I plan on sending the radio station) you should edit both the filename and the metadata.

You can even add embedded artwork on the mp3!

That's all. Hope that helped.

Questions/Comments? Let me know: ed@figfutures.com

But honestly, I've always hoped for better virtual whiteboards vs the real thing. I have terrible handwriting. I'm a tech native. Even in my school years, in the late 90s, I was the kid in class taking notes on a Palm Pilot PDA.

The most whiteboarding I usually do, is in a job interview. Otherwise, I'm just another guy in the room, working collaboratively in whatever way I'm needed. I'll whiteboard if I need to. But I prefer to prepare my work ahead of time. I'll create wireframes, mock schemas, and create articles to share with everyone in the meeting. The articles need to contain real, compelling user stories.

User stories need depth. You're trying to provide context to a problem. To get a team of people to rally around a solution. For stories to have depth, you need to prepare a drill-down.

How do I prepare stories?

Miro was recommended to me by a UX buddy a few months ago. Almost everyone I work with now uses it.

I instantly recognized it as Electronjs based app, it's speed, it's web components. Expertly designed. Beautiful and easy to use.

Here's my life story, in Mindmap Form

This might be my favorite solo brainstorming method so far.

I wonder if there's a way to do this collaboratively 🤔

It’s known for being brutally difficult. Your character will die. Many, many times.

But it’s also incredibly rewarding.

Product Manager is similar to Dark Souls.

You fail quickly. But you learn.

There are death traps at every corner.

But with experience, accumulated knowledge, you learn to avoid them.

What used to take you 5 hours now takes you 5 minutes.

Rapid death is akin to having a culture of continuous optimization.

• Short term goal: Make it to the next checkpoint without dying
• Long term mission: Beat the game. Meet your objectives.

This might be true for life too.

Notes on Chapter 49: Product Insights, from "Empowered: Ordinary People, Extraordinary Products" by Marty Cagan, Silicon Valley Product Group.

Goal: Leverage insights on few critical problems to focus on.

This is one most difficult but rewarding aspects of product strategy. To generate, identify, and leverage insights that will provide the foundation of the product strategy.

You're probably already familiar with the insights led Netflix to profitability. Insights around Facebook new user onboarding. Customer trials Salesforce and Slack.

There is no clear framework for this!

First, understand:

• Product strategy requires real effort and thought. "Good strategy does not pop out of some tool, matrix, triangle...". Identify 1 or 2 critical issues, and focus action and resources on those.
• This never happens without real preparation. Epiphanies may come. But you need the prep work. Strategic context is the foundation.
• Insights can come from anyone and anywhere. Sales, industry analysis, random comments, academic papers. Without preparation, you wouldn't recognize them.

"You never know what might help connect the dots."

Four consistent sources of insights:

1. Quantitative Insights

Product Data. Data that may be related to the business model, such as customer retention, sales execution, etc. Which customers respond best to your product? Run an analysis. It may help you find more customers like that. Or may help you understand how to bring that dynamic to other customers.

Get data, run a test, learn an insight. Leverage this learning.

2. Qualitative Insights

User Research. May not be statistically significant. But often profound.

1. Evaluative - what did we learn from testing new idea?
2. Generative - did we uncover any new opportunities that we aren't pursuing, but should?

3. Technology Insights

Solving problems in new ways. Team might not have the knowledge and experience.

4. Industry Insights

Trends, competition, analysis.

Not just business strategy, but PRODUCT strategy.

Common Problems

Learnings are ignored because feature teams are focused on serving the business. Consultants don't have the depth and tenure to really do this effectively. It's important for Product people to own this.

Shared learnings.

Always discuss. Leadership and team level. Getting relevant insights into the right hands at the right time. These need to be shared and communicated. But reports, emails, IM's are rarely effective for this.

Product Leaders are usually the first to connect the dots. So we need make sure that insights make it to the product leaders. In many cases, leaders are given the data they request. Not the data they "need". Leaders should share with the broader organization.

Takeaways

• Focus on a small number of truly important problems
• Identify insights that will move the needle
• Turn these insights into action--> Vision Pivots...
  Things are not always linear
• Vision -> Strategy -> Execute
• During work, insights can (and should!) change everything.

Back in November 2019, I had the opportunity to interview with one of the biggest tech companies in the world. Google. The G in FANG. I didn't get the job. And I left with mixed feelings about the process and Product Management in general. But it was an unforgettable experience.

Now, in 2021, a little older, a little wiser, I'm ready for a retrospective.

First, a little about the process.

Since it was different than anything I'd experienced before.

The on-site interview for Product Managers takes place on the Google Campus in Mountain View, California.

After a few phone screenings, the recruiter arranged my stay at a Google-owned apartment property. The GSuites:

I stayed one night at the suite. I arrived on a Sunday night. Interviewed on Monday. And returned home to Los Angeles the same afternoon.

The interview was a full-day experience on Monday. I took an Uber ride to the one of the buildings on the Google Campus for check-in. I left my travel bags with the receptionist, and they called my interview coordinator to take me to the first meeting room.

Google's Campus was huge. It reminded me of an actual college campus.

I met 5 different product manages that day. Varying ages and personalities.

4 interviews. And 1 lunch buddy.

Looking back, here's what I learned.

1) Product Management means different things to different people. Different companies. Different cultures.

I didn't set out on my career with an intent to get in to Product Management.

I've always enjoyed being the bridge between people and technology. I was the helpful all-knowing "computer nerd" for friends and family. I worked in Technology Sales at Best Buy during college. I had an affinity for working with Microsoft Excel. But My Bachelor of Arts Degree in Music from Cal State Northridge didn't prepare me for the tech and business world I'd end up in a decade later. In the mid-2000's, these worlds were not even closely related.

I was first introduced to Product Management when working at AT&T Interactive/YellowPages. At the time, the organization was still run through a mix of both Waterfall and Agile methodologies. As a Product Manager on the Ad Platforms Team, I was responsible for writing the Product Requirements Document (PRD) based on whatever Business Requirements were already written. It wasn't a bad process. But it was siloed work. And it led to the sort of Product Delivery that Silicon Valley Product Management Group's Marty Cagen describes as "Feature Teams" rather than truly empowered Product Teams.

Was a I really a Product Manager? This became a bit of an existential crisis for me.

When talking to other Product Managers, interviewing for other roles, I was asked questions about "the customer", "shipping" the product, "go to market plans". Lots of areas that I had not experienced in my work as a Product Manager.

Yet I received consistent positive feedback from my peers and leaders. So I had to be doing something right.

What did I ultimately learn from this?

2) There is a difference between following your intuition vs having a process.

Even if it's good intuition.

Apple's late Steve Jobs was known for having some of the best product sense of any technology leader in history. But his leadership didn't come from having a solid framework for it. He was following his heart. He knew intuitively, based on his experience, what was the right thing to do, at the right time.

But he wasn't always right. And there can never be another Steve Jobs. Jobs rise was his own making. He garnered his own reputation. He was a different animal.

Google can't risk hiring someone based on pure intuition.

How can a company like Google mitigate this risk? By developing criteria to test candidates against.

Meaning:

3) You need to have a framework to pass Google's Product Management Interview

I came in on interview day not knowing exactly what to expect. The internet is a treasure trove of information. I've read a lot of books that have helped me in my career. "Measure What Matters" by John Doerr. "Inspired" by Marty Cagan. "The Lean Startup" by Eric Ries. I had over 15 years of helping people with technology in different roles from Sales, Data Analyst, Ops Manager, Project Manager, and Product Manager.

But none of this prepared me for the first interview question.

"How would you design an ATM for elderly people in Florida?"

Oh man. I did NOT see this question coming.

"Well, I like to start with writing a story about the user and their needs." That was my first response. But how could I even do that, without understanding the customer?

I talked about "who" this product was for. I tried to build a persona. I talked about doing a bit of discovery. And I ended up making some pretty embarrassing assumptions. I fumbled through the answer in a way that could have only confused the interviewer.

Check out this blog post for the framework I've decided to use for Product Design and Product Sense.

4) You need to know a bit about business. Markets. Estimations. Pricing. Strategy.

This one might be obvious to some product people.

But to me, a Product Manager who has primarily worked on Feature Teams, a lot of that upfront work around "the market" was already done for me. Other PM's and BA's were responsible for "the problem". I was responsible for the "the solution".

There was hardly any room for me to influence anything around what the actual business problem was. If the business decided that they want to launch a new product feature--it was my job to deliver.

"Imagine we've developed a product for dog lovers. It's an automatic treat dispense. How would you price it? How would you estimate the market?"

I started by thinking about the cost of the parts of such a product by thinking about similar products. Hypotheticals around how such a product would work.

But that just led to a dead-end.

Then, it came to identifying how many dog owners might be interested in such a product. I know a little about How to Measure Anything. But I couldn't even remember simple facts like the population United States, so that I could estimate the number of dog owners, etc.

The interview tried to help me--but I was stumped.

I needed a framework. See my blog post on estimation.

A chance encounter.

On the way back to the airport, I shared an Uber ride with a recent-graduate who had just successfully landed the Google job. She gave me the tip that really got me thinking.

Emotional intellect. Empathy. Servant leadership. Technical prowess. Storytelling ability. Data fluency.

You can't just possess these traits and expect them to carry you through the interview process. Not at companies like Google. You need to be prepared. You need to know you have these traits. But you can't just say it. You need to show it.

My Uber buddy recommended the book Cracking The PM Interview: How to Land a Product Manager Job in Technology.

And it hit me.

5) "How to be a great product manager" and "how to pass the product management interview" are two very different search queries.

I love my job. I love product management. I love helping people with technology.

But it wasn't until focusing on the actual process

Why the mixed feelings?

Is Product Management a "hot job"? I know now that the answer is probably yes. In some ways. Google CEO Sundar Pichai's rise-to-fame stemmed from his product prowess. So the fire starts naturally, when the leader of a multi-billion dollar conglomerate

But how could I have known this earlier in my career? I grew up disconnected from Silicon Valley. My dad was an electrician. I was the first in my family to graduate from college. But I was a music industry guy in my years after college.

The problem with hot jobs, is that they motivate entire micro-economies to form around the training and creation of.

Once you distill a complex job like Product Management down to a set of training programs--how can you develop the intuition without the experience?

I felt the same way about MBA programs. But I realize that my perspective is outsider's perspective. It's not wrong. It's just that I don't know everything I need to know. And I never will--that's just the way it goes.

But it's ok.

Key Takeaways

In summary:

1. Memorize and practice a few frameworks. Product Strategy. Product Design. Product Sense. Estimation.
2. Think about a few real-life stories from your own experience. Think about what was difficult them
3. Study. Practice.
4. Know that you can be a seasoned product manager, and still not pass a product management interview.

Nobody asked me anything about Agile frameworks. Project Management. Data mining and analysis. Getting the job done.

There was no focus on the "how".

Product was all about understanding the "why" and the "what".

There is no possible way I could have done better on my first in-person interviews. Not the first time. I needed to experience this myself. I needed to fail. I had resources. But I didn't have any guidance. So this experience felt like a miserable failure to me at the time.

But that's how we learn.

And now I see Product Management through different eyes. Wisdom comes from the mastery of many different perspectives.  And I aim to be wise.