Paste This Into Claude, Never Hit a Token Limit Again — Transcript

Full transcript

1. 0:00You keep running out of Claude or Codex
2. 0:02or or chat GPT or Kimmy or whatever you
3. 0:04want and you don't do anything
4. 0:05unreasonable to run out of tokens. You
5. 0:07asked a handful of questions and it told
6. 0:09you to come back in 3 hours or 5 hours
7. 0:11or next week. On one working day, my
8. 0:14tracker recorded 3.77 billion tokens
9. 0:17moving through my Codex workspace. Of
10. 0:20that, 3.59
11. 0:23billion were reused input, almost 96%.
12. 0:27And look, I push these tools really
13. 0:28hard. That was 143 separate Codex
14. 0:31threads in one day.
15. 0:32But I did not type 3.77 billion tokens
16. 0:36and neither did you. So, if the answer
17. 0:38fails, the retry carries most of all of
18. 0:41that a second time and a third time and
19. 0:42a fourth time while you're telling
20. 0:44Claude or you're telling Codex, you got
21. 0:45to fix this. So,
22. 0:48this is the core idea that every single
23. 0:51rule that I'm about to give you comes
24. 0:53back to. The message you typed is the
25. 0:55tiniest part of the overall call. I'm
26. 0:58Nate B. Jones. I'm here to help you
27. 1:00build the life you want with AI. That's
28. 1:02what I'm going to show you how to do
29. 1:03today. We're going to go through all 15
30. 1:05and by the way, if you're a beginner,
31. 1:06this is for you. You're going to go
32. 1:08through things you can do initially. I
33. 1:10built a skill so you don't have to
34. 1:11remember all of it and if you're a
35. 1:13little bit more technical, we are going
36. 1:14to get to a really cool multi-agent
37. 1:16automated solution at the end. So, stay
38. 1:17tuned for that. Your 10th message, your
39. 1:1920th message, your 100th message in the
40. 1:22chat cost many more times than your
41. 1:25first and nobody ever raised the price
42. 1:27on you. It's just that every time you
43. 1:28hit enter, the way LLMs work, the entire
44. 1:32conversation gets wrapped up in a bow
45. 1:34and sent again from the top. That's how
46. 1:36LLMs pretend to have memory right now.
47. 1:38Watch how fast that compounds. Your
48. 1:41first message costs exactly what you
49. 1:44typed. That's the easy part. Your second
50. 1:46costs what you typed plus the answer
51. 1:49plus what you originally typed. Your
52. 1:5010th costs what you typed plus all the
53. 1:53nine previous conversations you already
54. 1:55paid for. And by message 30,
55. 1:58the thing you actually wrote just now is
56. 2:00a tiny tiny rounding error. That old
57. 2:03material, it has a name. It's called
58. 2:05reused input. The part of every request
59. 2:09the model has already seen. So, two
60. 2:11things to call out. One,
61. 2:13this is not getting fixed by the labs.
62. 2:17The labs are not going to magically fix
63. 2:19this tomorrow. It is up to us.
64. 2:21The labs aren't going to fix it because
65. 2:22frankly, they have an incentive to get
66. 2:24us using the product. And if we're using
67. 2:26that's great up to a point, right? If
68. 2:28they run out of compute, they're going
69. 2:29to sort of rein it back. But
70. 2:30fundamentally, the labs want us to use
71. 2:34their tokens. So, it's up to us to make
72. 2:36the most of the limits they give us. AI,
73. 2:39it's like this desk. And there's a
74. 2:41belief going around
75. 2:43that this is going to fix itself right
76. 2:44now. And that our desks can clean
77. 2:46themselves and the models are going to
78. 2:48get bigger context windows and they're
79. 2:50going to run longer and have more agents
80. 2:52getting more tools. And soon you can
81. 2:53point the thing at your work. And it's
82. 2:55just going to run, right? We In other
83. 2:57words, we believe we're going to be able
84. 2:59to have a messy desk and it's going to
85. 3:01be fine. But if I take all the stuff off
86. 3:03of my shelves and put it on the desk,
87. 3:05I'm not going to be very productive. I'm
88. 3:07going to be overloaded and stressed and
89. 3:09it's not going to work. It just doesn't
90. 3:10work that way. I have to pick one Lego
91. 3:12set for my desk. The more capable the
92. 3:15tool, the more tools you hand it, the
93. 3:17more material it puts in and the faster
94. 3:20you hit the wall. So, ironically, more
95. 3:21capability may end up giving you more
96. 3:24cleanup problems. And I talk about all
97. 3:26three in this video. But none of it
98. 3:29really takes the responsibility of
99. 3:31organizing your AI desk for you. And
100. 3:34reducing your token consumption is not
101. 3:36what any of these companies is graded
102. 3:38on.
103. 3:39It's your desk. You have to own it. Is
104. 3:42it fun to clean your desk? It can be
105. 3:43kind of boring, right? It's almost
106. 3:45always the boring stuff though that
107. 3:47gives you the superpowers. So, level one
108. 3:49is for everybody. It's how you keep your
109. 3:51desk clean. Nine habits, nothing that
110. 3:54you need to install, will work on any
111. 3:56AI. And some of it is advice we've known
112. 3:59for a while, and I'll tell you when it
113. 4:01is, but guess what? Old advice may still
114. 4:03be something you need to hear if it's
115. 4:04not something you're following right
116. 4:06now.
117. 4:07Then we're going to get to level two.
118. 4:09Level two is kind of like hiring someone
119. 4:11to keep your desk clean. I built a
120. 4:12skill, it's called Token Saver. It
121. 4:15installs into Codex, it installs into
122. 4:16Cloud Code, and it does most of level
123. 4:19one for you while you keep working
124. 4:21normally, which is kind of handy. Level
125. 4:24three is stopping the mess before it
126. 4:26ever gets to your desk. And that's a
127. 4:28software piece that I've been building
128. 4:30around my Ringer Multi-Agent Framework.
129. 4:32It's the most powerful option by far,
130. 4:35and I will tell you exactly where the
131. 4:36edges are and how it works when you are
132. 4:38ready at the end of the video. Level
133. 4:40one, clean your own desk. Rule number
134. 4:44one, you've got to edit your mistakes.
135. 4:46So, if you're writing something in AI
136. 4:48and you're like, "Oh, man, I didn't mean
137. 4:49to write that. I had a typo. I asked the
138. 4:51wrong thing." Do not say that was wrong
139. 4:53in the next chat. Just edit it, which
140. 4:56you can do, and resend the message. And
141. 4:58you may have heard of this, but almost
142. 5:00nobody does it. The model will give you
143. 5:02something wrong because your request was
144. 5:03unclear, and your instinct is to just
145. 5:05say, "No, that's not it." Well,
146. 5:08instead, hit that little edit button,
147. 5:11and make sure that you are actually
148. 5:13correcting the unclear request you had
149. 5:15before, and get it right. Rule number
150. 5:17two, ask related questions together, and
151. 5:21say how you want the answer to appear.
152. 5:24Not new advice, again, but absolutely
153. 5:26worth a minute for you to realize. If
154. 5:28you have multiple questions from the
155. 5:30same document set, put them all into one
156. 5:33query, and then specify what you want at
157. 5:36the end. What does it look like? Is it a
158. 5:37one-pager you want? Is it 150 words? Is
159. 5:41it just give me the bullets? Is it give
160. 5:42me the headline? Name it and say it.
161. 5:45Because that way you are reducing the
162. 5:48ambiguity that the AI is going to spend
163. 5:50tokens on to give you answers on. So,
164. 5:52just name what your questions are all at
165. 5:54once. Don't just string them along as
166. 5:55multiple questions and name the answer
167. 5:58you want in advance. Rule number three,
168. 6:00start a clean task when the job changes.
169. 6:04And this is this is a huge win and
170. 6:06people seem to resist this. I know, in
171. 6:08fact, a lot of people who believe they
172. 6:10are in romantic relationships with AI
173. 6:12because they didn't bother to do this.
174. 6:14Long conversations are really good while
175. 6:17you're focused on the same problem, but
176. 6:18they're really terrible for trying to
177. 6:21deal with token usage when you're trying
178. 6:25to get specific questions answered. And
179. 6:27we get away with this more now cuz
180. 6:28models are smarter. And so, you can
181. 6:31disambiguate more in that long context
182. 6:33window, but it's really token heavy and
183. 6:35you hit your usage limits faster. And
184. 6:37this drove the biggest measured change
185. 6:38when I was testing all of these out on
186. 6:41my own Codex and Claude installs. When
187. 6:43you are carrying a working conversation,
188. 6:46you carry so many reused tokens with
189. 6:49you. Like, you can carry 50,000,
190. 6:51100,000, a million, or as I've been
191. 6:53saying, if you're getting into the
192. 6:54billions, it's hundreds of millions of
193. 6:55tokens with you. Start from scratch.
194. 6:59Now, a clean task doesn't actually
195. 7:01report zero every time because, of
196. 7:03course, Codex and Claude will still send
197. 7:05their own starting instructions.
198. 7:07But what it does is it stops the old
199. 7:09conversation from writing along.
200. 7:12And you're not throwing the old task
201. 7:13away. You can keep it. Just stop making
202. 7:15the next job that you that you want it
203. 7:17to do carry that task. Rule four is
204. 7:19related. Carry the answer that you care
205. 7:22about, not the argument. Let's say you
206. 7:24have a multi-step process. You have a
207. 7:25research piece, you want to have a
208. 7:27document writing piece. This was big for
209. 7:29me. Make sure that you separate out the
210. 7:33stages of the job and that when you have
211. 7:35an artifact that's produced after a
212. 7:37stage, like a research report, that's
213. 7:39what you carry forward into the next
214. 7:42step so that you only have that to give
215. 7:45to your AI to start. You don't have
216. 7:4810 million, 100 million tokens of
217. 7:50research that you're handing, much of
218. 7:52which is irrelevant because you had to
219. 7:54guide the research along the way. Be
220. 7:56precise.
221. 7:58You don't need to stack your bad first
222. 8:00draft and your three rounds of criticism
223. 8:02and the sources you rejected and the
224. 8:04model's reasoning all into the next
225. 8:05step. You can actually just take the
226. 8:08result you got and move on to the next.
227. 8:10Keep your desk clean. Rule five, ask for
228. 8:13only the answer you need. Input gets a
229. 8:15lot of attention because it gets so big,
230. 8:17but output costs you twice. It costs you
231. 8:19once when it's written very expensively
232. 8:22and again when it becomes part of the
233. 8:23input on the next turn and again when
234. 8:25there's a turn after that and again when
235. 8:26there's a turn after that, you keep
236. 8:27getting billed on it. So, remember, when
237. 8:30when we first got like the capability to
238. 8:33write a lot from AI, I saw a lot of
239. 8:35people who would go tell deep research
240. 8:37to write them 50-page papers. But, what
241. 8:40I see a lot more of now is people
242. 8:42saying, "Can I get it in 50 words? Can
243. 8:44Can you please write it precisely in a
244. 8:46way that I can understand it and give me
245. 8:48what matters?" We need more of that
246. 8:50because when you do that, you're saving
247. 8:52not just on the output tokens, you're
248. 8:54saving on every single response that
249. 8:56comes after that. And so, ask exactly
250. 8:59for what you need. If you need a
251. 9:00paragraph, ask for that. If you need
252. 9:02JSON, ask for that. If you need five
253. 9:04bullets, get five bullets. Rule six is
254. 9:06really simple and it's not something
255. 9:07people understand a lot. You want to
256. 9:10search the file yourself whenever you
257. 9:12can and don't make the model search the
258. 9:13file. Yes, the model can search the file
259. 9:16now. It's a great convenience, it's also
260. 9:18a massive token burner.
261. 9:20You want to be in a position to say,
262. 9:23"Hey, I searched it. I found these
263. 9:25things. This is what you should focus
264. 9:27on. This is what I'm including in this
265. 9:28snippet for you. You don't need to go
266. 9:30read this entire file." Rule seven is
267. 9:32related. You want to send the lightest
268. 9:34useful form of the source. In other
269. 9:37words,
270. 9:38if the words matter and the layout
271. 9:40doesn't, just convert it to text. I've
272. 9:43talked about this before. You don't have
273. 9:45to send a PDF just cuz the source came
274. 9:48to you in a PDF, convert it to markdown,
275. 9:51convert it to text, and just paste the
276. 9:54text in. It's so much more efficient.
277. 9:57Don't just be lazy and say, I can throw
278. 10:00a PDF and 18 screenshots in and it's all
279. 10:02fine and it will sort it out because as
280. 10:04tempting as it is, cuz it does sort it
281. 10:06out, eat your token bill so so fast.
282. 10:09Make sure that you take the time to
283. 10:12actually get your sources in order.
284. 10:15Clean your desk. It's common theme. Rule
285. 10:17nine ties into those of you who have
286. 10:19open brain. Keep your answers around
287. 10:21somewhere you can find them. And like if
288. 10:23you have open brain, it may be in your
289. 10:24open brain database or you may have your
290. 10:26own database that you use, but really
291. 10:29really important. If you are working on
292. 10:32a particular problem set and you're
293. 10:33working around the edges of it, you have
294. 10:35multiple conversations. The more you
295. 10:37keep information about that in a place
296. 10:39you can look it up easily, the more the
297. 10:42AI is going to be able to look it up and
298. 10:44not have to go recreate it and not have
299. 10:46to go dig for it in the sources and
300. 10:47recalculate it. It's just going to be
301. 10:49able to get that particular piece of
302. 10:50data out of the database and be done
303. 10:52with it and that is so so much more
304. 10:53token efficient. So, if you're not using
305. 10:56open brain, you can check it out. I have
306. 10:58a lot of videos on it. I can link it
307. 11:00here. It's super easy to get started on,
308. 11:02but make sure that you have a system
309. 11:04that allows you to go and retrieve data
310. 11:07for stuff that you would look up
311. 11:08multiple times otherwise because the
312. 11:10cheaper you can make that, the more
313. 11:12you're saving tokens. Let's say you're
314. 11:14like, "Nate, I love this. I'm lazy. I
315. 11:17don't do this. I don't remember to do
316. 11:19this. Please help me." That's what I
317. 11:21built the skill for. I built the skill
318. 11:23called token saver. One command will
319. 11:25install it. You can get it into Codex,
320. 11:27you can get it into Claude Code,
321. 11:28wherever you do your skills, and then it
322. 11:30will keep working with you the way you
323. 11:33already work. And so, you can just say
324. 11:34use the token saver skill for this job
325. 11:36and it will handle a lot of the tedious
326. 11:38parts of level one for you. It'll search
327. 11:40before opening large sources. It will
328. 11:43send selected passages instead of whole
329. 11:45files. It will run exact work as code
330. 11:47wherever it can. It saves the version
331. 11:49you accepted and builds your next
332. 11:51request from that result plus your
333. 11:52change. It keeps answers to the length
334. 11:55that you asked for and it stops
335. 11:57pointless retries over and over. So,
336. 11:59there's a good things about it and I
337. 12:02want you to use it and find ways to
338. 12:04continue to improve it. And that's
339. 12:05something that I love about our
340. 12:06community in the slack is that we find
341. 12:08ways to improve each other's skills. So,
342. 12:10I'm launching this. I've tested it. My
343. 12:12team tested it. We love it. I'll also
344. 12:14call out that this helps you with the
345. 12:16next three rules I'm about to give you
346. 12:18that are really difficult to do by hand.
347. 12:20And so, I want you to like listen to
348. 12:22these, understand them, but realize the
349. 12:24skill is going to help you get there.
350. 12:26So, rule 10, load only the tools the job
351. 12:29can use. Now, this is something where I
352. 12:31quite frankly in six or eight months
353. 12:34expect the models to be good enough to
354. 12:36solve this problem, but they're not
355. 12:38reliably good enough at it today. Every
356. 12:40tool you connect carries a description
357. 12:42right now. What it does, when to use it,
358. 12:45what arguments it takes. That
359. 12:47description is model input before the
360. 12:49model does anything at all.
361. 12:51Now, Anthropic has published some work
362. 12:54that makes this whole problem space
363. 12:56very, very real. A typical setup with
364. 12:58several tool servers connected like say
365. 13:00GitHub and Slack and Sentry and Grafana.
366. 13:03It burns roughly 55,000
367. 13:06tokens in tool definition
368. 13:08before Claude does anything. Now, we're
369. 13:10starting on more advanced rules. This is
370. 13:13stuff where the skill can be supported,
371. 13:14but you also have to use your head.
372. 13:16Sometimes starting clean on a thread
373. 13:18like I recommended in the sort of the
374. 13:20beginner section, it's not very
375. 13:21practical. Let's say you're debugging a
376. 13:23particular system and the model needs a
377. 13:25decision from you and you can't restart
378. 13:28it because it's in the middle of the
379. 13:29task. This is an area where compaction
380. 13:32and context editing become really
381. 13:33important. OpenAI supports compaction
382. 13:36for long-running work, and it carries
383. 13:37forward the state, and it turns out it
384. 13:40needs fewer tokens as a result, but
385. 13:41you're depending on their native
386. 13:42compaction capabilities. Anthropic
387. 13:45supports context editing, which clears
388. 13:47old tool results out and and clears
389. 13:50thinking blocks before the next request,
390. 13:52and that's also super helpful. And so,
391. 13:54the skill helps with it with that, but I
392. 13:55also need to be honest with you that
393. 13:56this is just something you as an
394. 13:58advanced AI user should be aware of. You
395. 14:00should know where your context window
396. 14:01is, and you should recognize that when
397. 14:05you clear out old material, you are
398. 14:07depending on a approximated version of
399. 14:12the initial prompt, approximated
400. 14:14versions of the initial responses that
401. 14:17the model makers are using to enable you
402. 14:20to continue work on a long-running
403. 14:21thread.
404. 14:23That is not perfect, but it's a whole
405. 14:24lot better than just hitting a wall,
406. 14:26which is what we used to do. So, the
407. 14:28takeaway for you is use the skill,
408. 14:30understand where your context window is,
409. 14:32and then make sure that you actually
410. 14:35anticipate in your work
411. 14:38the consequences of hitting that wall.
412. 14:40Put the signal early. The skill also
413. 14:43helps when you want to figure out what
414. 14:45model to use because sometimes a smaller
415. 14:48model helps you get the job done
416. 14:49cheaper, but you have to trade off the
417. 14:51context, etc. So, what the skill does is
418. 14:53what it looks at the at the question or
419. 14:55the problem you're tackling, and it
420. 14:57comes up with, at least an initial
421. 14:59opinion on whether the model you're
422. 15:01using is correct or not. You can
423. 15:02obviously disagree, you can move on, you
424. 15:04can say, "No, I want to use this model,"
425. 15:05but at least you have a first blush
426. 15:07approximation at what a token-efficient
427. 15:10model solution is. I I like to say, like
428. 15:13if you're doing a serious task, use the
429. 15:16absolute dumbest model that will still
430. 15:19get the work done for you. And the more
431. 15:21you work with AI, the more you have a
432. 15:23feel for what that line of dumbest model
433. 15:25is. And so, the skill just kind of helps
434. 15:27you get there, helps you take a guess at
435. 15:29that if you're new at that. I hear a lot
436. 15:31about prompt caching. That's what we're
437. 15:32talking about with this rule. Prompt
438. 15:34caching is really, really important if
439. 15:37you're doing repeated work. It's
440. 15:38especially important with API work. It's
441. 15:40not something that I would recommend for
442. 15:42people who are doing just initial desk
443. 15:44work. It's not a clear your desk feature
444. 15:46if you're a regular knowledge worker.
445. 15:47It's something where it's an API
446. 15:49feature. You're going to be in a
447. 15:50position where you can cache a prompt or
448. 15:52cache part of what you're sending, and
449. 15:54that makes it much, much more efficient
450. 15:56to send because you're not sending the
451. 15:57whole message back and forth. And that's
452. 15:59really all you need to know. And if
453. 16:00you're someone who's already diving into
454. 16:01this, you're like, "Yeah, Nate, I do
455. 16:03prompt caching, or at least I know what
456. 16:04prompt caching is." And you're off to
457. 16:06the races. And if you're someone who's
458. 16:07like, "What is an API and what is prompt
459. 16:09caching?" Well, by definition, you don't
460. 16:11need it. And you can focus on all the
461. 16:13other good habits that I just talked
462. 16:14about to keep your desk clean and make
463. 16:15sure that you're not running into your
464. 16:17AI limits. And this is where the Ringer
465. 16:19multi-agent framework comes in.
466. 16:20Everything so far shares a single
467. 16:23ceiling.
468. 16:24And that ceiling is that a skill cannot
469. 16:27make the call it is inside of any
470. 16:30smaller. Like, if you're in the middle
471. 16:31of a conversation and you're using the
472. 16:33skill,
473. 16:34it's going to do its very best, but it's
474. 16:36going to sort of work on things going
475. 16:38forward. By the time the model reads the
476. 16:40skill, the request initially has already
477. 16:43been sent, right? The conversation, the
478. 16:44standing instructions, the tool
479. 16:46definitions, the hidden setup I talked
480. 16:47about. A lot of that is in the envelope
481. 16:50before the skill ever gets invoked.
482. 16:53And so, what I want to do is think about
483. 16:55that initial request and how we hook
484. 16:57into that and make that cleaner. And
485. 16:59that's the problem I'm solving with
486. 17:01Ringer and level three. And yes, it is
487. 17:03absolutely a bit more of an advanced
488. 17:04solution. Don't don't be scared of it.
489. 17:06You can absolutely do it. I'm just
490. 17:08telling you honestly what it's actually
491. 17:10going to take. Ringer runs locally
492. 17:12between your AI and the model provider.
493. 17:16And before the request goes up to the
494. 17:18model provider,
495. 17:19it can return an answer without a model
496. 17:22call in some instances. It can run a
497. 17:24fixed local recipe with no model call in
498. 17:26some instances. It can select only the
499. 17:28useful passages to send through in some
500. 17:30instances, or it can forward a small
501. 17:32request under hard limits, or it can
502. 17:34even stop it entirely. It is not another
503. 17:36chat window. You don't go somewhere else
504. 17:38to use it. Instead, it is effectively an
505. 17:42intermediary that hooks in, and it helps
506. 17:45to constrain the size of what you're
507. 17:48sending with all of these tricks built
508. 17:50in. Rule number 14, the the
509. 17:52second-to-last one, you want to make
510. 17:55sure that you are able to enforce hard
511. 17:57limits if you're serious about token
512. 17:59usage. So, if you want to say, "I only
513. 18:02want to send packets of a certain size
514. 18:05out, or I only want to get back packets
515. 18:08of a certain size, or I want to make
516. 18:09sure I have a hard limit on my call, so
517. 18:11I'm never sending 10 million tokens."
518. 18:14That's something you can enforce with an
519. 18:16in-between intermediary like Ringer.
520. 18:18It's not something you can really do
521. 18:20without that.
522. 18:21The other thing that's important is that
523. 18:23Ringer allows you to take advantage of
524. 18:25what I talked about with OpenBrain. So,
525. 18:27OpenBrain has, "Oh, if it's got an
526. 18:29answer, we can go get it." Well, Ringer
527. 18:31can go hit OpenBrain and come back, or
528. 18:33your database of choice and come back,
529. 18:35and say, "We've already had an accepted
530. 18:37answer here. We already talked about
531. 18:38this last week. We've got this response.
532. 18:40Is this what you mean?"
533. 18:42And that saves you the call, right? It
534. 18:43saves you the entire 100% of the call.
535. 18:45We have our desk here.
536. 18:47The nine things that we're talking about
537. 18:49initially are you picking up your pen
538. 18:51and your paper and your LEGOs and
539. 18:53keeping the desk clean. And then the
540. 18:54skill is kind of like someone who comes
541. 18:56in and cleans your desk for you every
542. 18:58night. And then Ringer is really a
543. 19:01magical system that keeps your desk
544. 19:04clean for you
545. 19:05before the desk ever gets messy. I'm
546. 19:08always telling you to think big on this
547. 19:09channel, and I don't want token limits
548. 19:11to be the thing that holds you back. And
549. 19:13so this video is all about making sure
550. 19:15that you can literally 10x the value of
551. 19:17your tokens and get where you want to
552. 19:19go. I'm saying 10x for a reason and I
553. 19:20was able to audit down all of the tokens
554. 19:24that I've been using and say where they
555. 19:26actually going, what am I wasting and
556. 19:29how do I make sure that I'm putting my
557. 19:30tokens toward their maximum value so I'm
558. 19:33not wasting the dollars I'm spending on
559. 19:34subscriptions. So better tools are
560. 19:36coming and they will get better but
561. 19:39you're still going to have to keep your
562. 19:40desk clean.
563. 19:41And everything is linked below. The 15
564. 19:43rules are written up in full. I have the
565. 19:45skill for you over on Substack and
566. 19:48Ringer is there and if you want a whole
567. 19:50introduction to Ringer, I have a whole
568. 19:51video on that. And if you tuned out and
569. 19:53you're the kind of person that wants
570. 19:54Ringer, you can get that started. If
571. 19:56you've tried any of the tools I've
572. 19:57mentioned, if you tried the skill, if
573. 19:59you've tried Ringer, let me know below.
574. 20:01If you're just trying it for the first
575. 20:02time, let me know below, too. Let me
576. 20:04know what your experience is and if you
577. 20:06know someone who's running out of AI,
578. 20:08share this video with them. We don't
579. 20:10want them to run out of AI.