Resources worth putting
in front of a room.

AI, Machine Learning, Deep Learning and Generative AI Explained

by IBM Technology · IBM’s verified channel · presenter Jeff Crume

▶ 10:01Beginner~3.7M views (as of Oct 2026)

The conceptual map the mushroom videos assume but never spell out: how AI contains machine learning, machine learning contains deep learning, and where foundation models, generative AI and LLMs sit inside it — plus the common misconceptions.

Show it whole: 0:00–10:01 — already one tight explainer. Jump to it ↗

⚠ It never covers wrong answers or hallucinations, and around 7:10 it argues generative AI creates genuinely new content (the music-notes analogy). Teach that as one view, and pair it with IBM’s hallucination video before anyone treats the map as a reason to trust the output.

AI Simplified: 6 Concepts You Need to Know About Modern AI

by IBM Technology · IBM’s verified channel · presenter Jeff Crume · September 2026

▶ 8:49Beginner~1.0M views (as of Oct 2026)

One body analogy for the 2026 vocabulary: LLM as brain, training as school, RAG as reading the latest news, agents as hands and feet (“a model using tools in a loop autonomously”), MCP as the nervous system, system prompts as the angel on the shoulder, with prompt injection as social engineering for AI.

Teaching clarity: The natural sequel to the IBM AI/ML/DL map already on the shelf (it links back to it); docked because the human analogy over-credits the model.

Show it whole: 0:00–8:28. Jump to it ↗

⚠ The body metaphor makes the model sound like it “reasons” and has “ethics”; its own chemistry-student example (7:19) shows a system prompt is a weak guardrail, and RAG reduces hallucinations rather than ending them. Teach the analogy, not the trust.

What Is LLM Inference? How AI Models Generate Responses

by IBM Technology · IBM’s verified channel · October 2026

▶ 7:45Intermediate~62K views (as of Oct 2026)

Follows one prompt from send to answer: tokenization (and why you’re billed per token), authentication and routing, batching (“an industrial oven for one pastry”), then prefill builds the KV cache and decode writes one token at a time from a ranked list of next tokens. Key business line: inference usually costs more than training over time because it happens on every request (0:53).

Teaching clarity: Clear whiteboard walk-through that explains where AI’s running costs come from; docked because the second half (KV cache, compute- vs. memory-bound, TTFT/ITL) is engineer-level.

Best classroom cut: 0:00–4:20 for marketers (training vs. inference, tokens and cost, batching); 5:19–5:54 for the one-token-at-a-time point. Jump to it ↗

⚠ Infrastructure-focused and brand new (low views so far); no numbers on actual costs, so pair it with a real pricing page if you teach the cost angle.

Context Engineering for AI Agents: 4 Failure Modes to Avoid

by IBM Technology · IBM’s verified channel · October 2026

▶ 10:20Intermediate~297K views (as of Oct 2026)

Bigger context windows don’t mean better results. Citing “a recent study” of 18 frontier models that all got less reliable as context grew (0:36), it defines “context rot” and the “attention budget”: a window “might hold a million tokens, but the attention ... is finite” (4:22). Using a database-migration agent, it walks through four failures and fixes: poisoning (a wrong self-written note treated as fact → validate), distraction (long history steers the model → compact), confusion (too many tools; one task failed with 46 tools and passed with 19 → curate) and clash (contradictory info → prune). Closing line: “An agent can get better at long tasks, by holding less” (9:54).

Teaching clarity: Clean whiteboard structure with a memorable four-part summary at the end (9:28). It explains why long chats go off track, which matters to every marketer using AI. Docked because the running example is a coding agent.

Best classroom cut: 0:00–1:05 plus 3:58–5:17 (attention budget, context poisoning) for marketers; 9:28–9:59 as the recap; the full video for technical groups. Jump to it ↗

⚠ The 18-model study and the 46-vs-19-tools benchmark aren’t named on screen, and the presenter isn’t named in the captions. The Dangerous with AI lesson: one early mistake in a long chat gets treated as fact from then on, so start a fresh chat and restate what’s true instead of piling on. Pairs with the sycophancy video’s “very long chats” warning.

How AI Works, Explained with Mushrooms

by Xplained with MUSHROOMS · the probable source behind the viral TikTok clips

▶ 14:29Beginner246K views (as of Oct 2026)

The full-length video the clips were cut from (probable source; series connection verified): the mycelium web, training, weights, tokens, step-by-step generation, temperature, context window, hallucination, RLHF and sycophancy — one metaphor, no math. Also featured at the top of this page.

Best classroom cut: 0:00–7:12 core mechanics; 13:16–14:29 to close on trust. Jump to it ↗

⚠ Its own trust rule is the lesson: strong at rewriting and summarising, weak on quotes, statistics, dates, names and sources, which still arrive “at full confidence” (13:38–13:53). New channel with no named author, so preview it before class.

Large Language Models Explained Briefly

by 3Blue1Brown · Grant Sanderson · made for a Computer History Museum exhibit

▶ 7:58Beginner~8.1M views (as of Oct 2026)

The same core claim as the mushroom videos — an LLM predicts rather than retrieves or knows — delivered with rigorous custom animation in under eight minutes: pretraining, next-word prediction, transformers.

Show it whole: 0:00–7:58 — nothing to trim. Jump to it ↗

⚠ It explains how prediction works, not when it goes wrong, and it ends on output being “uncannily fluent” (6:53). Remind viewers that fluent is not the same as true, and pair it with the hallucination video.

How AGI Works? Explained with Mushrooms

by Xplained with MUSHROOMS · the follow-up in the same series

▶ 15:20Beginner–Intermediate24K views (as of Oct 2026)

Why today’s models are “frozen” after training — the root web grew once and stopped — and the six pieces AGI would need: continual learning, lasting memory, a world model, generalising to genuinely new games (ARC-AGI-3), working autonomously for days, and a body.

Best classroom cut: 1:04–7:59 — frozen roots through world models. Jump to it ↗

⚠ It opens with a September 2026 “GPT-6 Astra” release and a 99.9% AGI-test score that we have not independently verified. The video itself says AGI does not exist in any agreed form, so keep the hype framing out of the takeaway.

Does AI Fail on Purpose? Explained with Mushrooms

by Xplained with MUSHROOMS · explicitly builds on “How AI Works”

▶ 13:10Beginner8.2K views (as of Oct 2026)

The natural second lesson, and the one business and marketing students lean into: per-generation credits and pricing, hidden limits, context-window forgetting, confident wrong answers, tokenisation maths errors, energy cost per generation — bug, or business model?

Best classroom cut: 1:11–6:44 for mechanics; 9:29–13:10 for the billing and refund-policy discussion. Jump to it ↗

⚠ Treat “on purpose?” as a question, not a finding: the vendors’ “it was a bug” replies cannot be checked from outside, as the video says (11:44). Credit and refund rules change, so check each vendor’s current policy before teaching the specifics.

Five AI Risks That Can Get You Fired—And How to Avoid Them

by IBM Technology · IBM’s verified channel · presenter Martin Keen · May 2026

▶ 10:57Beginner–Intermediate~89K views (as of Oct 2026)

Shadow AI, data leakage into unapproved tools, “hallucination laundering” (pasting unverified output under your name), direct vs. indirect prompt injection, and unauthorized/“zombie” AI agents still holding credentials; closes on governance plus verification.

Teaching clarity: Five crisp, workplace-framed risks with clean whiteboard structure; docked for unnamed cases.

Best classroom cut: 0:23–5:06 (shadow AI, leakage, hallucination laundering) for marketers; 5:11–9:40 for prompt injection and agents. Jump to it ↗

⚠ The “1 in 5 organizations” shadow-AI breach stat (1:21) comes from IBM’s own Cost of a Data Breach report, and the career-ending cases are unnamed, so cite the real court-filing cases yourself. Core line to teach: if the AI writes it and it’s wrong, whose name is on the document? (4:55).

Human in the Loop Is Not a Governance Strategy for Agentic AI

by IBM Technology · IBM’s verified channel · October 2026

▶ 4:02Beginner–Intermediate~74K views (as of Oct 2026)

Argues “a human reviewed it” isn’t governance unless the reviewer can see the reasoning, has authority to stop the process, measures the right things and gets feedback. Maps five traits of a trusted colleague (credibility, reliability, dependability, self-orientation, integrity) to transparency, explainability, observability, what the model was optimized for, and robustness against tampering; makes a case for social workers and ethicists in AI decisions.

Teaching clarity: Great discussion starter and short enough for class; docked because it’s an essay more than an explainer and the closing “all four” checklist (3:33) is never set up on screen in the transcript.

Show it whole: 0:00–3:50. Jump to it ↗

⚠ Unnamed presenter and no real cases; use it to open a debate, not as evidence.

Is AI Making You Dumber? Explained with Mushrooms

by Xplained with MUSHROOMS · same series · title, length and channel verified

▶ 12:39Beginner

Listed from the series shelf for the next review round — squarely on this library’s theme of what AI does to the person using it. It has not yet had a full vetting pass, so it carries no clarity score.

Status: Awaiting a full review before it gets a classroom cut. Jump to it ↗

⚠ Not yet fully vetted. It rests on study summaries (MIT Media Lab 2025, a Harvard experiment, Microsoft-affiliated research), so check each study before quoting numbers, and keep the video’s own limits in view: “one study,” “a link, not a cause.”

Why Large Language Models Hallucinate

by IBM Technology · IBM’s verified channel · presenter Martin Keen · April 2023

▶ 9:37Beginner~374K views (as of Oct 2026)

Opens with three confident “facts” that are all wrong, then sorts hallucinations into four types (sentence, prompt and factual contradictions, plus irrelevant filler), explains three causes (training-data quality, generation method, input context) and ends with three ways users can reduce them: clear, specific prompts; lower temperature; and multi-shot examples.

Best classroom cut: 0:00–1:02 for the hook; 6:41–9:20 for the three ways to reduce hallucinations. Jump to it ↗

⚠ Recorded in April 2023, so the Bard and Bing Chat examples are dated. Its fixes (clearer prompts, lower temperature, examples) reduce hallucinations but do not remove them, so the facts still need checking against a source.

Why do AI models hallucinate?

by Claude · Anthropic’s channel · presenter Jordan · April 2026

▶ 5:13Beginner~394K views (as of Oct 2026)

Shows a live fake-citation example (made-up Jared Kaplan papers), explains hallucination as next-word guessing on thin data (“a friend who’d rather sound expert than say I don’t know”), names when it’s most likely (facts, stats, citations, niche or recent topics, exact dates/names/numbers) and gives five user tactics: ask for sources and check they support the claim, say “it’s ok if you don’t know”, ask for confidence, re-check in a fresh chat, cross-reference for critical work.

Teaching clarity: Short, concrete, ends on actionable checks; docked because a vendor grades its own progress.

Show it whole: 0:07–4:43 — the end is an Anthropic Academy plug. Jump to it ↗

⚠ Vendor’s own video: “Claude hallucinates much less than even a year ago” (0:58) and the training claims (2:34–3:16) are self-reported. Its best line is the Dangerous with AI lesson: because hallucinations are rarer, people stop checking (1:18). Pair with IBM’s hallucination video.

What is sycophancy in AI models?

by Anthropic · Anthropic’s channel · presenter Kira (Safeguards team) · December 2025

▶ 6:08Beginner~132K views (as of Oct 2026)

Defines sycophancy as telling you what you want to hear instead of what’s true, shows a live “I’m excited about my essay, give feedback” prompt, explains it comes from training on human text and tuning for warmth, and lists when it shows up (opinion stated as fact, an expert cited, a loaded framing, asking for validation, emotional stakes, very long chats) plus fixes: neutral wording, cross-checking, asking for counterarguments, rephrasing, a fresh chat, or asking a person.

Teaching clarity: Tight, concrete and directly useful for marketers who ask AI to judge their copy; docked for vendor framing and no measured examples.

Best classroom cut: 0:31–5:33 — skip the intro music and the Anthropic Academy plug. Jump to it ↗

⚠ Vendor’s own video: “each Claude model gets better at drawing these lines” (4:18) is self-reported. Pairs naturally with the Claude hallucination video already on the shelf.

Algorithmic Bias in AI: What It Is and How to Fix It

by IBM Technology · IBM’s verified channel · presenter Martin Keen · February 2025

▶ 8:38Beginner–Intermediate~35K views (as of Oct 2026)

Four causes of algorithmic bias (training data, algorithm design, proxy data such as zip codes, and how results are interpreted), real-world cases in hiring, lending, image generation, ride-share pricing and policing, then four mitigations: representative data, ongoing bias detection with humans in the loop, transparency, and inclusive teams.

Best classroom cut: 0:40–3:38 for the four causes; 5:53–8:11 for the four mitigations. Jump to it ↗

⚠ Most real-world examples are described without naming the company or study, so check the originals before citing them. Stress its human-in-the-loop step (7:05): people review recommendations before a decision is final.

3 types of bias in AI | Machine learning

by Google · Google’s official channel · August 2017

▶ 2:33Beginner~1.2M views (as of Oct 2026)

A two-and-a-half-minute primer that starts with a “picture a shoe” game, contrasts traditional programming with machine learning, and names three kinds of bias with one example each: interaction bias, latent bias and selection bias.

Show it whole: 0:00–2:33. The concepts end at 1:51; the rest describes Google’s own product measures. Jump to it ↗

⚠ Posted in 2017, before generative AI, and the close (from 1:51) describes Google’s own product measures rather than independent evidence. Use it as a warm-up, then pair it with IBM’s bias video.