[0:09] [Music]
[0:12] Let's explore one of the most practical
[0:14] skills when working with AI. Crafting
[0:16] effective prompts. This might sound
[0:18] technical or complicated, and some
[0:20] guides certainly make it seem that way,
[0:22] but at its heart, it's surprisingly
[0:24] straightforward. Prompting is simply how
[0:26] we apply this course's description
[0:27] competency in practice. clearly
[0:30] communicating what we want, how we want
[0:31] it done, and how we want to interact
[0:33] with our AI assistant throughout the
[0:35] entire process. Think of prompting like
[0:37] explaining a task to a helpful new
[0:39] colleague who's eager to assist, but
[0:41] needs clear directions and expectation
[0:43] setting to do their best work. We'll be
[0:45] using Claude throughout this section,
[0:47] but these tips can be carried over to
[0:49] many other AI systems. You might have
[0:52] heard the term prompt engineering tossed
[0:54] around. Prompt engineering is simply the
[0:56] practice of designing effective
[0:58] instructions for AI systems like Claude.
[1:01] It's about crafting your questions and
[1:02] providing context in ways that help AI
[1:04] assistants understand exactly what you
[1:07] want. What's fascinating is that
[1:09] effective prompting blends familiar
[1:11] human communication skills with a few
[1:13] considerations specific to AI. Many
[1:16] principles that make for good human
[1:17] conversation, such as being clear,
[1:19] providing relevant context, and giving
[1:21] concrete examples, also apply when
[1:23] working with AI. Yet, there are
[1:25] differences, such as being more explicit
[1:27] about things humans could naturally
[1:28] infer, and accommodating the AI's
[1:31] limited context window, and sometimes,
[1:33] depending on the AI you're working with,
[1:35] using specific formatting that machines
[1:37] can easily process. As AI assistants
[1:40] continue to evolve, prompting best
[1:42] practices evolve, too. What works with
[1:44] today's AI systems may be different from
[1:46] what works with tomorrow's.
[1:48] Experimentation is key to discovering
[1:50] what works best for your specific needs.
[1:52] In this video, we'll mainly explore six
[1:54] foundational prompting tips that will go
[1:56] a long way toward helping you
[1:57] effectively communicate and collaborate
[1:59] with Claude and other AI systems. They
[2:01] are give Claude context, show examples
[2:04] of what good looks like, specify output
[2:07] constraints, break complex tasks into
[2:10] steps, ask Claude to think first, and
[2:13] define Claude's role, style, or tone.
[2:15] The first principle is simple but
[2:17] powerful. Be specific and clear about
[2:19] what you want, why you want it, and
[2:21] perhaps most surprisingly, who you are.
[2:24] Let's take a simple prompt. Tell me
[2:26] about climate change. How can we improve
[2:29] this by giving Cloud more context? A
[2:31] more specific contextrich version could
[2:33] look like, explain three major impacts
[2:36] of climate change on agriculture in
[2:38] tropical regions with examples from the
[2:40] past decade. Our baseline prompt was
[2:43] vague and leaves clawed guessing about
[2:44] our interests, level of knowledge, and
[2:46] the depth of detail we're looking for,
[2:48] such as geography and time span. We can
[2:50] even add more context by providing
[2:52] information not just about what we're
[2:54] looking for, but why we're asking and
[2:56] how we'll be using that information that
[2:58] Claude gives us. Now, our prompt looks
[3:00] like this. Explain three major impacts
[3:02] of climate change on agriculture in
[3:04] tropical regions with examples from the
[3:06] past decade. I'm preparing for a job
[3:09] interview at an agricultural research
[3:10] lab in Indonesia. I have a degree in
[3:13] ecology, but no specific knowledge on
[3:15] climate change. write a summary of key
[3:17] concepts that would help me speak
[3:18] intelligently in the interview. All this
[3:21] added context helps tailor Claude's
[3:23] response to your specific situation and
[3:25] knowledge level. This kind of background
[3:26] information is something we naturally
[3:28] provide in human conversations, but
[3:30] might forget to include when talking
[3:31] with Claude. Sometimes showing is better
[3:34] than telling. Providing examples of the
[3:36] kind of output you're looking for can be
[3:38] incredibly effective. This is sometimes
[3:40] called fshot prompting or nshot
[3:42] prompting in technical circles where n
[3:44] is the number of examples given but it's
[3:47] really just about showing the AI
[3:48] examples for it to emulate. For
[3:50] instance, take the following prompt.
[3:52] Please convert this technical statement
[3:54] to plain language. The platform
[3:56] implements end-to-end encryption
[3:57] protocols to safeguard data integrity.
[4:00] Clog may already be able to do this to
[4:02] your satisfaction. So we definitely
[4:04] recommend you just try first without
[4:06] examples and see where it leads you. But
[4:08] let's say you have a very specific style
[4:10] you want Claude to follow, and it's
[4:12] harder to explain than to give examples.
[4:15] Your refreshed prompt could look
[4:16] something like this. Here are two
[4:19] examples of how to convert technical
[4:20] jargon into plain language. Original,
[4:23] the quantum algorithm exhibits quadratic
[4:25] speed up. Plain, the new method solves
[4:27] problems roughly twice as fast as
[4:29] previous methods. Original, the
[4:32] interface leverages intuitive design
[4:33] paradigms. Plain, the design is easy to
[4:36] understand and use. Now, please convert
[4:39] this complex technical manual to plain
[4:41] language. When providing examples, aim
[4:43] to cover the full diversity of possible
[4:45] prompts, such as examples that cover
[4:47] different cases or styles. This helps
[4:50] Claude better understand the broad range
[4:51] of the pattern you want it to follow.
[4:53] Being clear about output constraints,
[4:55] such as the desired format and length of
[4:57] Claude's response, or the language you
[4:59] want Claude to code in, or the color of
[5:01] the buttons on the web page you want
[5:03] Claude to design, also helps ensure you
[5:05] get exactly what you need. Here's an
[5:08] example of clear and detailed
[5:09] description to ensure Claude delivers
[5:11] exactly what you're looking for. Create
[5:14] a clean, modern, single page art
[5:16] portfolio website. Include these main
[5:18] sections: hero, about me, skills,
[5:19] portfolio projects experience and
[5:21] contact. Make the navigation menu sticky
[5:24] and responsive with hamburger menu on
[5:25] mobile. Use a sunset color palette and
[5:28] add a dark light mode toggle in the
[5:30] navigation. Guidance like this helps
[5:32] cloud structure its response to match
[5:34] your expectations. When you have a
[5:36] complicated request, breaking it down
[5:38] into smaller steps helps Cloud follow
[5:40] your thinking and deliver better
[5:42] results. Think about it this way. If you
[5:45] ask a friend to do something for you
[5:46] without specifying how, there's a chance
[5:48] that they may not do it the way you
[5:50] intended them to. We've all been there.
[5:52] Listing out task steps ensures that
[5:54] Claude follows the process you want to
[5:56] in order to accomplish its task. This is
[5:59] sometimes called chain of thought
[6:00] prompting. For example, instead of
[6:02] asking Claude to analyze this quarterly
[6:04] sales data, you might say, "I'd like to
[6:06] analyze this quarterly sales data.
[6:08] Please approach this by looking through
[6:10] our sales records to identify the top
[6:11] performing products, comparing current
[6:14] quarter results to the previous quarter,
[6:16] highlighting any unusual trends or
[6:17] patterns, and then suggesting possible
[6:19] reasons for these trends. By default,
[6:21] you may not need to do this, especially
[6:23] for tasks that are relatively
[6:24] straightforward. Furthermore, modern
[6:27] reasoning models or extended thinking
[6:29] models are increasingly capable of
[6:31] performing step-by-step reasoning on
[6:33] their own, but you can still guide this
[6:36] process to ensure it aligns with your
[6:37] needs. The more variance there is in
[6:40] ways to execute the task well, or the
[6:42] more that proper task execution relies
[6:44] on experience and knowledge you've
[6:45] gained as a domain expert, the more you
[6:48] should consider taking the time to
[6:49] translate that knowledge into Claude.
[6:51] Relatedly, sometimes it can be helpful
[6:53] to explicitly give AI assistants like
[6:55] Claude space to work through its process
[6:57] first before executing its task. This
[7:00] approach helps Claude produce more
[7:02] thorough and well-considered responses.
[7:04] For example, you can add this to your
[7:06] prompt. Before answering, please think
[7:08] through this problem carefully. Consider
[7:11] the different factors involved,
[7:12] potential constraints, and various
[7:14] approaches before recommending the best
[7:16] solution. As I mentioned, modern
[7:18] reasoning or extended thinking models by
[7:20] default think before acting. But if
[7:22] you're working with an AI assistant that
[7:24] does not think first by default, you can
[7:26] still prompt the AI to do so. I want to
[7:28] note the importance of giving the AI
[7:30] assistant space to think before doing
[7:32] its task, not after. If you want that
[7:35] thinking to increase the quality of the
[7:37] AI's work, just like how having space to
[7:39] think before you act is different than
[7:41] acting first, then being asked to
[7:43] explain your thinking afterwards. As a
[7:45] side benefit, this also allows you to
[7:47] better see where the AI assistant might
[7:49] be going astray and thus where you could
[7:51] hone your description competency further
[7:53] by providing more guidance. Specifying
[7:56] how you want cloud to communicate and
[7:58] behave can significantly change how it
[8:00] approaches a task. By specifying the
[8:02] level of expected expertise, the
[8:04] perspective you want it to take, or its
[8:06] communication style, you can guide both
[8:08] Claude's interaction with you and the
[8:10] final result of what it produces. Simply
[8:12] put, who do you want the AI to act as?
[8:15] For example, take this prompt. Please
[8:17] explain how rainbows form from the
[8:19] perspective of an experienced science
[8:21] teacher speaking to a bright 10-year-old
[8:23] who's interested in science. This is
[8:25] also a good way to brainstorm or get
[8:27] feedback. You can specify a general role
[8:29] or even ask Claw to take on the persona
[8:31] of a specific figure, such as Richard
[8:32] Fineman, when asking for physics
[8:34] explanations. Here's another example. As
[8:37] a UX design expert, review this website
[8:40] wireframe and suggest three improvements
[8:42] focusing on user navigation and
[8:43] accessibility. Perhaps the most powerful
[8:45] technique is asking Claude to help
[8:47] improve your prompt. When you're not
[8:49] sure how to ask for something or how to
[8:51] improve your prompt, describe to Claude
[8:53] your issue or situation and ask it to
[8:55] make your prompt better or write your
[8:56] prompt for you. I'm trying to get you,
[8:59] Claude, to help me with goal. I'm not
[9:01] sure how to phrase my request to get the
[9:04] best results. Can you help me craft an
[9:06] effective prompt for this? Here's where
[9:08] Claude and other AI assistants may vary
[9:10] most in terms of performance. So, we
[9:12] suggest you experiment with different
[9:13] models as part of practicing delegation.
[9:17] Effective prompting is iterative and
[9:19] experimental. AI systems and best
[9:21] practices are constantly evolving. So,
[9:23] what works today may change tomorrow.
[9:25] Your first attempt won't always yield
[9:27] the perfect result, and that's expected.
[9:29] When a response isn't quite what you
[9:31] need, try refining your approach by
[9:33] playing around with any of the
[9:34] techniques we mentioned, such as add
[9:37] more specificity or context. Provide
[9:40] examples of your desired output. Break
[9:42] the task into smaller steps and try a
[9:45] different technique or combination of
[9:47] techniques. You can also ask for
[9:49] variations such as, "Can you give me
[9:50] three different versions of this?" You
[9:53] can request different formats such as,
[9:54] "Instead of a paragraph, could you
[9:56] present this in an interactive
[9:58] artifact?" Note that artifacts are a
[10:00] unique way that Claude can create
[10:01] outputs that may be easier to understand
[10:03] or more interesting to digest. You can
[10:05] also check confidence, such as for
[10:07] factual questions, you can ask, "How
[10:09] confident are you about this answer?"
[10:12] You can also reset the conversation
[10:13] entirely. Sometimes starting a fresh
[10:15] conversation gives better results than
[10:17] trying to correct the conversation
[10:19] that's gone off track. Use each
[10:21] interaction as feedback to improve your
[10:23] next prompt. Over time, you'll develop
[10:25] an intuition for how to communicate
[10:27] effectively with all AI systems. As you
[10:30] apply these techniques in practice,
[10:31] here's some guidance to recap. Some
[10:34] patterns consistently work well.
[10:36] Starting with a clear task overview
[10:37] statement, including format
[10:39] specifications and examples, setting
[10:41] explicit constraints or requirements,
[10:43] providing rich and relevant background
[10:45] information, and common mistakes to
[10:47] avoid are assuming that Claude can read
[10:49] your mind, or overloading a single
[10:52] prompt or conversation with multiple
[10:53] unrelated tasks, being too vague about
[10:56] what success looks like, and not
[10:57] providing feedback on previous
[10:59] responses. To recap, effective
[11:01] communication with AI systems like
[11:03] Claude combines timeless human
[11:04] communication principles with AI
[11:06] specific techniques. The approaches
[11:08] we've covered will serve you well across
[11:10] different AI systems. These six
[11:12] principles together with the secret
[11:13] weapon of asking Cloud for help form a
[11:16] solid toolkit for applying the
[11:17] description competence to your AI
[11:19] interactions. Iteration and practice
[11:21] here is the key to Swift improvement and
[11:24] mastery. Remember that prompt
[11:26] engineering is an evolving practice. As
[11:28] models improve, some specific techniques
[11:30] become less necessary. However, these
[11:33] principles of good communication are
[11:35] still relevant even if the way we apply
[11:37] them changes. Maintain a spirit of
[11:39] experimentation and adapt your approach
[11:42] based on your results.