Learn · 13 series · 67 chapters

Explore the AI series.

Start from scratch or choose what to understand. Videos, visual explanations and sources along a clear path.

Start from scratch →
Find your path4 connected paths · start with any series

Understand AI

From the concept to how models generate and reason.

  1. 01AI and Generative AI Foundations
  2. 02From the Caves to AGI
  3. 03Multimodality in Generative AI
  4. 04Reasoning Models

Build systems

Agents, context, voice, code and inference.

  1. 08AI Agents
  2. 11Context Engineering, Memory & MCP
  3. 09Realtime Voice Agents
  4. 10Coding Agents & Agent Harnesses
  5. 12LLM Inference Engineering & Economics

Evaluate and protect

From tasks to permissions and evidence.

  1. 08AI Agents
  2. 07AI Security
  3. 13Evaluating AI Systems in Production

Understand the impact

Energy, wellbeing and infrastructure.

  1. 05AI, GDP, Well-being and Energy
  2. 06Data Centers in Space
AI and Generative AI Foundations · 4 chapters← All series
Series 01 / 13 · 4 chapters

AI and Generative AI Foundations

What AI is, how it learns and how it differs from generative AI.

Open video page and transcript ↗
AI and Generative AI Foundations
Before you startNo prior knowledge needed.Read the original introduction →

The path

4 chapters
  1. 01
  2. 02
  3. 03
  4. 04
Start with chapter 1 →See the learning map
From the Caves to AGI · 5 chapters← All series
Series 02 / 13 · 5 chapters

From the Caves to AGI

Follow the ideas from representing the world to training models.

Open video page and transcript ↗
From the Caves to AGI
Before you startAI and Generative AI FoundationsRead the original introduction →

The path

5 chapters
  1. 01
  2. 02
  3. 03
  4. 04
    Chapter 4

    Scale

    Read →
  5. 05
Start with chapter 1 →See the learning map
Multimodality in Generative AI · 5 chapters← All series
Series 03 / 13 · 5 chapters

Multimodality in Generative AI

Discover how text, images, audio and video connect.

Open video page and transcript ↗
Multimodality in Generative AI
Before you startAI and Generative AI FoundationsRead the original introduction →

The path

5 chapters
  1. 01
  2. 02
  3. 03
  4. 04
  5. 05
Start with chapter 1 →See the learning map
Reasoning Models · 5 chapters← All series
Series 04 / 13 · 5 chapters

Reasoning Models

Inspect what changes when a model spends more compute answering.

Open video page and transcript ↗
Reasoning Models — test-time compute, chains of thought and systematic failures
Before you startAI and Generative AI FoundationsRead the original introduction →

The path

5 chapters
  1. 01
  2. 02
  3. 03
    Chapter 3

    Test-Time Compute

    Read →
  4. 04
  5. 05
Start with chapter 1 →See the learning map
AI, GDP, Well-being and Energy · 4 chapters← All series
Series 05 / 13 · 4 chapters

AI, GDP, Well-being and Energy

Separate energy, economic growth and wellbeing when examining AI impact.

Open video page and transcript ↗
AI, GDP, Well-being and Energy
Before you startAI and Generative AI FoundationsRead the original introduction →

The path

4 chapters
  1. 01
  2. 02
  3. 03
  4. 04
Start with chapter 1 →See the learning map
Data Centers in Space · 4 chapters← All series
Series 06 / 13 · 4 chapters

Data Centers in Space

Explore the physical constraints of putting data centres beyond Earth.

Open video page and transcript ↗
Data Centers in Space
Before you startAI, GDP, Well-being and EnergyRead the original introduction →

The path

4 chapters
  1. 01
  2. 02
  3. 03
  4. 04
Start with chapter 1 →See the learning map
AI Security · 5 chapters← All series
Series 07 / 13 · 5 chapters

AI Security

Distinguish instructions from data and inspect which actions a system permits.

Open video page and transcript ↗
AI Security — attacks and defenses
Before you startAI AgentsRead the original introduction →

The path

5 chapters
  1. 01
    Prompt injection — when a document can change what the system does

    Can a document authorize an action?

    Read →
  2. 02
    Jailbreaks — when a model can be pushed into answering incorrectly

    Does refusing an answer also protect tools?

    Read →
  3. 03
    Poisoning — when a dangerous instruction stays in the system

    What happens when a contaminated source is stored?

    Read →
  4. 04
    Red teaming — test the full path before an incident

    What does an attack test establish?

    Read →
  5. 05
    Production controls — limit actions when the model fails

    Where should an unauthorized action stop?

    Read →
Start with chapter 1 →See the learning map
AI Agents · 5 chapters← All series
Series 08 / 13 · 5 chapters

AI Agents

Follow a realistic task: inspect, act and verify the result.

Open video page and transcript ↗
AI Agents — from answering to acting
Before you startAI and Generative AI FoundationsRead the original introduction →

The path

5 chapters
  1. 01
    What an AI agent is—and is not

    What changes when the tool fails?

    Read →
  2. 02
    The anatomy of an agent: tools, memory, and state

    Is a name enough to execute an action?

    Read →
  3. 03
    How to evaluate an AI agent

    Does a good answer mean the task is done?

    Read →
  4. 04
    Agent security: prompt injection, identity, and permissions

    Can external text change the recipient?

    Read →
  5. 05
    From demo to production: how to operate an agent

    When should the agent stop?

    Read →
Start with chapter 1 →See the learning map
Realtime Voice Agents · 6 chapters← All series
Series 09 / 13 · 6 chapters

Realtime Voice Agents

See turns, interruptions and the audio that actually reaches the listener.

Open video page and transcript ↗
Voice architectures: where the text boundary lives
Before you startAI Agents

The path

6 chapters
  1. 01
    Voice architectures: where the text boundary lives

    Why does architecture change the audio path?

    Read →
  2. 02
    Turn-taking: detecting speech is not deciding the turn

    Does a pause mean the turn is over?

    Read →
  3. 03
    Latency budget: measure the critical path, not dashboard sums

    Where is time spent before hearing a reply?

    Read →
  4. 04
    Tools and state: execute actions without breaking the conversation

    Does cancelling generation clear queued audio?

    Read →
  5. 05
    WebRTC, SIP, and telephony: follow the real audio path

    Why can audio arrive late?

    Read →
  6. 06
    Evaluating a voice agent: turn evidence, observability, and reliability

    Can a fast call still fail?

    Read →
Start with chapter 1 →See the learning map
Coding Agents & Agent Harnesses · 6 chapters← All series
Series 10 / 13 · 6 chapters

Coding Agents & Agent Harnesses

Observe code changes, permissions, tests and recovery on a concrete task.

Open video page and transcript ↗
What an agent harness is: the runtime that turns a model into a coding agent
Before you startAI AgentsContext Engineering, Memory & MCP

The path

6 chapters
  1. 01
    What an agent harness is: the runtime that turns a model into a coding agent

    What turns a proposed change into a verified one?

    Read →
  2. 02
    Context, workspaces, and sandboxing for coding agents: isolating state is not isolating execution

    Where can the agent write?

    Read →
  3. 03
    Specs and planning in coding agents: turning a request into a verifiable task contract

    What does completing a step mean?

    Read →
  4. 04
    Tools and permissions in coding agents: who can do what, with which credential, under which approval

    Is a proposed command authorized?

    Read →
  5. 05
    Tests and verifiers for coding agents: when the harness can say a task is done

    Does passing one example cover edge cases?

    Read →
  6. 06
    Long-running coding agents: memory, subagents, recovery, merge, and observability

    Where does an interrupted task resume?

    Read →
Start with chapter 1 →See the learning map
Context Engineering, Memory & MCP · 6 chapters← All series
Series 11 / 13 · 6 chapters

Context Engineering, Memory & MCP

Inspect which information is selected, stored and passed to the model.

Open video page and transcript ↗
Context engineering vs prompt engineering: what enters the model, when, and why
Before you startAI Agents

The path

6 chapters
  1. 01
    Context engineering vs prompt engineering: what enters the model, when, and why

    Does all available information reach the model?

    Read →
  2. 02
    Context budgets, prioritization, compaction, and provenance: what to preserve when everything will not fit

    What do you keep when context does not fit?

    Read →
  3. 03
    Agent memory architectures: working, episodic, semantic, and persistent state

    What persists between two turns?

    Read →
  4. 04
    Retrieval and context assembly: freshness, relevance, conflict, and grounding

    What happens when two sources conflict?

    Read →
  5. 05
    MCP: hosts, clients, servers, tools, resources, prompts, lifecycle, and trust boundaries

    Does connecting a tool authorize its actions?

    Read →
  6. 06
    Skills, plugins, subagents, and hooks: context isolation, authority, and evaluation

    What should a subagent receive?

    Read →
Start with chapter 1 →See the learning map
LLM Inference Engineering & Economics · 6 chapters← All series
Series 12 / 13 · 6 chapters

LLM Inference Engineering & Economics

Follow tokens, memory and time to understand model-serving cost.

Open video page and transcript ↗
Prefill vs decode: TTFT, TPOT, throughput, and the latency budget
Before you startAI and Generative AI FoundationsReasoning Models

The path

6 chapters
  1. 01
    Prefill vs decode: TTFT, TPOT, throughput, and the latency budget

    What happens before and after the first token?

    Read →
  2. 02
    KV cache, memory hierarchy, continuous batching, and PagedAttention

    What is reused for the next token?

    Read →
  3. 03
    Quantization, parallelism, and memory/performance/quality trade-offs

    What changes with fewer numeric levels?

    Read →
  4. 04
    Speculative decoding, prefix caching, and other latency optimizations

    When can a prefix be reused?

    Read →
  5. 05
    Model routing, fallback, caching, and workload-aware serving

    Must all tasks follow the same route?

    Read →
  6. 06
    Benchmarking inference: cost/task, throughput, latency, energy, and hardware constraints

    Are two latencies comparable under different workloads?

    Read →
Start with chapter 1 →See the learning map
Evaluating AI Systems in Production · 6 chapters← All series
Series 13 / 13 · 6 chapters

Evaluating AI Systems in Production

Inspect the evidence behind a metric and verify a task’s outcome.

Open video page and transcript ↗
What to evaluate: model, component, system, workflow, and trajectory
Before you startAI AgentsContext Engineering, Memory & MCP

The path

6 chapters
  1. 01
    What to evaluate: model, component, system, workflow, and trajectory

    Are you evaluating an answer or the system?

    Read →
  2. 02
    Offline eval sets: curation, hard negatives, contamination, and versioning

    What does an overly easy test set hide?

    Read →
  3. 03
    LLM-as-a-judge and human evaluation: calibration, bias, variance, and agreement

    Do the automated judge and reference agree?

    Read →
  4. 04
    Evaluating agent and tool trajectories: success, efficiency, recovery, and policy compliance

    Is reaching the goal enough if permission is violated?

    Read →
  5. 05
    Online evaluation: shadow, canary, A/B, guardrails, and regression gates

    What signal stops a rollout expansion?

    Read →
  6. 06
    Observability, failure taxonomies, and production → eval → repair feedback loops

    How does a failure become a reproducible test?

    Read →
Start with chapter 1 →See the learning map