Chroma. Kelly Hong walking through Chroma's research on 18 models — why needle-in-haystack scores are misleading, how performance degrades with ambiguity and distractors, why even simple string-repetition tasks degrade past 500 tokens. Short, evidence-based, and exactly the case the article needs you to take seriously before getting to the engineering moves.
Model names, pricing and capabilities change quickly. Use this for the decision pattern, then verify current model behavior before adopting it.
Understand how long context can fail under ambiguity and distractors, then design tests around that risk.
None beyond basic LLM vocabulary — it's a short research talk.
Last reviewed: May 18, 2026
Continue through the same learning path with the next curated companion videos.
18 minutesSee how approval gates can be implemented as explicit workflow checkpoints rather than informal manual review after something goes wrong.
4 minutesUnderstand why multilingual embeddings matter for private internal search and where local retrieval can reduce data-exposure risk.
17 minutesSet up a multi-agent coding workflow with explicit review boundaries so generated changes stay small, tested and owned.
Hand-picked external courses that go deeper on this topic.
DeepLearning.AIIsa Fulford · Andrew Ng
Ninety minutes to a year's worth of intuition. If you've started writing code that calls an LLM — or you're about to — this is the most efficient course online for closing the gap between "playing with ChatGPT" and "shipping a feature that calls an LLM."
Coursera · Vanderbilt UniversityDr. Jules White
The academic complement to DeepLearning.AI's short course — same discipline, longer arc, written for people who don't code. Dr. White teaches prompting as a set of reusable patterns (Ask for Input, Outline Expansion, Fact Check List, Menu Actions) rather than tricks. After this you'll prompt LLMs like a designer, not a guesser.
Microsoft LearnMicrosoft Learn
Prompt engineering, but for the tool most office workers will actually touch first. Microsoft's own four-part prompting framework (goal, context, source, expectation) is a genuinely useful mental model, and unlike the generic ChatGPT prompting courses already in our catalog, this one is grounded entirely in Microsoft 365 Copilot's specific quirks and grounding behavior.
