WEBVTT - Still arguing with AI over a bad first draft? Try these 4 tips 0:00:00.910 --> 0:00:05.110 Do you still have one of those prompt formulas saved somewhere? 0:00:05.530 --> 0:00:09.890 Maybe some acronym, an Instagram reel, swore would get you 0:00:10.050 --> 0:00:14.370 better answers from AI. And maybe you still paste it in, 0:00:14.490 --> 0:00:17.890 swap a few words, hit send, and what comes back 0:00:18.160 --> 0:00:22.340 still isn't what you wanted. Part of the trouble is 0:00:22.380 --> 0:00:25.460 that most of us still talk to AI the way 0:00:25.680 --> 0:00:31.520 Google trained us to. A few keywords shouted into a box. Today, 0:00:31.840 --> 0:00:35.120 Neo and I look at why those formulas caught on 0:00:35.300 --> 0:00:39.880 and what has changed since, including how memory now fills 0:00:40.100 --> 0:00:43.659 in some of the gaps for you. And today, you're 0:00:43.700 --> 0:00:47.519 going to learn how to brief AI properly, what to 0:00:47.680 --> 0:00:50.280 ask it so it shows you better options, and how 0:00:50.320 --> 0:00:53.240 to get it to challenge you right down to whether 0:00:53.520 --> 0:00:57.180 you should be making the thing at all. By the 0:00:57.260 --> 0:00:59.960 end of today's show, you're going to have a simpler 0:01:00.020 --> 0:01:03.380 way of talking to AI that gets you better work 0:01:03.480 --> 0:01:06.220 back with far less back and forth. 0:01:12.190 --> 0:01:15.690 Welcome to How I AI with me, Dr. Amantha Imber 0:01:15.830 --> 0:01:18.630 and Neo Applin, head of Inventium AI. 0:01:19.560 --> 0:01:22.679 Each episode, we share one practical way to use AI 0:01:22.740 --> 0:01:24.520 better at work and in life. 0:01:25.080 --> 0:01:28.100 No fluff, no tech jargon, just things you can use 0:01:28.160 --> 0:01:28.840 straight away. 0:01:29.800 --> 0:01:34.390 Prompt formulas used to be really, really big in AI 0:01:34.560 --> 0:01:41.450 in like 2023, 2024, maybe marginally in 2025, where, you know, 0:01:41.470 --> 0:01:44.950 there were all these fancy acronyms that we had to 0:01:45.010 --> 0:01:50.390 remember that the Instagram influencers were telling us in reels 0:01:50.730 --> 0:01:53.440 to apparently get better output from AI. 0:01:53.900 --> 0:01:57.660 So, I mean, why did we have, like, why was 0:01:57.720 --> 0:02:00.840 there such a focus on prompt formulas, Neo? 0:02:01.720 --> 0:02:04.120 They were cool. Everyone had one and I wanted to 0:02:04.200 --> 0:02:05.920 collect them. No, I didn't want to collect them all. 0:02:05.960 --> 0:02:08.179 I found it very difficult to remember. Did you? It 0:02:08.860 --> 0:02:11.900 was a prep, edit or goal. I can't remember all 0:02:11.940 --> 0:02:15.740 those different things that they had for these acronyms. Yeah, 0:02:16.139 --> 0:02:20.070 the reason was first off, these were pretty early models. 0:02:20.470 --> 0:02:22.390 They were pretty good. Like if you look back to 0:02:22.490 --> 0:02:24.250 AI a couple of years back, they were pretty good, 0:02:24.270 --> 0:02:26.660 but they were not as good as what we have today. 0:02:27.240 --> 0:02:29.680 And so people were trying to find a way to 0:02:29.720 --> 0:02:31.540 get the model to actually work with them in a 0:02:31.600 --> 0:02:33.680 way which was a little bit more reliable and actually 0:02:33.700 --> 0:02:37.760 would actually be better. And so each of the companies 0:02:37.800 --> 0:02:40.480 and different people had different ways to be able to 0:02:40.720 --> 0:02:44.380 tell AI what we wanted. But it was a flawed 0:02:44.540 --> 0:02:48.750 perspective in a couple of ways. One was what we 0:02:48.770 --> 0:02:52.850 were trying to do was send it one command, a 0:02:53.070 --> 0:02:56.790 magical command, this big long prompt that had all the 0:02:56.870 --> 0:02:59.280 words in it and all the things. And then it 0:02:59.310 --> 0:03:02.200 would be like an abracadabra. It would then open up 0:03:02.260 --> 0:03:04.580 the thing and then suddenly you'd get a great response 0:03:04.639 --> 0:03:09.299 from AI. And maybe that was the case for the day. 0:03:09.820 --> 0:03:12.690 And I don't think it was really... true, but I 0:03:12.710 --> 0:03:14.169 think a lot of people felt a lot of comfort 0:03:14.230 --> 0:03:16.450 in that I've got this magical thing called AI. I'm 0:03:16.690 --> 0:03:18.070 not really quite sure how to talk to it, but 0:03:18.130 --> 0:03:20.170 here I've got a formula. I'll just grab the formula, 0:03:20.350 --> 0:03:22.430 change a couple of things and then send it and 0:03:22.450 --> 0:03:25.210 then it will be great. And that's where the kind 0:03:25.250 --> 0:03:27.630 of reliance came from. And it does make sense. We're 0:03:27.650 --> 0:03:29.470 all getting used to it and people wanted to find 0:03:29.500 --> 0:03:30.190 their way through it. 0:03:31.080 --> 0:03:35.840 So what has changed so much that really these prompt 0:03:35.900 --> 0:03:39.290 formulas are really don't matter in the way that they 0:03:39.470 --> 0:03:42.590 did previously. So what are the big changes that have happened? 0:03:43.490 --> 0:03:46.370 The first is that these models have gotten better. So 0:03:46.430 --> 0:03:51.000 the AI companies, Anthropic and OpenAI and all the others, 0:03:51.280 --> 0:03:53.590 they've done a better job with these models. And the 0:03:53.630 --> 0:03:57.180 models are better at understanding what you really mean, not 0:03:57.300 --> 0:04:00.540 what you say. The problems that we as humans have 0:04:00.600 --> 0:04:04.140 are we're really lazy with our language. And so AI 0:04:04.220 --> 0:04:07.590 can But earlier, it was very literal. And so it 0:04:07.610 --> 0:04:09.210 would go, oh, you want this? And then you go, no, 0:04:09.230 --> 0:04:12.170 it's not really what I meant. It's what I said, sure, 0:04:12.210 --> 0:04:14.550 but it's not what I meant. Now, AI has a 0:04:14.590 --> 0:04:18.830 much better understanding of what good looks like. Sometimes it 0:04:18.890 --> 0:04:20.910 has a better idea of what good looks like than 0:04:20.950 --> 0:04:23.810 what we do. And so then it can kind of 0:04:23.870 --> 0:04:26.570 read between the lines. And there is really the big 0:04:26.630 --> 0:04:29.409 benefit of it. So rather than us having to teach 0:04:29.630 --> 0:04:33.330 AI what good is, because they've elevated their models, they've 0:04:33.350 --> 0:04:35.940 something a lot smarter than they used to be, AI, 0:04:36.120 --> 0:04:38.279 and here's actually probably a problem, is AI thinks it 0:04:38.300 --> 0:04:40.619 probably knows better than you do in a lot of ways. 0:04:40.700 --> 0:04:44.440 And so we don't need to use magical phrases or 0:04:44.740 --> 0:04:47.520 cut and paste major prompt libraries to be able to 0:04:47.560 --> 0:04:49.720 get a pretty good output out of it. 0:04:50.680 --> 0:04:56.290 So Also, memory, I believe, is something that has changed 0:04:56.350 --> 0:04:59.420 as well that impacts on how we have a conversation 0:04:59.560 --> 0:05:02.000 with the AI in a way that gets us really 0:05:02.020 --> 0:05:02.760 good output. 0:05:03.300 --> 0:05:05.740 Yeah, I should have mentioned that. That's the other part. 0:05:05.800 --> 0:05:08.440 So the memories in these models, AI does know a 0:05:08.480 --> 0:05:10.160 fair bit about you. It depends on the different tools 0:05:10.180 --> 0:05:13.430 you've got. There's different types of the memories there. But 0:05:13.490 --> 0:05:18.310 in a broad kind of framework, Think about AI having 0:05:18.350 --> 0:05:21.100 a little notebook about you. And it's about three pages-ish 0:05:21.750 --> 0:05:25.220 of notebook about you. And it depends on the different 0:05:25.320 --> 0:05:27.260 models as to how they do this. But generally, it's 0:05:27.279 --> 0:05:29.920 like who you are, what you do, what you're generally 0:05:30.060 --> 0:05:34.220 working on. And it does really like to know about 0:05:34.240 --> 0:05:36.190 the things that you care about. So if you think 0:05:36.230 --> 0:05:40.950 like your children, your dogs, your pets, your hobbies and passions, 0:05:41.029 --> 0:05:44.440 it certainly likes to scribble those away. also how you 0:05:44.500 --> 0:05:46.900 try to talk to it. And then it starts to 0:05:46.940 --> 0:05:49.380 get an idea of the human that it's dealing with 0:05:49.640 --> 0:05:52.859 so it can more likely personalize things for you. And 0:05:52.920 --> 0:05:55.520 so it's then kind of guessing based on your role, 0:05:55.580 --> 0:05:58.099 your job, what you've done with it before, I got 0:05:58.140 --> 0:06:00.359 an idea of what you're trying to achieve here. So 0:06:00.400 --> 0:06:02.480 it'll help you do it again. So you don't have 0:06:02.520 --> 0:06:05.750 to be as explicit. But then again, I'd also argue 0:06:06.050 --> 0:06:08.359 that can be a bit dangerous because Just because I 0:06:08.400 --> 0:06:11.700 did something once or I've worked in a particular area 0:06:11.779 --> 0:06:13.930 doesn't mean that I'm going to do that kind of 0:06:14.010 --> 0:06:16.950 thing again in the future. And so I find the 0:06:16.990 --> 0:06:19.230 memory is a bit of a two-edged sword. For most people, 0:06:19.270 --> 0:06:22.180 it's magnificent and you don't have to prompt as hard 0:06:22.330 --> 0:06:25.020 with those magical prompts because it already kind of understands 0:06:25.040 --> 0:06:26.840 what you're after. But if you're doing lots of new 0:06:26.900 --> 0:06:29.740 or different things, then how you worked yesterday doesn't help 0:06:29.779 --> 0:06:30.260 you today. 0:06:31.360 --> 0:06:34.620 So given all these, given all these changes, how should 0:06:34.680 --> 0:06:37.120 we be prompting AI now? 0:06:37.160 --> 0:06:38.969 What's the best way to have a conversation with it? 0:06:38.980 --> 0:06:39.800 All right. 0:06:39.820 --> 0:06:43.230 I'm going to give you four tips. Tip number one, 0:06:43.940 --> 0:06:47.050 we've got to forget what we've learned from Google. Google's great. 0:06:47.130 --> 0:06:49.930 Don't get me wrong. Love Google. Google's awesome. But what 0:06:49.970 --> 0:06:52.930 we've learned from Google over all of these years is 0:06:53.410 --> 0:06:55.790 I will shoot it a, I call it a caveman 0:06:55.850 --> 0:07:00.779 phrase of keywords. So it might be a best washing 0:07:00.820 --> 0:07:03.560 machine 2026 or something like that. And then you'll get 0:07:03.600 --> 0:07:05.360 a pretty crap response. And so you might go best 0:07:05.420 --> 0:07:10.380 washing machine, family, grass stains, 2026 or something like that. 0:07:10.640 --> 0:07:13.390 And so we did like yelling at our AI at 0:07:13.470 --> 0:07:15.990 Google and we've got like four or five words and 0:07:16.010 --> 0:07:17.670 then we'll get this response back. Now I do all 0:07:17.690 --> 0:07:19.790 the work here because I'm hitting another word. I've got 0:07:19.810 --> 0:07:22.010 to read another eight pages and all those kinds of things. 0:07:22.450 --> 0:07:25.880 And what I found watching people is they will effectively 0:07:26.390 --> 0:07:28.080 put a couple of words around their Google phrase and 0:07:28.100 --> 0:07:31.000 give that to AI. So it's the equivalent of what 0:07:31.100 --> 0:07:33.380 is the best washing machine I should get for my 0:07:33.440 --> 0:07:36.860 family in 2026? It's still pretty the same thing, you know, 0:07:37.180 --> 0:07:39.860 and people do shout their commands at AI, which is 0:07:40.100 --> 0:07:42.700 write me a report that does blah. And of course, 0:07:42.730 --> 0:07:44.020 then you get a report and then you go, that's 0:07:44.060 --> 0:07:46.020 not a good report. That's a terrible report. And then 0:07:46.050 --> 0:07:48.190 you're arguing with your AI and you're spending a lot 0:07:48.210 --> 0:07:52.310 of time and energy of your emotional energy and also 0:07:52.370 --> 0:07:54.970 tokens as well to be able to get it. So instead, 0:07:55.530 --> 0:07:58.210 tip number two, tip number one is forget what we've 0:07:58.230 --> 0:08:01.840 learned from Google. Tip number two is treat it like 0:08:01.880 --> 0:08:05.780 a consultant. So you can imagine any of the kind 0:08:05.820 --> 0:08:08.340 of consultant you would love to be able to hire. 0:08:08.360 --> 0:08:11.650 They can be $ 3, 000 a day consultants if you want to. 0:08:12.610 --> 0:08:15.310 When you bring someone into your business as a consultant, 0:08:15.730 --> 0:08:17.890 you won't just go, write me a report or give 0:08:17.910 --> 0:08:20.630 me the analysis. Instead, you'll be going, okay, so here's 0:08:20.650 --> 0:08:22.600 what we've done before. Here's what I'm going to do. 0:08:22.620 --> 0:08:25.560 Here's where I need to go. What are your thoughts 0:08:25.600 --> 0:08:28.810 on this? You're going to have a conversation. do the 0:08:28.910 --> 0:08:33.210 same thing with AI. And so use direct natural language. 0:08:33.890 --> 0:08:37.199 This is where you and I always recommend dictate to AI. 0:08:37.240 --> 0:08:39.340 You're more likely to deal with it like a human 0:08:39.380 --> 0:08:42.199 by giving it the background, the context, what you're trying 0:08:42.220 --> 0:08:44.660 to do, what you've done before, all of those kinds 0:08:44.679 --> 0:08:48.360 of things. And if you're dealing with it like that consultant, 0:08:48.800 --> 0:08:51.800 you're more likely going to give it more background, more context, 0:08:52.020 --> 0:08:54.800 and help it to understand what you're truly after. 0:08:56.320 --> 0:08:59.530 And I should clarify, by dictate, you don't mean be 0:08:59.550 --> 0:09:05.670 a dictator, but press the little microphone button that is 0:09:05.750 --> 0:09:08.920 in all the major AI platforms. It's so much easier 0:09:08.940 --> 0:09:11.480 to speak these things than type these things. 0:09:11.940 --> 0:09:13.340 Sorry, Neo, continue. 0:09:13.380 --> 0:09:14.470 What is tip number three? 0:09:15.360 --> 0:09:17.620 Or just on the treat it like a consultant, I 0:09:17.660 --> 0:09:21.480 would also give a consultant documents. And so rather than 0:09:21.500 --> 0:09:24.260 just say, do this thing, give it the background, the 0:09:24.300 --> 0:09:26.750 context in a document. Say here, read section five, they'll 0:09:26.770 --> 0:09:28.469 give you a nice bit of background. We would do 0:09:28.510 --> 0:09:31.010 that with people, but we'd never do that with Google. 0:09:31.170 --> 0:09:33.250 And so we've got to start giving it much more 0:09:33.309 --> 0:09:35.309 background and context, and it does a much better job 0:09:35.330 --> 0:09:40.530 that way. Tip number three is be inquisitive. So if 0:09:40.650 --> 0:09:43.790 I've got a consultant, they've got a brain, they're really smart. 0:09:43.830 --> 0:09:45.550 That's why I'm paying them to come in to help 0:09:45.600 --> 0:09:47.680 me with these things. And so I'll go, hey, what 0:09:47.700 --> 0:09:50.099 do other clients do? What do other industries do? Is 0:09:50.120 --> 0:09:52.780 that a better way to do that? Can I achieve 0:09:52.800 --> 0:09:55.630 this in a different way that'll land better for my 0:09:55.670 --> 0:09:59.390 customers or clients? So being inquisitive, because it can actually 0:09:59.410 --> 0:10:03.140 show you and teach you ways of doing these things. 0:10:04.260 --> 0:10:06.200 And you might find there's a better way to get 0:10:06.260 --> 0:10:08.100 these results you're trying to do and try to get 0:10:08.160 --> 0:10:08.820 out of AI. 0:10:09.380 --> 0:10:12.560 What is the final tip, Neo? Tip number four. 0:10:13.420 --> 0:10:18.550 Ultimately, challenge it. So if you're a customer of one 0:10:18.570 --> 0:10:20.960 of these consultants, Yeah, they'll come in, they'll tell you 0:10:20.980 --> 0:10:24.300 what different industries and different clients and different companies are doing. 0:10:24.679 --> 0:10:27.200 But you then have to go, yeah, that's great, but 0:10:27.260 --> 0:10:30.290 I'm not those companies or that's not my industry. And 0:10:30.330 --> 0:10:34.290 so it doesn't really apply to me. And so absolutely, 0:10:34.350 --> 0:10:37.699 when you have this conversation with the AI, say, here's 0:10:37.720 --> 0:10:39.860 my instructions, what I'm trying to do. Before you get 0:10:39.880 --> 0:10:42.160 to do the thing, you might actually go, hey, what 0:10:42.200 --> 0:10:43.670 a better way to do it. And it might come 0:10:43.710 --> 0:10:46.130 up with something and you go, that may not land 0:10:46.230 --> 0:10:48.670 for me. There's something you need to look at and 0:10:49.010 --> 0:10:52.830 kind of really judge if you'd like, is that right? 0:10:53.110 --> 0:10:55.510 And if it's not right, challenge it and tell it 0:10:55.570 --> 0:10:57.670 why it's not right so it can then understand what's 0:10:57.690 --> 0:11:00.900 not right. And so this way, you're not just accepting it, 0:11:01.260 --> 0:11:04.120 you're more likely going to understand or get it also 0:11:04.140 --> 0:11:07.059 to understand what you're trying to achieve. Another part there 0:11:07.100 --> 0:11:10.420 is also get it to challenge you as well. So 0:11:10.700 --> 0:11:12.800 I get mine to challenge me not only just on 0:11:12.880 --> 0:11:16.140 how I can create the thing, but also should I 0:11:16.179 --> 0:11:18.630 be even creating that thing in the first place? Are 0:11:18.650 --> 0:11:21.450 there even better ways to achieve that? And so get 0:11:21.490 --> 0:11:23.610 it to challenge you as well as you challenge it 0:11:23.850 --> 0:11:28.670 when it gives you the response back or the document 0:11:28.720 --> 0:11:30.860 or the output or the research or whatever it is, 0:11:31.220 --> 0:11:33.120 and you'll find that you actually get a much better result. 0:11:33.140 --> 0:11:35.980 But ultimately, out of those four tips, it's really deal 0:11:36.020 --> 0:11:39.860 with it like a human being insofar as... It doesn't 0:11:39.880 --> 0:11:41.760 know enough about you. You've got to brief it. It 0:11:41.800 --> 0:11:44.330 can do a good job. It could also misunderstand things. 0:11:44.490 --> 0:11:46.469 You can learn from it. It can learn from you. 0:11:46.809 --> 0:11:49.429 But ultimately, you are the customer of this. And so 0:11:49.450 --> 0:11:51.230 it's got to make sure it lands with you. So 0:11:51.309 --> 0:11:52.250 check those things off. 0:11:52.990 --> 0:11:54.710 Amazing. Thank you for sharing, Neo. 0:11:54.750 --> 0:11:58.930 And hopefully our wonderful listeners have got some very practical 0:11:59.010 --> 0:12:02.220 tips to talk to AI, perhaps in a slightly different 0:12:02.260 --> 0:12:05.949 way today. Thank you so much for listening to today's 0:12:05.970 --> 0:12:09.400 episode of How I AI. If this show got you 0:12:09.420 --> 0:12:12.560 thinking about how you could use AI better, that is 0:12:12.640 --> 0:12:16.480 exactly what we help people do at Inventium AI. We 0:12:16.520 --> 0:12:21.120 run AI training programs for organizations and also for individuals. 0:12:21.540 --> 0:12:25.110 So if you want to upskill your entire organization or 0:12:25.140 --> 0:12:30.420 maybe just yourself, you can find out more at inventium.ai. Hey, 0:12:30.490 --> 0:12:32.670 and once more favor before you go, if you've got 0:12:32.690 --> 0:12:35.829 a moment, like just five seconds, I'd love it if 0:12:35.870 --> 0:12:36.610 you could leave us. 0:12:36.590 --> 0:12:38.809 A review and hit follow wherever you're listening. 0:12:39.150 --> 0:12:41.890 It really helps the show reach more people and it 0:12:41.950 --> 0:12:44.689 helps us create even better episodes for you. 0:12:45.190 --> 0:12:46.910 Thank you so much for listening and we'll see you 0:12:46.950 --> 0:12:48.810 next time on How I AI. 0:12:49.530 --> 0:12:52.530 How I AI was hosted by me, Amanda Imber and 0:12:52.670 --> 0:12:55.210 Neo Applin. A big thank you to Martin Imber who 0:12:55.250 --> 0:12:58.980 does our sound editing and Jem Rubio for production support. 0:12:59.380 --> 0:13:02.300 And thank you to John Kilby who composed the theme music.