One technical idea per video. Full evidence one click away.
Explore 46 native-English explanations published on 5sigmas. Every watch page connects the short explanation to its original chapter and related material.
46 videos available
0:48
Production Controls
Which controls limit damage when an AI system reads external content, uses tools and one defense fails.
Separate document reading from actions · Give every tool only the permissions it needs
0:48
Red Teaming
How to test from the document that enters the system to the action it can execute, before an incident occurs.
The threat model comes before the benchmark · Separate what the model can do from what the system executes
0:48
Poisoning
What happens when a document or memory preserves a dangerous instruction and the system uses it again later.
Storing a datum does not make it true · Persistent memory is already a measurable attack surface
0:48
Jailbreaks
How an attacker tries many ways of asking for the same thing and which controls remain necessary after a refusal.
Refusing a request does not create a perfect boundary · Trying many variants changes the cost of the attack
0:48
AI Security
A series on how an instruction hidden in a document can influence an AI system, how that risk can persist and which controls constrain actions.
Contents · 1. An instruction hidden in a document can change what the system does
0:32
From demo to production
What it takes to run an AI agent in production: budgets, retries, idempotency, observability, asynchronous work, fallbacks, and criteria for not using an agent.
Budgets before promises · Retries, idempotency, and terminal failures
0:32
Agent security
Why an agent that reads external data can be manipulated into acting, and which controls reduce risk: instruction separation, least privilege, approval, and auditability.
Direct and indirect prompt injection · Authorization must live outside the prompt
0:32
How to evaluate an AI agent
Evaluating an agent requires measuring the whole task, traces, tools, cost, and recovery from failure. A convincing final answer is not enough.
The unit of evaluation is a task · Four dimensions worth measuring
0:32
The anatomy of an agent
How an AI agent works internally: context, planning, tools, memory, state, and runtime. A tool call is a contract, not magic.
The minimal loop · Tool calling: from text to a contract
0:32
What an AI agent is—and is not
The difference between a chatbot, workflow, copilot, and agent. An agent is not just an LLM with tools: it is a system that decides actions within explicit boundaries.
A scale of autonomy · 1. Direct response
0:32
AI Agents
Five chapters to understand what an AI agent is, how it uses tools, how to evaluate it, which risks it introduces, and what it takes to operate one in production.
Contents · 1. What an agent is—and is not
0:53
The real footprint of a data center
Water, energy, minerals and data-center lifecycle: what AI consumes, why impact varies by location, and how to measure its footprint.
1. The comparison that calibrates the conversation · 2. Water: withdrawal, consumption and the technology that determines it
0:53
What a data center in space actually is
What processing data in orbit really means, which use cases make sense today, resilient storage and megaprojects with decades-long horizons.
1. What it means to process data in orbit · Processing observation data
0:53
Energy, heat and connectivity
Why the cold of space does not mean free cooling, where the real advantage of orbital energy lies, and what limits the connection to Earth imposes.
1. Why the cold of space does not mean free cooling · The real scale of radiators
0:53
Why now
Why orbital computing is being discussed now. The pressure AI compute demand places on terrestrial infrastructure and the limits that make space relevant.
1. The explosion in compute demand · 2. Terrestrial bottlenecks
0:53
Data Centers in Space
Technical analysis of orbital data centers: heat dissipation, latency, launch cost and real viability compared with terrestrial infrastructure.
Contents · 1. Why now
0:48
Prompt Injection
How an instruction hidden in a document can enter an AI system and which controls separate reading from action.
1. The problem starts when a document reaches the model · 2. The instruction can appear in a document search
0:20
Proactive and reactive agents and tool calls
A technical walkthrough of Reactive / Proactive Agent and the conversational contract it encapsulates: immediate response, asynchronous work, and deferred completion.
The problem is not the tool, but time · The first turn accepts work; it does not promise results
1:00
Reasoning Risks and Production Controls
Overthinking, prompt injection and agent hijacking in reasoning models with tools. Design criteria for bounding risk in production.
1. Overthinking: when more reasoning degrades the answer · 2. Quality vs cost vs latency in a real product
0:59
Latency, Streaming and Product Design
TTFT, streaming and perceived-latency thresholds in reasoning models. RouteLLM, design patterns and production session-cost management.
1. Perceived-latency thresholds · Dynamic routing: RouteLLM
1:00
Test-Time Compute
Test-time compute as a second scaling axis. The three levers—more steps, more candidates and more structure—and their quality, cost and latency tradeoffs in reasoning models.
1. What test-time compute is and why it matters · 2. The three levers
1:02
AI and GDP Today
Why AI's macroeconomic impact takes time to appear in GDP, where it appears earlier, and which signals are most indicative of what is happening.
1. Why macro impact takes time to arrive · The four mechanisms behind the lag
0:59
How Reasoning Fails
Sycophancy, shortcut learning, specification gaming and cascading failures: the failure modes of reasoning models, how to detect them and how to mitigate them.
1. Failure types · 1.1 Shortcuts — shortcut learning
1:02
Measurement: GDP vs Well-being
Why GDP does not capture real well-being, which dimensions matter most, and when subjective well-being diverges from material well-being.
1. What GDP measures and what it leaves out · What GDP leaves out
0:58
What It Means for an LLM to Reason
What reasoning means for a language model, what o1 and DeepSeek R1 added, and why evaluating reasoning requires looking at steps, cost and failure modes.
1. What "reasoning" means for a human · 2. What LLMs can do that looks like reasoning
0:45
Reasoning Models
Five chapters on how LLMs reason: test-time compute, failure modes, latency, cost and risks when models use tools.
Contents · 1. What "reasoning" means for an LLM
1:02
AI as an Electrical Technology
What AI implies in compute and energy terms, why demand can grow even as hardware improves, and where the real bottlenecks are.
1. What "compute" means in practice · Training
1:02
Electricity and Well-being
Why reliable, affordable electricity enables real gains in health, logistics and industry, and the difference between having kilowatts and having supply quality.
1. The four main channels · Health
1:02
AI, GDP, Well-being and Energy
Quantitative analysis of AI's impact on energy, productivity and well-being using real World Bank, IEA and Penn World Table data rather than speculative projections.
Contents · 1. Electricity → well-being: the real mechanisms
1:02
Risks of Multimodal AI Systems
Visual prompt injection, privacy, context leakage and tool manipulation when a multimodal system can observe and act.
1. Visual prompt injection · 2. System leakage and tool manipulation
1:02
Evaluating Multimodal Systems
How to evaluate multimodal systems without confusing benchmarks with real capability: OCR, audio, grounding, reasoning and metric failures.
1. What it means to evaluate grounding · 2. The problem of benchmark contamination
1:02
Multimodal System Architectures
Four multimodal architecture families, their differences in quality, cost and latency, and when each way of combining modalities makes sense.
1. Visual encoder + connector + language model · 2. Fusion through cross-attention
1:02
Alignment: From Pairs to Interactions
How models learn that different signals describe the same content, and why data quality determines the robustness of multimodal alignment.
1. Image–text pairs: the foundation and its limits · 2. Beyond the pair: aligning multiple modalities
1:02
The Real Problem of Multimodality
What it means to integrate text, images, audio and other modalities, and how perception, alignment, reasoning, generation and action fit together.
1. A modality is not only an input type · 2. The problem is not adding modalities, but crossing them without destroying them
1:02
Multimodality in Generative AI
What it means to build systems that can perceive, align, reason, generate and act across text, image, audio, video, documents and other signals from the world.
Contents · 1. The real problem: what counts as multimodality
1:03
Beyond the Transformer
How the field is trying to move beyond pure Transformer scaling by combining tools, search, inference-time memory, world models and robotics.
1. Why the Transformer is no longer a complete map · 1.1 Truth, uncertainty and hallucination
1:03
Learn — from rules to data
How AI moved from hand-written rules to learning from data: expert systems, statistics, neural networks and the AlexNet breakthrough.
1. The age of rules: when intelligence was written by hand · The first symbolic systems
1:03
Mechanize — from calculation to computing
How humanity automated calculation: from the first physical mechanisms to the separation of program and hardware, and the theoretical foundations of modern computing.
1. From automating calculations to programming procedures · The first calculators: automation is not programming
1:03
Represent — from counting to calculus
How symbols, numbers, algebra and calculus made it possible to represent and manipulate the world, building the mathematical foundation of modern AI.
1. We invented languages to describe the world · Before writing, we were already counting
1:03
From the Caves to AGI
An intellectual history of AI: from the first mathematical abstractions to foundation models. Mathematics, philosophy and computing in context.
Contents · 1. Represent (≈ 40,000 BCE – 1700)
1:03
AGI: Artificial General Intelligence
AGI means artificial general intelligence. This chapter explains its definitions, DeepMind's and OpenAI's levels, and what would still be required to reach it.
1. The definition problem · 2. The definitions in dispute
1:03
Classical AI vs Generative AI
Technical comparison between classical AI and generative AI: inputs, outputs, determinism, explainability, and when to use rules, ML, LLMs, RAG, or agents.
1. The five differences · 1.1 Inputs and outputs
1:03
What is Generative AI?
How generative AI works: from embeddings and the Transformer to foundation models. Scaling laws, LLMOps, and differences between LLMs, RAG, and agents.
1. Embeddings: Translating text into numbers · 2. The Transformer: The architecture that changes everything
1:03
What is AI?
What Artificial Intelligence is, how it works and how it evolved: from heuristics and Machine Learning to neural networks and foundation models.
1. The General Framework: AI, ML, DL and GenAI · 2. How do these systems learn?
0:53
AI and Generative AI Foundations
Introductory series on AI and generative AI: what they are, how they work, how they differ and what AGI means. For technical professionals and decision-makers.
Contents · 1. What AI is and how it evolved
1:03
Scale — deep learning to foundation models
A short video explanation of Scale — deep learning to foundation models.
1. 2012: when scale stopped being a detail · 2. The Transformer and massive pretrainingNo videos match this search.