If boot camps and college curricula taught real-world Data Science, this is how their programs would look. A student’s first assignment would be to answer a business question with Machine Learning and then deploy their model to an actual production environment. That model would fail, but the student wouldn’t know it until they got an F. They’d put monitoring in place to prevent them from being caught off guard again.

ChatGPT and Stable Diffusion have brought futurists and thought leaders (aka AI con artists) back to data science. The same thing happened in 2010 when Big Data became the buzzword bingo winner.

What happens next is up to us. We can all do 4 things:
1/7 1. Call out the copycats.

2. Point out snake oil.

3. Deny the doomsayers.

4. Curate and share expert content.
2/7

Microsoft's reported $10B investment in OpenAI has little to do with ChatGPT or GPT-4. For Microsoft, this is the latest step in a strategic journey they began in 2019. OpenAI is proving the viability of an emerging business model while Microsoft is teaching a masterclass in monetizing AI. This is a large bet on 3 key pillars of growth for Microsoft:

Change your approach to learning data science because the job you’re training for doesn’t exist. There’s no data scientist, just like there’s no software engineer. The term is a category. Target a phase of the data science lifecycle to build a realistic learning path. 1/5 Data Phase Common Roles: Data Engineers & Data Analysts

Stages:

1. Cataloging
2. Acquisition
3. Transformation
4. Metadata & Annotation
5. Data Quality & Cleaning
6. Sampling, Analysis, & Serving
7. Maintenance
2/5

If you're hiring for skills, your process is broken. In 12-18 months, critical skills for most roles will be different. Platforms, tools, frameworks, models, and programming languages evolve year to year.

Here's what we should be hiring for: 1/7 Businesses need people who:

1. Identify opportunities to improve early.
2. Learn with the opportunity and business impacts in mind.
3. Rapidly advance from knowledgeable to capable.
4. Apply learning to deliver business value.
5. Teach others to do the same.
2/7

Early career leaders struggle to write a clear, compelling resume. After reading it, I have no idea if they're a capable leader & can't hire them.

I'll teach you to fix that in 3 parts:
1. Common mistakes.
2. Critical resume focus areas.
3. How to explain leadership.
1/6 The most common leadership resume mistakes:

Focused on what was important to their last IC role.

Explains what they did, not why it matters.

Missing an explanation of their leadership style and core traits.

A lack of ownership for change management and team development.
2/6

Data scientists are finally building products with complex models, and we’re making all the same mistakes software engineers did 20 years ago. We’re packaging up the solutions to technical challenges and calling them products. 1/6 That approach has never made money. Today, most machine learning model supported products are still just interesting technology. They put the technology at the center and stall in the pilot phase. 2/6

Machine learning will be more successful when we build models to complete a task with people vs. complete a task alone. ML based products fail to be adopted because we build for model autonomy vs. human-machine teaming. 1/11 ChatGPT is a good example. It was built to answer questions alone, not answer questions with people. When I ask ChatGPT a question, I have an objective. ChatGPT needs the ability to ask follow up questions to understand my needs. 2/11

We're in the AI hype cycle phase where few data scientists can differentiate hype from substance or practical from vaporware. Any new model with a frontend is crowned the next great innovation without critical thought into feasibility or reliability in real-world scenarios.
1/9 The most significant moves in machine learning are happening in traditional sciences and research settings. ML researchers are chasing parameter counts while scientists are working to improve reliability. They are integrating models into their research process.
2/9

Data scientists spend significant amounts of time gathering requirements. Why? Most users don't know how to articulate their data and model needs. This is a critical part of data and model literacy training.

#DataScience #MachineLearning They are used to providing functional requirements, and models need something different—reliability requirements.

I have talked about the high-level forces at play, but it's worth explaining them with specific examples. Here are 12 companies that recently announced layoffs. Amazon is reducing headcount in unprofitable business units that won't do well in a downturn. Wearables are one example. This fits into the discretionary spending category.

A senior data scientist must have these 9 capabilities to be effective in a business setting.

1. Apply causal methods to business problems.

2. Experiment or study design to validate/refute models and assumptions in the data. 3. Build model reliability requirements from user or customer requirements.

4. Evaluate multiple trained models for new lines of exploration.

5. Production grade model development resulting in work products that integrate with the machine learning platform.

If I give you this data, what do you think will change? What will you be more successful at? The business makes hundreds, sometimes thousands, of data requests per year, and it’s critical for the data team to understand the objectives behind each one. I cannot be effective without understanding my users and stakeholders. The easiest way to start that process is to understand each data request.

Data teams expect the business to be data and model literate but fail to see the need for the data team to be business and strategy literate. Literacy training is a two-way street. Businesses are assessing data capabilities without the data team represented.
They are building data and AI strategies without consulting the data team.
They prioritize and sometimes estimate initiatives with 0 data team input. Why?

McDonald's CEO summed up senior leadership's sentiment on advanced machine learning during their July earnings call.

"Good at garnering headlines but not practical, and the economics don't pencil out." The comments were directed at robotics being used for workforce automation, but I have heard the same things said across industries and use cases. The current downturn presents a massive opportunity for our field to prove we can deliver significant value.

Impostor syndrome is common for new leaders. It has a powerful grip because we fear being revealed and rejected. It attacks the 2 most crucial team constructs. The first is our role and value, our sense of belonging. The second is acceptance from the team and validation from our leaders. We are not part of a team without acceptance and belonging.

Great decisions start with great questions. Do you want a simple, 3-level framework to focus your thoughts and come up with the right questions?

This is a critical leadership and product management trait that I can teach to you in about a minute. Level 1: Information Gathering. Ask broad, open questions about the events. I need enough of the facts before starting to formulate high-quality questions.

Why is all the career advice built for people at junior and mid-career stages? None of it works for people with 10+ year careers. What’s different? Resumes. It’s impossible to fit everything on a single page or even cover our most important capabilities. Senior++ people have bodies of work. They’ve moved beyond resumes and need a more polished professional presence.

Data Scientists leaving big tech companies are about to be swept up by traditional industries who are desperate for talent. Most won’t do well. Why? 1. The data science tools, data infrastructure, and IT support they depend on aren’t there.

2. The budget and data team size are much smaller.

3. Specialized roles don't exist and data scientists are expected to play multiple roles.

29 tech companies laid people off this week, and Apple announced a selective hiring freeze until the middle of next year. Every one of the Big Tech companies has implemented hiring freezes or layoffs. They’re all saying the same thing:
1/5 They are pausing less-critical initiatives and winding down unprofitable business units. Everyone is refocusing on core business or accelerating their path to profitability.

Data scientists, data engineers, and machine learning engineers face a strange hiring market.
2/5

This is the toughest job market in 10 years, especially for people looking for their first data science role. Here are the 4 main challenges data science job seekers face.

Starting salaries are dropping.

Fewer internships are converting to permanent roles. Companies are cutting back or dropping entry-level hiring altogether.

The qualifications and job requirements are becoming stricter.