Стоит ли учить ML в 2026? Или поезд уже ушёл?

Main message / lesson

• The speaker argues that ML/AI education is still worth it in 2026 and that the “train has left” idea is wrong.
• ML demand is framed as booming globally, with:
  • high hiring needs
  • relative job stability vs. some IT roles
  • lower barriers to entry thanks to tools like ChatGPT and accessible learning resources.
• AI is presented as augmenting work (especially routine tasks), not fully replacing the end-to-end responsibilities ML engineers/teams handle (infrastructure, iteration loops, production constraints, etc.).

---

Evidence and concepts used to support ML value

1) Growing demand for AI/ML-related roles

The speaker claims:

• At Stanford, AI is the most in-demand major, with many students choosing it.
• World Economic Forum analyses indicate the fastest growth among “data scientist” roles (with other related specialties also increasing).
• Hiring growth is described as very large (speaker mentions demand increasing “about four times year on year” in one market example).
• Analogy: early tech hype cycles (e.g., websites in the 2000s, finance/business education in the 1990s) were “real,” implying ML is the next such wave.

2) ML job resilience vs broader IT layoffs

• While large companies may lay off IT staff (examples named), data/ML work continues to be funded and hired because companies want profit from data.
• The speaker argues ML is “revolutionary,” making it relatively more protected from layoffs than parts of general IT.

3) “AI won’t replace programmers” — it changes what programmers/engineers do

The speaker addresses fears that code-writing automation (GPT-like tools) will eliminate developers:

• Historical analogy:
  • handwritten code → assemblers → object-oriented languages → GPT tools
  • Each productivity jump allegedly increased total number of programmers, because technology became cheaper and scalable.
• Core claim about AI limits:
  • GPT can handle small/routine pieces of work and drafts.
  • But it cannot (yet) reliably do end-to-end tasks such as:
    • full iterative development loops
    • scalable infrastructure meeting requirements
    • production-grade implementation from start to finish
• Conclusion: engineers remain necessary, but the role shifts toward iteration, systems integration, and production responsibility.

---

Instructional / career methodology (implicitly presented, plus explicit steps)

A) How to evaluate whether ML study is “still worth it”

Consider:

• University and industry trends (e.g., Stanford demand)
• Labor market reports (e.g., WEF-style analytics)
• Actual vacancy growth
• Hiring stability compared with other IT areas

B) How to position yourself for ML hiring (what “requirements” changed into)

The speaker describes shifting expectations over time:

• Earlier era (2018–2019):
  • Junior devs needed basic Python and ML libraries
  • Example tasks: fine-tuning models, plotting graphs, working with data notebooks
• More recent era (2020s):
  • Requirements shift toward deployment to production
  • Developers are expected to understand:
    • differences between training vs production data
    • correct model training
    • differences between offline vs online metrics
• Even more recent expectation (as described):
  • Juniors should understand:
    • why ML is needed
    • business use cases
    • the benefits and impacts of decisions

C) Practical hiring-focused learning pathway (stated through examples and “what it takes”)

• Entering ML may be possible with little/no prior coding:
  • Use AI tools (e.g., ChatGPT) to accelerate understanding and prototyping
  • Build projects (example: Kaggle-style beginnings)
  • Focus on ML basics and CS fundamentals rather than only tool-specific knowledge
• For people with an IT background:
  • Entry is easier because prior experience covers:
    • development skills
    • metrics/analytics thinking
    • communicating with business stakeholders
• Recommended “minimum foundation”:
  • Fundamental computer science is important (how computers/programming languages work)
  • Coding fluency alone isn’t sufficient—foundations support the whole stack (ML, deep learning, generative AI)

D) Role of ML tools in daily work

• Prompts/ChatGPT usage is described as widely applicable:
  • “in absolutely all jobs” where more thinking is required (except mostly manual tasks)
• Future expectation:
  • more specialized models will accelerate or replace parts of workflows (beyond text generation)

---

Where ML is used (main application areas listed)

• Finance and insurance
  • risk prediction
  • algorithmic trading
  • fraud detection
• E-commerce
  • supply chain/logistics optimization (example: “how Wildberries thinks”)
  • dynamic price formation
  • product demand forecasting
  • personal recommendations
• Healthcare and medicine
  • diagnosing diseases
  • drug research: speeding discovery (example claim: “more than 100-fold”)
  • sorting/searching chemical bonds more intelligently
• Production and industry
  • maintenance forecasting
  • flow optimization
  • quality control
• Marketing and user experience
  • segmentation
  • recommendation systems
  • targeting
  • collecting/analyzing users
  • building user-behavior models
• Energy and mineral exploration
  • site selection for oil rigs / minimizing the chance of unprofitable wrong locations
  • cost optimization and geographic problem solving (example attributed to work involving BP)

---

Forecast: what will grow in 2026 (categories emphasized)

• GenAI (Generative AI) / generative neural networks
  • text generation, video generation, photo generation, audio generation
  • example claim: GenAI replacing parts of computer graphics to reduce cost and speed production
• On-device / “secrets”
  • speaker meaning: optimization for small devices (e.g., running complex email-like or model-dependent systems on phones)
• ML Ops
  • growth driven by the need to bring heavy deep learning/generative models into production
  • requires highly skilled specialists (often transitioning from ML engineering and including operations responsibilities)
• Interpretability and safety of ML
  • increased focus on understanding models and ensuring safe usage

---

Microsoft “AI could replace jobs” claim (method described)

• The speaker references a Microsoft study and summarizes its approach:
  • evaluated ~200,000 requests (like prompting ChatGPT)
  • assessed the quality of outputs from an AI system responding to those requests
  • example outcome:
    • certain text-heavy roles (translator, historian) may be reduced because AI performs them well
• Caveat emphasized:
  • AI impact differs by job type:
    • some jobs require more than text—they require complex system thinking, creativity, and iteration/engineering practices.

---

Course promotion / training-related callouts

• The video includes a promotional segment for an ML base course:
  • aimed at analysts, developers, managers, and those considering ML/data science roles
  • includes a pre-registration link in the video description (per speaker)

---

Speakers / sources featured

Speakers

• Alexander Dyakovsky (video narrator/speaker; ML expert)

Named external sources/persons

• Andrew Ng (Stanford professor; lecture referenced; described as “father of machine learning”)
• World Economic Forum (report/analytics on job trends)
• Microsoft (study on which jobs AI could replace; ~200,000 request evaluation described)
• British Petroleum (BP) (referenced via an attributed employee example: “Valery Babushkin”)
• Avito (speaker’s current work context and an ML Ops example)
• Yandex Taxi (speaker’s previous experience)
• Google (examples: ML in Search; also claims about internal AI writing much code)
• Amazon, Walmart, Microsoft (examples of layoffs mentioned)
• HTHunter (analytics source referenced for Russia/CIS hiring context)
• Stanford University (demand/trends referenced)

AI tool / platform referenced

• ChatGPT / GPT (and “OpenAI” by implication as the model ecosystem referenced)
• Kaggle (used as an example of learning/handling data earlier)