You're Asking Claude for Too LITTLE (Here's Proof) — Transcript

Full transcript

1. 0:00This gumball machine was made with one
2. 0:02simple prompt. You can put coins in
3. 0:04there. You can turn this crank, get
4. 0:05these gumballs out of here. You can see
5. 0:08through it. You can look at the labels
6. 0:10to figure out how it works. You can zoom
7. 0:12in to see all the different parts. Most
8. 0:14people think getting results like this
9. 0:16means writing super complicated prompts,
10. 0:18but it really doesn't. In fact, the guy
11. 0:20who runs engineering for Claude says
12. 0:21that most of us are just asking for way
13. 0:23too little. So, what's the difference
14. 0:25between asking Claude to give you a
15. 0:27paragraph explanation of a gumball
16. 0:29machine and something like this? That's
17. 0:31what we're going to explore in today's
18. 0:32video. My name's Casey. I ran a
19. 0:34marketing agency for over a decade, but
20. 0:36for the last 3 years I've been helping
21. 0:37teams push the limits of what Claude can
22. 0:39do using techniques like we're going to
23. 0:41go through today. But, I also want to
24. 0:42share a resource that can help you
25. 0:44manage super complex projects with
26. 0:46Claude. There's a link to that in the
27. 0:47description. Check that out if you're
28. 0:49interested. But, back to that gumball
29. 0:50machine. I'm also going to share a link
30. 0:52in the description of this video where
31. 0:53you can find a skill that will simply
32. 0:57allow you to describe an everyday object
33. 0:59and let AI draw an interactive
34. 1:02see-through version of it with labels to
35. 1:05help you understand these different
36. 1:07objects. It's projects like these that I
37. 1:09think really illustrate what this lead
38. 1:11engineer at Anthropic was trying to get
39. 1:13across when he wrote this article about
40. 1:15how to be more ambitious with our use of
41. 1:17AI. He breaks this down into basically
42. 1:20four categories. Speed is the one that
43. 1:22we're all probably most familiar with
44. 1:24using AI to get your job done faster,
45. 1:27get a task completed faster. So, we're
46. 1:29not going to focus too much on that one.
47. 1:30The other three are a lot more
48. 1:32interesting. Altitude is a big one where
49. 1:34we focus on a goal rather than a
50. 1:36specific task. Depth is another one. So,
51. 1:40when we are looking at the tasks to
52. 1:41complete that goal,
53. 1:43trying to complete them in a way that a
54. 1:45expert in that field might do it. And
55. 1:48then, range is another one that I think
56. 1:50is really great where we're looking at
57. 1:52all the different possible solutions,
58. 1:54all the different, you know, fields that
59. 1:56are out there and borrowing from a wide
60. 1:58variety of
61. 2:00of expertise that the AI models might
62. 2:03have or that's available to us.
63. 2:05This in particular reminds me of Charlie
64. 2:07Munger. He talks a lot about this
65. 2:09lattice work of mental models. So,
66. 2:11borrowing from the best of what every
67. 2:13single field has figured out and
68. 2:14avoiding being what Charlie Munger calls
69. 2:17the man with the hammer syndrome. So,
70. 2:20the man with the hammer basically sees
71. 2:23every problem as a nail. There are a lot
72. 2:25of different solutions to solve any
73. 2:26specific problem and that's where this
74. 2:29range comes into play. This color
75. 2:31figures example does a great job of what
76. 2:33Felix describes as creating the thing
77. 2:35that creates the thing. You could think
78. 2:37about, you know, asking Claude to just
79. 2:39explain how a washing machine works or
80. 2:41how an espresso machine works or even,
81. 2:44you know, maybe you you want to create a
82. 2:45graphic of that, but with this project
83. 2:48they went a
84. 2:49way farther than that and created this
85. 2:52version that you can now see inside that
86. 2:55anybody can use to
87. 2:58to install and, you know, explore
88. 3:01everyday objects
89. 3:02using Claude here and using this skill.
90. 3:04Another example here is a cardiologist
91. 3:07who built a 3D tool to help his
92. 3:09trainings learn how to read these
93. 3:11coronary angiograms and you could, you
94. 3:13know, think of the small version of
95. 3:15this, just, you know, here's some text
96. 3:17explaining how an angiogram works, but
97. 3:20he went definitely way further and built
98. 3:22something that his team could actually
99. 3:24practice on. So, that's another takeaway
100. 3:26that you might want to try instead of
101. 3:27just asking it to explain something, you
102. 3:29know, build something that you or your,
103. 3:31you know, co-workers can practice on and
104. 3:33work through. Other ways to start
105. 3:34thinking about altitude and pushing the
106. 3:36limits of all of this is asking Claude,
107. 3:38"What's the goal beyond the goal?" So,
108. 3:40instead of fixing this one bug, maybe
109. 3:42you want to say, "How do we eliminate
110. 3:44all bugs from this software or all bugs
111. 3:46of this type?" and build the, you know,
112. 3:48rebuild the software, rebuild the team
113. 3:50structure in a way to completely avoid,
114. 3:53you know, any of the problems that you
115. 3:54might run into on a day-to-day basis.
116. 3:56So, not just addressing the symptoms,
117. 3:58but potentially the root cause of these
118. 3:59different problems that you might be
119. 4:01running into. Another technique in this
120. 4:02area that I've been exploring is using
121. 4:04Claude to design better ways to do
122. 4:06quarterly updates and, you know, figure
123. 4:08out ways that it can detect problems
124. 4:10earlier and ideally back testing this on
125. 4:13old transcripts if if you have those
126. 4:15available and then putting different
127. 4:17forcing mechanisms in place to uh
128. 4:20enforce this going forward. Now, when it
129. 4:21comes to the category of depth or
130. 4:23actually building tools that a pro in
131. 4:26that area or an expert in whatever area
132. 4:28you're working in might um
133. 4:30want to work in, nothing comes close to
134. 4:31this wildfire analysis prototype that I
135. 4:34came across here. You know, obviously
136. 4:37this is way more robust than some sort
137. 4:39of a report about how a wildfire uh
138. 4:42moved through a valley. This was pulled
139. 4:44from actual data and allows people to
140. 4:46visualize in many different ways how the
141. 4:49fire spread with all these different uh
142. 4:51parameters such as wind and, you know,
143. 4:54dryness and all of these different
144. 4:56things so that you can uh play with the
145. 4:58data and learn a lot more about
146. 5:00potential future fires and how to
147. 5:03mitigate those wildfires that are, you
148. 5:05know, seem to be always present these
149. 5:07days. Another killer project here is
150. 5:09someone who built a new font from their
151. 5:12own uh handwriting. So, they took all of
152. 5:15their, you know, 120 pages of their
153. 5:17handwriting, they cleaned it all up and
154. 5:20they went through this various processes
155. 5:22of uh isolating the different uh letters
156. 5:25and building a complete font from this.
157. 5:27So, you know, rather than just finding a
158. 5:30uh handwritten font, this is going very,
159. 5:33very deep into building something
160. 5:35completely bespoke and completely unique
161. 5:38here and asking way more of Claude than
162. 5:41most people would. I'm not sure if you
163. 5:42saw my last video, but I chained
164. 5:44together a couple different tools to
165. 5:45build a very robust, um, competitor
166. 5:48dashboard. I'll I'll link to this uh in
167. 5:51the description. I used a technique
168. 5:52where uh I encouraged Claude to
169. 5:55continuously loop through uh and iterate
170. 5:58on um my request until it was really
171. 6:01happy with the result. And this really
172. 6:04pushed it to the next level. Uh this was
173. 6:06something that I really would have loved
174. 6:08to have when I was running my uh agency.
175. 6:10This is all updated whenever I want.
176. 6:13It's pulled from live data. Uh so, this
177. 6:16is another example of really pushing
178. 6:19things to the limit where I could have
179. 6:20just asked for a quick competitor
180. 6:21analysis.
181. 6:23Um you know, the next level up from that
182. 6:25might have been a basic dashboard, but
183. 6:27then building something that, um,
184. 6:29automatically updates and pulls from
185. 6:31various different uh tools here would
186. 6:33be, you know, really going in depth
187. 6:36with, um, you know, pushing Claude to
188. 6:38its limits. When it comes to
189. 6:39implementing this depth category or
190. 6:41pushing Claude to go deeper than ever
191. 6:43before, a couple things to keep in mind
192. 6:44might be insisting that Claude uses the
193. 6:47tools that pros use. So, that dashboard
194. 6:49that I just showed you actually pulls
195. 6:51from professional tools that I used in
196. 6:53my digital marketing agency, rather than
197. 6:56just, you know, the training data that
198. 6:57Claude has access to. Similarly, that
199. 7:00wildfire map pulled from, you know,
200. 7:02topographical resources. Another thing
201. 7:05to keep in mind is to pick an
202. 7:06unreasonable goal and treat it as a
203. 7:08reasonable, you know, Claude doesn't
204. 7:09know the difference, so you might as
205. 7:10well push it to go as deep as possible
206. 7:12that way. Another technique that can be
207. 7:14valuable here is challenging what you
208. 7:15think you know. So, if, you know, for
209. 7:17some reason your software tool might be
210. 7:19churning a lot, push Claude to find that
211. 7:21in the actual data, rather than just
212. 7:22resting on that assumption. Other things
213. 7:24that you might add to your prompts is to
214. 7:25ask Claude, you know, what would a
215. 7:27professional in this category ask for
216. 7:29that I didn't ask for here. You might
217. 7:31want to use the raw starting material,
218. 7:33like with that handwriting example, they
219. 7:35started with very raw information there,
220. 7:38rather than, you know, a blog post that
221. 7:40was about a blog post that was about a
222. 7:42blog post, you know, there's a lot of
223. 7:45drift out there in the world. So, going
224. 7:47and getting that raw data can be really
225. 7:49helpful. And always thinking what might
226. 7:50be an outside check you can run to prove
227. 7:53that this is right or to prove that this
228. 7:54is the right direction. The next
229. 7:56category here is range or pulling from
230. 7:58all sorts of different disciplines to
231. 8:00help you achieve specific goals. Before
232. 8:02we jump into that, I want to remind you
233. 8:04to grab this project management
234. 8:06resource. There's a link in the
235. 8:07description to that. I use this thing
236. 8:08every single day. It's a short video as
237. 8:10well as a template that you can use over
238. 8:12and over again with any different
239. 8:14project that you're working on. This
240. 8:16really helps me keep Claude
241. 8:17laser-focused on the work at hand. It
242. 8:19kills this document sprawl. As you know,
243. 8:21AI can create tons and tons of
244. 8:23documents. That can be a very, very big
245. 8:25problem as you work through these
246. 8:27different complex projects and you can
247. 8:30reuse this over and over and over again.
248. 8:32So, make sure to check that out. Now,
249. 8:34let's get into some ideas about range.
250. 8:37One of my favorite free tools in this
251. 8:38category is called Archify. I'm going to
252. 8:40have a link to this and all the other
253. 8:42resources in the description here.
254. 8:44Archify pulls the best of you know,
255. 8:47process documentation and process
256. 8:50mapping and pulls it into a format that
257. 8:52almost anybody can use. Consultants can
258. 8:55definitely use it. Marketers, anybody
259. 8:57dealing with complex systems as we all
260. 8:59are can use Archify to visualize
261. 9:04you know, their systems or the systems
262. 9:05that they're working on. So, this pulls
263. 9:07together a lot of different
264. 9:09resources. It pulls together from
265. 9:10design. It's beautifully designed and
266. 9:12pulls a little bit together from
267. 9:14you know, teaching as it shows people
268. 9:16these paths how different
269. 9:19pieces of software and systems connect
270. 9:22to each other. Here's what this looks
271. 9:24like. Allows you to really see some you
272. 9:26know, in-depth information and it's very
273. 9:29easy to build these. What I've been
274. 9:30doing is just you know, taking
275. 9:32transcripts from calls or meetings and
276. 9:34pulling those into Claude and pointing
277. 9:37it at this. It can, you know, map out
278. 9:39everything that we spoke about in a
279. 9:40visual way that everybody can agree,
280. 9:42yes, this is what we're talking about,
281. 9:43or no, this is not what we're talking
282. 9:45about. Can also use Claude in this way
283. 9:47to, you know, analyze something from,
284. 9:49you know, what might a CFO say about
285. 9:51this? What might a COO say about this?
286. 9:53What might a, you know, investor say
287. 9:56about this? And get, you know, various
288. 9:58perspectives on whatever you're looking
289. 10:00into. Uh, depending on what you're
290. 10:02working on, you can look at it in
291. 10:03different, you know, scientific levels.
292. 10:06You know, what might a chemist say about
293. 10:08this? What might a physicist say about
294. 10:09this? So, you can borrow from the best
295. 10:12of all of these different fields and
296. 10:13pull that into helping you to decide and
297. 10:16figure out what the best option, what
298. 10:18the best course of action for you is for
299. 10:19whatever you might be facing. You can
300. 10:21even take that one step further, not
301. 10:23just asking it, you know, what's what
302. 10:24somebody think, what would somebody say
303. 10:26about this, but pull it into actions,
304. 10:28pull it into verbs, you know, what would
305. 10:29that person do with this information?
306. 10:31What would that person want uh, from,
307. 10:34you know, a specific example. So, I
308. 10:37think there's a lot to be said about,
309. 10:38you know, changing out uh, nouns for
310. 10:41verbs. So, instead of, hey, we need to
311. 10:43create a deck, actually, what we need is
312. 10:45to get in line on a specific decision.
313. 10:49You know, what are the best paths to get
314. 10:50that to happen? It may not even be a
315. 10:52deck at all. There may be a many
316. 10:53different other routes you can take to
317. 10:55get to whatever your end goal might be.
318. 10:57Here's another really cool project that
319. 10:58combines illustration along with
320. 11:00animation and chess to create something
321. 11:02that really wasn't possible before AI.
322. 11:05So, I think this is another great
323. 11:06example of range and drawing from
324. 11:08different disciplines to create things
325. 11:10that were absolutely not possible
326. 11:12before. So, I'm going to link to all of
327. 11:14these things along with that project
328. 11:16management resource. That'll all be in
329. 11:18the description. And remember, when you
330. 11:19sign up for that project management
331. 11:20resource, we're going to be connected
332. 11:22via email, and you can respond to any
333. 11:24email that I send. I'd love to hear from
334. 11:26you with any challenges you might be
335. 11:28facing to see if I can help, or any wins
336. 11:30that you've had, anything that you may
337. 11:31have created, uh you know, that you're
338. 11:33proud of, send that my way. I'd love to
339. 11:35see it. It helps me improve these
340. 11:37videos. And if you haven't played with
341. 11:39the new Claude dashboard feature yet, I
342. 11:41created a video all about that. That's
343. 11:43your next step here. So, check that out.
344. 11:44I'll see you over on that video.
345. 11:46>> Make your dreams come true. True. True.