WEBVTT 00:00:00.000 --> 00:00:03.040 Imagine sitting in a massive high -stakes board 00:00:03.040 --> 00:00:06.259 meeting. You pull up a gorgeous spreadsheet generated 00:00:06.259 --> 00:00:09.160 by your AI tool. The numbers look absolutely 00:00:09.160 --> 00:00:11.179 perfect. You present them confidently to your 00:00:11.179 --> 00:00:13.439 boss. But right in the middle of your pitch, 00:00:13.859 --> 00:00:18.620 you realize the terrifying truth. The AI just 00:00:18.620 --> 00:00:21.359 quietly invented half the revenue data out of 00:00:21.359 --> 00:00:27.280 thin air. Welcome to the deep dive. We are exploring 00:00:27.280 --> 00:00:29.879 an incredible source guide today. It tackles 00:00:29.879 --> 00:00:33.140 the honesty gap in modern AI. It is honestly 00:00:33.140 --> 00:00:35.359 a massive vulnerability for anyone using these 00:00:35.359 --> 00:00:37.740 tools right now. Our mission is fixing this exact 00:00:37.740 --> 00:00:41.100 problem. We are going to unpack why AI invents 00:00:41.100 --> 00:00:43.159 these answers, even when the truth is staring 00:00:43.159 --> 00:00:44.780 it right in the face. Yeah, we really need to 00:00:44.780 --> 00:00:47.020 understand that. We will reveal a strict three 00:00:47.020 --> 00:00:49.579 -step prompt method. It is designed to catch 00:00:49.579 --> 00:00:51.859 these hidden mistakes perfectly. and we will 00:00:51.859 --> 00:00:54.079 show how to totally transform your final data 00:00:54.079 --> 00:00:56.000 review. It completely flips the script on how 00:00:56.000 --> 00:00:57.579 you audit information. You are never going to 00:00:57.579 --> 00:01:00.039 look at AI outputs the same way again. But before 00:01:00.039 --> 00:01:02.119 we can actually fix the machine, we have to understand 00:01:02.119 --> 00:01:04.500 why it lies to us, and more importantly, why 00:01:04.500 --> 00:01:06.840 we keep falling for it. The sources refer to 00:01:06.840 --> 00:01:09.719 this as the honesty gap. It is a really fascinating 00:01:09.719 --> 00:01:12.420 concept. AI tools are getting incredibly smart 00:01:12.420 --> 00:01:15.500 right now. They really are. They read massive 00:01:15.500 --> 00:01:19.079 hundred page documents in seconds. They summarize 00:01:19.079 --> 00:01:22.760 chaotic meetings effortlessly, but their honesty 00:01:22.760 --> 00:01:25.640 just is not. keeping pace with their capability. 00:01:25.859 --> 00:01:28.379 They seem fundamentally incapable of admitting 00:01:28.379 --> 00:01:30.659 their own ignorance. Right. And that is the core 00:01:30.659 --> 00:01:33.420 issue here. They cannot easily say, I don't know, 00:01:33.840 --> 00:01:36.700 the entire system is built around this deep desire 00:01:36.700 --> 00:01:38.980 to be helpful. The architecture is literally 00:01:38.980 --> 00:01:41.959 designed to please the user. Exactly. So if a 00:01:41.959 --> 00:01:44.180 specific fact is completely missing from the 00:01:44.180 --> 00:01:47.450 text, the machine basically panics inside. I 00:01:47.450 --> 00:01:50.530 picture the AI like a terrified, eager -to -please 00:01:50.530 --> 00:01:52.590 intern. Oh, that's a great way to put it. It 00:01:52.590 --> 00:01:54.530 is their first day on the job. You ask them for 00:01:54.530 --> 00:01:56.310 a critical file. They cannot find it anywhere. 00:01:56.409 --> 00:01:59.189 Right. But they are so terrified of looking incompetent, 00:01:59.290 --> 00:02:01.689 they just draft a fake one. They hand it to you, 00:02:01.730 --> 00:02:04.069 desperately hoping you won't notice. It is a 00:02:04.069 --> 00:02:06.269 people -pleasing mechanism gone totally rogue. 00:02:06.849 --> 00:02:09.409 The AI feels like it failed its primary mission, 00:02:09.629 --> 00:02:12.449 so it just uses its logic pathways to guess the 00:02:12.449 --> 00:02:14.490 answer. It fills the void to keep you happy. 00:02:14.569 --> 00:02:17.389 Yeah. But these guesses do not come with any 00:02:17.389 --> 00:02:19.930 bright red warning labels. The sources call these 00:02:19.930 --> 00:02:22.889 things silent mistakes. They look exactly like 00:02:22.889 --> 00:02:26.270 real verifiable facts. They really do. Think 00:02:26.270 --> 00:02:29.189 about asking a tool to review a complex legal 00:02:29.189 --> 00:02:32.610 contract. It stands the document and sees conflicting 00:02:32.610 --> 00:02:36.090 payment clauses. Page two says 30 days, page 00:02:36.090 --> 00:02:39.289 six says 45 days. Instead of flagging that massive 00:02:39.289 --> 00:02:42.050 discrepancy, it just quietly picks one. The conflict 00:02:42.050 --> 00:02:43.590 is completely hidden from you. You would never 00:02:43.590 --> 00:02:46.189 even know there was an issue or think about messy 00:02:46.189 --> 00:02:48.689 disorganized meeting notes someone mutters they 00:02:48.689 --> 00:02:50.550 will try to check on a project next week right 00:02:50.550 --> 00:02:53.710 a vague comment the AI grabs that non -committal 00:02:53.710 --> 00:02:56.090 promise it turns it into a hard concrete deadline 00:02:56.090 --> 00:02:58.629 it suddenly assigns a specific date in person 00:02:58.629 --> 00:03:00.289 but nobody actually agreed to that in the meeting 00:03:00.289 --> 00:03:03.930 at all exactly but the sources highlight a second 00:03:03.930 --> 00:03:06.270 equally dangerous problem here and fortunately 00:03:06.270 --> 00:03:09.639 that problem is us Humans are a massive part 00:03:09.639 --> 00:03:12.180 of this entire failure loop. We suffer deeply 00:03:12.180 --> 00:03:14.400 from something called automation bias. Meaning 00:03:14.400 --> 00:03:17.460 we blindly trust machines just because their 00:03:17.460 --> 00:03:20.419 output looks smart. Yeah, we really do. It is 00:03:20.419 --> 00:03:23.620 the illusion of clean data. A table full of completely 00:03:23.620 --> 00:03:26.039 wrong numbers looks incredibly professional. 00:03:26.139 --> 00:03:28.620 It is formatted perfectly. It has bold headers 00:03:28.620 --> 00:03:31.520 and crisp alignment. It looks exactly like a 00:03:31.520 --> 00:03:33.340 table full of perfectly right numbers. I have 00:03:33.340 --> 00:03:35.990 to make a vulnerable admission here. I still 00:03:35.990 --> 00:03:38.330 wrestle with trusting a clean spreadsheet just 00:03:38.330 --> 00:03:40.110 because the formatting looks professional. We 00:03:40.110 --> 00:03:42.990 all fall for it constantly. My brain just assumes 00:03:42.990 --> 00:03:45.969 the data must be accurate. It is a very dangerous 00:03:45.969 --> 00:03:49.110 illusion. The perfect presentation actively tricks 00:03:49.110 --> 00:03:51.590 our brains. It lowers our cognitive defenses. 00:03:52.030 --> 00:03:54.960 We feel safe. So we check the data much less 00:03:54.960 --> 00:03:57.879 carefully. Then those tiny silent errors just 00:03:57.879 --> 00:04:00.300 stay embedded in the work. And they slowly grow 00:04:00.300 --> 00:04:02.879 into massive business problems over time. We 00:04:02.879 --> 00:04:04.860 let our guard down completely because the packaging 00:04:04.860 --> 00:04:07.520 is pretty. And that is the danger of the honesty 00:04:07.520 --> 00:04:10.289 gap. The raw intelligence of the tool keeps going 00:04:10.289 --> 00:04:13.530 up, but the transparency stays exactly the same. 00:04:13.870 --> 00:04:15.750 Let's pause and really drill down into the mechanics 00:04:15.750 --> 00:04:19.730 of this. Why exactly does the AI feel so much 00:04:19.730 --> 00:04:22.290 pressure to fill in those blanks? Because its 00:04:22.290 --> 00:04:24.990 underlying programming explicitly rewards generating 00:04:24.990 --> 00:04:28.250 an output over ensuring strict accuracy. Ah, 00:04:28.430 --> 00:04:31.170 so it literally views a blank space as a failure. 00:04:31.439 --> 00:04:34.680 Beat. Exactly. It is fighting its own core directive. 00:04:34.879 --> 00:04:37.379 Since its default programming views empty spaces 00:04:37.379 --> 00:04:40.699 as failures, we have to actively rewire its incentive 00:04:40.699 --> 00:04:43.199 structure. We need to build a completely new 00:04:43.199 --> 00:04:46.279 set of rules. The source outlines a very strict 00:04:46.279 --> 00:04:48.779 three -step method for this. This is where you 00:04:48.779 --> 00:04:50.860 finally take control back from the machine. Step 00:04:50.860 --> 00:04:53.819 one is surprisingly simple, but incredibly powerful. 00:04:54.420 --> 00:04:57.019 You must explicitly tell the machine to leave 00:04:57.019 --> 00:05:00.379 things blank. An empty space is actually honest. 00:05:00.579 --> 00:05:02.680 It is the most honest thing the AI can do. It 00:05:02.680 --> 00:05:04.939 tells you exactly what data is missing. It highlights 00:05:04.939 --> 00:05:07.259 exactly what needs your human attention. But 00:05:07.259 --> 00:05:09.800 you also need to enforce a strict grounding rule 00:05:09.800 --> 00:05:12.560 here. Let's define that. A grounding rule just 00:05:12.560 --> 00:05:15.279 limits the AI strictly to the provided text. 00:05:15.680 --> 00:05:18.879 Yeah, you literally tell it, stay with the text 00:05:18.879 --> 00:05:21.660 only. You are building a fence around its knowledge 00:05:21.660 --> 00:05:25.100 base. A common fatal mistake is forgetting to 00:05:25.100 --> 00:05:27.529 include that grounding line. If you forget it, 00:05:27.649 --> 00:05:30.529 the AI just brings in outside knowledge. It sneaks 00:05:30.529 --> 00:05:32.709 out to the internet to guess the answer. It breaks 00:05:32.709 --> 00:05:35.649 the fence and wanders off. Exactly. And there 00:05:35.649 --> 00:05:38.310 is another crucial part to step one. You must 00:05:38.310 --> 00:05:42.269 always ask the AI why it left a blank. Otherwise, 00:05:42.350 --> 00:05:45.019 you are just staring at an empty box. You waste 00:05:45.019 --> 00:05:47.079 hours hunting through the document for the reason 00:05:47.079 --> 00:05:49.560 yourself. Right. A short explanatory note saves 00:05:49.560 --> 00:05:51.600 you massive amounts of reading. It might say, 00:05:52.019 --> 00:05:54.459 the text mentions revenue but omits the exact 00:05:54.459 --> 00:05:56.759 year. Now you know exactly what the problem is. 00:05:56.980 --> 00:06:00.060 That brings us to step two. This step... actively 00:06:00.060 --> 00:06:02.319 changes the machine's reward system. This is 00:06:02.319 --> 00:06:05.939 huge. You must explicitly give the AI a new mathematical 00:06:05.939 --> 00:06:09.180 rule. You tell it, a wrong answer is three times 00:06:09.180 --> 00:06:11.279 worse than a blank. This is like pulling that 00:06:11.279 --> 00:06:13.699 terrified intern into your office. You give them 00:06:13.699 --> 00:06:16.259 a totally new job description. You clearly explain 00:06:16.259 --> 00:06:18.519 that guessing wrong now costs them their annual 00:06:18.519 --> 00:06:21.560 bonus. And it changes their entire behavior instantly. 00:06:22.379 --> 00:06:24.680 They stop guessing immediately because the risk 00:06:24.680 --> 00:06:27.170 is just too high. Leaving it blank is totally 00:06:27.170 --> 00:06:31.029 fine, but a wrong guess is fatal. But there is 00:06:31.029 --> 00:06:34.110 a huge hidden trap here that people fall into. 00:06:34.750 --> 00:06:37.430 The source is worn against this constantly. You 00:06:37.430 --> 00:06:40.290 should never ask the AI for a confidence score. 00:06:40.610 --> 00:06:43.129 Like asking if it is 9 out of 10 sure about a 00:06:43.129 --> 00:06:45.230 fact. Yeah, people love asking for those scores, 00:06:45.290 --> 00:06:47.790 but that score just gives the machine an excuse 00:06:47.790 --> 00:06:50.689 to lie confidently. It essentially lets the tool 00:06:50.689 --> 00:06:53.759 decide its own level of trustworthiness. An AI 00:06:53.759 --> 00:06:56.540 will happily invent a fake fact and then proudly 00:06:56.540 --> 00:06:59.079 give it a 10 out of 10 confidence rating. It 00:06:59.079 --> 00:07:01.860 is grading its own homework. Exactly. You should 00:07:01.860 --> 00:07:04.800 always use blank spaces and written reasons instead. 00:07:05.220 --> 00:07:07.939 Force it to explain itself. Do not just ask for 00:07:07.939 --> 00:07:10.939 a shiny number. That perfectly sets up step three. 00:07:11.180 --> 00:07:14.139 You have to force the AI to show its exact source. 00:07:14.279 --> 00:07:17.000 This is the auditing safety net. You force the 00:07:17.000 --> 00:07:19.579 AI to label every single answer it gives you. 00:07:19.639 --> 00:07:22.259 It must use the word extracted or the word inferred. 00:07:22.410 --> 00:07:24.610 Let's define those carefully. Extracted means 00:07:24.610 --> 00:07:26.970 the fact is a direct quote from the provided 00:07:26.970 --> 00:07:30.209 document. Yes, it is pulled right from the text. 00:07:30.449 --> 00:07:33.610 And inferred means the AI used logic to guess 00:07:33.610 --> 00:07:36.850 from the surrounding context. For every single 00:07:36.850 --> 00:07:39.790 inferred answer, you most demand a one -sentence 00:07:39.790 --> 00:07:43.110 explanation. You need to see its underlying logic 00:07:43.110 --> 00:07:46.220 clearly laid out. The sources say a common mistake 00:07:46.220 --> 00:07:49.279 is not separating these answers clearly. If you 00:07:49.279 --> 00:07:51.839 do not separate them, you completely destroy 00:07:51.839 --> 00:07:54.300 your safety net. You will end up mixing hard 00:07:54.300 --> 00:07:57.079 facts with algorithmic guesses. You won't know 00:07:57.079 --> 00:07:59.420 which data points are actually real. You will 00:07:59.420 --> 00:08:01.839 end up treating a fragile guess like a rock -solid 00:08:01.839 --> 00:08:04.259 fact. That defeats the whole purpose of using 00:08:04.259 --> 00:08:06.740 the system. You are right back to trusting the 00:08:06.740 --> 00:08:09.540 shiny spreadsheet. You really are. You have to 00:08:09.540 --> 00:08:12.180 force the machine to show its work. When it literally 00:08:12.180 --> 00:08:14.529 writes the word inferred, It is admitting it 00:08:14.529 --> 00:08:17.670 made a leap. You, the human, can then judge if 00:08:17.670 --> 00:08:19.850 that specific leap actually makes sense. Let 00:08:19.850 --> 00:08:22.629 me ask about that penalty concept again. Does 00:08:22.629 --> 00:08:24.769 telling it three times worse actually change 00:08:24.769 --> 00:08:27.750 the math inside the model's head? Yes. Explicitly 00:08:27.750 --> 00:08:30.389 weighting the penalties actively shifts the AI's 00:08:30.389 --> 00:08:33.169 probability folders. We basically just hack its 00:08:33.169 --> 00:08:35.710 incentives by making guessing too expensive. 00:08:36.250 --> 00:08:39.269 Beat. Exactly. We adjust the internal math, deciding 00:08:39.269 --> 00:08:42.149 which word comes next. Mid -roll sponsor, read 00:08:42.149 --> 00:08:45.129 placeholder, So we have hacked the incentives. 00:08:45.389 --> 00:08:47.690 We have completely rewired the rules of engagement. 00:08:47.850 --> 00:08:50.750 We have. But how does this actually change our 00:08:50.750 --> 00:08:53.289 Tuesday morning workflow? We need to look at 00:08:53.289 --> 00:08:55.850 how this fundamentally transforms our data auditing. 00:08:56.289 --> 00:08:59.049 The sources provide some brilliant real -world 00:08:59.049 --> 00:09:02.009 applications for this exact framework. It moves 00:09:02.009 --> 00:09:04.629 it from theory into daily practice. Let's fool 00:09:04.629 --> 00:09:08.220 them. Imagine you are analyzing Facebook ad reports 00:09:08.220 --> 00:09:10.899 for your marketing business. The raw data exports 00:09:10.899 --> 00:09:13.980 are often incredibly messy or missing entirely. 00:09:14.299 --> 00:09:16.580 Without these strict rules, the AI just guesses 00:09:16.580 --> 00:09:19.019 the cost per lead. It wants to give you a complete 00:09:19.019 --> 00:09:21.639 pretty table. Right, it uses a weak generalized 00:09:21.639 --> 00:09:24.559 formula to invent a number. It fills the cell 00:09:24.559 --> 00:09:26.779 so you won't be mad. But with these new rules 00:09:26.779 --> 00:09:29.440 applied, it just leaves that cell blank. It tells 00:09:29.440 --> 00:09:32.100 you exactly which specific day has a data error. 00:09:32.379 --> 00:09:34.899 It prevents you from making terrible budget choices 00:09:34.899 --> 00:09:37.419 based on a hallucination. Let's shift context 00:09:37.419 --> 00:09:39.860 and look at cryptocurrency news next. Oh, the 00:09:39.860 --> 00:09:42.019 stakes are incredibly high there. You are reading 00:09:42.019 --> 00:09:47.200 a dense 40 page white paper about a brand new 00:09:47.200 --> 00:09:50.279 coin. It is full of complex jargon and chaotic 00:09:50.279 --> 00:09:53.159 formatting. You really just want the exact total 00:09:53.159 --> 00:09:56.539 supply and the launch date. A normal lazy prompt 00:09:56.539 --> 00:09:59.000 makes the machine guess the launch date. If it 00:09:59.000 --> 00:10:01.379 is not in the paper, it pulls random rumors from 00:10:01.379 --> 00:10:03.980 the Internet. It breaks the fence again. Exactly. 00:10:04.539 --> 00:10:07.759 But our strict system extracts the exact total 00:10:07.759 --> 00:10:11.340 supply cleanly. It labels it extracted. And it 00:10:11.340 --> 00:10:13.960 leaves the watch date completely blank if it 00:10:13.960 --> 00:10:16.879 is genuinely unclear. You always get safe verified 00:10:16.879 --> 00:10:19.759 information that way. What about a legal context? 00:10:20.100 --> 00:10:22.259 Let's say we are reviewing sponsor contracts. 00:10:22.440 --> 00:10:24.600 This is honestly a perfect daily use case for 00:10:24.600 --> 00:10:27.620 this method. You are scanning a messy PDF. Page 00:10:27.620 --> 00:10:30.240 two clearly says payment is due in 30 days. But 00:10:30.240 --> 00:10:32.419 buried down on page six, another clause says 00:10:32.419 --> 00:10:35.440 45 days. Under the old system, it just picks 00:10:35.440 --> 00:10:37.919 one and hides the conflict. Under our new system, 00:10:38.080 --> 00:10:40.539 the AI spots a conflict immediately. It leaves 00:10:40.539 --> 00:10:43.419 the final payment term blank. Yes. And in the 00:10:43.419 --> 00:10:46.279 reason column, it points out the exact Error. 00:10:46.500 --> 00:10:49.240 It says, conflict between page two and page six. 00:10:49.820 --> 00:10:52.080 Now you can actually call your partner and clarify 00:10:52.080 --> 00:10:54.779 before signing anything. You avoid a massive, 00:10:55.080 --> 00:10:57.139 expensive headache down the road. The source 00:10:57.139 --> 00:10:59.659 guide provides a final master prompt to tie this 00:10:59.659 --> 00:11:03.059 together. It is a simple copy paste, seven rule 00:11:03.059 --> 00:11:06.139 numbered list. It forcefully locks the AI into 00:11:06.139 --> 00:11:09.000 this strict structural format. It outputs a very 00:11:09.000 --> 00:11:11.580 specific table. It has a reason column and a 00:11:11.580 --> 00:11:14.980 source column. It also demands exact page numbers 00:11:14.980 --> 00:11:17.559 for everything it finds. This essentially creates 00:11:17.559 --> 00:11:20.600 what they call the three -step fast review process. 00:11:20.899 --> 00:11:22.899 It completely changes how you spend your time. 00:11:23.210 --> 00:11:26.330 You no longer read every single line of the AI's 00:11:26.330 --> 00:11:28.970 output. First, you focus entirely on the blanks. 00:11:29.169 --> 00:11:30.929 You read the short reason it left them empty. 00:11:31.149 --> 00:11:33.110 You are just checking its homework on the missing 00:11:33.110 --> 00:11:35.710 pieces. Second, you check the inferred boxes. 00:11:36.149 --> 00:11:38.330 You read that one sentence to see if the machine's 00:11:38.330 --> 00:11:40.129 logic actually holds up. Right. You make sure 00:11:40.129 --> 00:11:43.049 its logical leap is not totally crazy. Finally, 00:11:43.169 --> 00:11:45.470 you look quickly at the extracted boxes. You 00:11:45.470 --> 00:11:47.629 can basically trust those cited page numbers 00:11:47.629 --> 00:11:52.889 completely. Whoa. Beat. Imagine reducing a massive 00:11:52.889 --> 00:11:55.529 stressful contract review into just scanning 00:11:55.529 --> 00:11:58.450 three blank boxes. It completely eliminates the 00:11:58.450 --> 00:12:00.909 exhausting stress of hunting for constant errors. 00:12:01.610 --> 00:12:04.350 The table acts like a map. It tells you exactly 00:12:04.350 --> 00:12:06.730 where the remaining risks are located. You do 00:12:06.730 --> 00:12:09.269 not have to guess or panic anymore. It fundamentally 00:12:09.269 --> 00:12:11.169 changes how you interact with the information. 00:12:11.669 --> 00:12:13.850 You are suddenly auditing the process itself, 00:12:14.029 --> 00:12:16.330 not just passively reading words. You graduate. 00:12:16.509 --> 00:12:19.370 It makes you a true manager of the AI tool. You 00:12:19.370 --> 00:12:21.950 stop being a passive, vulnerable consumer of 00:12:21.950 --> 00:12:23.950 its automated guesses. You become an editor. 00:12:24.490 --> 00:12:27.129 But let me push back on that last step. Can we 00:12:27.129 --> 00:12:29.230 entirely skip reading those extracted parts of 00:12:29.230 --> 00:12:31.289 the document now? You still do a quick glance, 00:12:31.429 --> 00:12:33.690 but the exact page citations mean the heavy lifting 00:12:33.690 --> 00:12:36.210 is completely done. So we stop hunting for needles 00:12:36.210 --> 00:12:39.789 and just audit the machine's doubts. Beat. Exactly. 00:12:40.210 --> 00:12:42.289 Let's take a step back and recap the big idea 00:12:42.289 --> 00:12:46.019 here today. Artificial intelligence is not inherently 00:12:46.019 --> 00:12:50.000 bad, and it is not intentionally deceptive. It 00:12:50.000 --> 00:12:52.320 does not have a malicious agenda. It is just 00:12:52.320 --> 00:12:56.000 an incredibly powerful, eager tool. But it desperately 00:12:56.000 --> 00:12:58.600 needs strict mathematical boundaries to work 00:12:58.600 --> 00:13:01.539 safely. We simply have to prioritize honesty 00:13:01.539 --> 00:13:05.019 over helpfulness every single time. When we actively 00:13:05.019 --> 00:13:07.659 do that, we take control back from the machine. 00:13:07.840 --> 00:13:10.570 We stop letting it guess just to please us. We 00:13:10.570 --> 00:13:13.070 demand clarity. I highly recommend you try this 00:13:13.070 --> 00:13:16.490 out today on one single document. Take a messy 00:13:16.490 --> 00:13:19.350 weekly report or a short transcript. Apply these 00:13:19.350 --> 00:13:22.129 three rules and just see the difference for yourself. 00:13:22.190 --> 00:13:24.470 You are going to catch structural issues so much 00:13:24.470 --> 00:13:26.690 faster than before. You will feel incredibly 00:13:26.690 --> 00:13:28.750 more in control of your daily business data. 00:13:28.909 --> 00:13:32.090 It makes you wonder. Two secs silence. If we 00:13:32.090 --> 00:13:35.429 can train an AI to value a blank space and an 00:13:35.429 --> 00:13:37.519 honest I don't know, over a confident guess. 00:13:37.960 --> 00:13:39.779 Maybe we should start demanding the same standard 00:13:39.779 --> 00:13:41.159 from ourselves in our own meetings.