1 00:00:00,520 --> 00:00:04,320 Speaker 1: If you've ever fallen down a Google rabbit hole trying 2 00:00:04,320 --> 00:00:07,480 Speaker 1: to prepare for a workshop, or write a strategy paper, 3 00:00:08,039 --> 00:00:10,680 Speaker 1: or just get your head around a new topic, you 4 00:00:10,800 --> 00:00:16,480 Speaker 1: are going to love this episode. Today, Neo Applan, Inventium's 5 00:00:16,720 --> 00:00:21,400 Speaker 1: Jenai Guru, is back to explain one of Jenai's most 6 00:00:21,520 --> 00:00:27,640 Speaker 1: powerful hidden tools, deep research. It's like outsourcing four hours 7 00:00:27,680 --> 00:00:36,680 Speaker 1: of online digging in five minutes. Welcome to How I Work, 8 00:00:36,920 --> 00:00:41,040 Speaker 1: a show about habits, rituals, and strategies for optimizing your day. 9 00:00:41,640 --> 00:00:45,360 Speaker 1: I'm your host, Doctor Amantha Imber. Before we get into 10 00:00:45,360 --> 00:00:48,599 Speaker 1: today's episode, I have been geeking out over the research 11 00:00:48,680 --> 00:00:51,760 Speaker 1: on AI adoption in the workplace, and the data is 12 00:00:51,920 --> 00:00:56,440 Speaker 1: really clear. The people who master Jenai now aren't just 13 00:00:56,480 --> 00:01:01,120 Speaker 1: getting ahead, they are creating entirely new level of productivity 14 00:01:01,320 --> 00:01:05,039 Speaker 1: that seemed impossible just like a year ago. Because in 15 00:01:05,120 --> 00:01:09,280 Speaker 1: a nutshell, there are two types of professionals, those who 16 00:01:09,319 --> 00:01:13,319 Speaker 1: are fluent in AI and those left scrambling to catch up. 17 00:01:14,200 --> 00:01:18,120 Speaker 1: Inventium's Genai Productivity Upgrade is how you can make sure 18 00:01:18,160 --> 00:01:20,960 Speaker 1: that you are in that first group. It is a 19 00:01:21,040 --> 00:01:23,679 Speaker 1: twelve week course designed to move you from dabbler to 20 00:01:23,720 --> 00:01:27,880 Speaker 1: productivity machine. No fluff, just practical strategies that will pay 21 00:01:27,920 --> 00:01:31,320 Speaker 1: off from week one, saving you at least ten hours 22 00:01:31,360 --> 00:01:34,560 Speaker 1: every single week. You'll learn how to make AI sound 23 00:01:34,600 --> 00:01:37,840 Speaker 1: exactly like you use AI as your second brain to 24 00:01:37,920 --> 00:01:40,440 Speaker 1: excel at your job and use it to become a 25 00:01:40,520 --> 00:01:45,080 Speaker 1: much more effective leader, and so much more. Whether you're 26 00:01:45,120 --> 00:01:48,680 Speaker 1: a complete beginner or already dabbling, we've got you covered, 27 00:01:48,960 --> 00:01:52,040 Speaker 1: starting with prompting fundamentals and going all the way through 28 00:01:52,080 --> 00:01:57,320 Speaker 1: to advanced automations and digentic AI week kickoff on July fourteen, 29 00:01:57,680 --> 00:02:01,919 Speaker 1: and spots are limited. Visit inventi do au forward slash 30 00:02:02,120 --> 00:02:05,960 Speaker 1: Jenai hyphen cohort to secure your place now and there's 31 00:02:05,960 --> 00:02:08,280 Speaker 1: a link to that in the show notes. You've got 32 00:02:08,320 --> 00:02:12,079 Speaker 1: nothing to lose because there's a seven day money back guarantee, 33 00:02:12,520 --> 00:02:14,640 Speaker 1: so head to the show notes click on the link 34 00:02:14,760 --> 00:02:18,960 Speaker 1: to read more about the program today. Okay, neo, what 35 00:02:19,160 --> 00:02:24,079 Speaker 1: exactly is deep research and what problem is it solving? Oh? 36 00:02:24,160 --> 00:02:26,079 Speaker 2: You know how when you've got to do a lot 37 00:02:26,080 --> 00:02:27,680 Speaker 2: of research and you think I'm going to have to 38 00:02:27,680 --> 00:02:30,799 Speaker 2: spend a couple of hours working with Google to learn 39 00:02:30,840 --> 00:02:33,400 Speaker 2: about this thing, to research about this thing, so it 40 00:02:33,520 --> 00:02:36,400 Speaker 2: might be a lot of manual research that you need 41 00:02:36,440 --> 00:02:39,640 Speaker 2: to do instead of doing that manually, which can take 42 00:02:39,720 --> 00:02:41,880 Speaker 2: many hours, so you can learn the topic and know 43 00:02:41,960 --> 00:02:44,960 Speaker 2: that you're getting the right information. You can automate that 44 00:02:45,280 --> 00:02:49,560 Speaker 2: with using Jenai tools and the deep research function. So 45 00:02:49,600 --> 00:02:51,760 Speaker 2: it's great if you want to do things like market trends, 46 00:02:52,440 --> 00:02:56,680 Speaker 2: best practices, science updates and all those kind of things. Effectively, 47 00:02:56,720 --> 00:02:59,360 Speaker 2: you're getting the ai to be your research assistant and 48 00:02:59,440 --> 00:03:01,800 Speaker 2: it goes away. It plans what the research is going 49 00:03:01,840 --> 00:03:04,639 Speaker 2: to be, and it goes and searches so many websites 50 00:03:04,639 --> 00:03:04,840 Speaker 2: for you. 51 00:03:05,720 --> 00:03:11,080 Speaker 1: Okay, So which genai tools actually offer deep research and 52 00:03:11,160 --> 00:03:12,000 Speaker 1: how do they differ? 53 00:03:13,600 --> 00:03:16,280 Speaker 2: The main functions are pretty much the same between the 54 00:03:16,840 --> 00:03:20,560 Speaker 2: different ais. So chatchpt does it on the paid and 55 00:03:20,760 --> 00:03:23,840 Speaker 2: the free version. You get five deep researchers a month 56 00:03:23,960 --> 00:03:26,680 Speaker 2: on the free and thirty a month on the paid. 57 00:03:27,280 --> 00:03:31,200 Speaker 2: Perplexity has it, as well as Gemini, which is Google's 58 00:03:31,360 --> 00:03:35,840 Speaker 2: Genai tool. And yeah, they're all very good. They're slightly 59 00:03:35,840 --> 00:03:38,440 Speaker 2: different in the way that they approach the results. However, 60 00:03:38,640 --> 00:03:41,920 Speaker 2: I'd say play with one many of them. They're awesome 61 00:03:41,960 --> 00:03:43,440 Speaker 2: at what they do. 62 00:03:43,440 --> 00:03:44,880 Speaker 1: Do you have a personal. 63 00:03:44,480 --> 00:03:48,520 Speaker 2: Favorite At the moment, I'm enjoying Gemini, and the reason 64 00:03:48,560 --> 00:03:51,120 Speaker 2: for that is that I can use it as much 65 00:03:51,120 --> 00:03:53,960 Speaker 2: as I want to. It's free at the moment. So 66 00:03:54,000 --> 00:03:58,320 Speaker 2: I do many deep researchers, but look, all of them 67 00:03:58,360 --> 00:03:59,960 Speaker 2: are excellently in the way that they operate. 68 00:04:00,960 --> 00:04:03,040 Speaker 1: And I think what's useful with Gemini If you are 69 00:04:03,160 --> 00:04:06,960 Speaker 1: on the Google ecosystem, it exports the deep research report 70 00:04:07,000 --> 00:04:10,200 Speaker 1: to Google Docs and then chat, GPT and Perplexity export 71 00:04:10,280 --> 00:04:12,960 Speaker 1: to word and PDF, so that is useful as well. 72 00:04:13,880 --> 00:04:17,000 Speaker 1: Now I want to know what is actually happening in 73 00:04:17,040 --> 00:04:20,799 Speaker 1: the background when AI is doing its deep research. 74 00:04:21,600 --> 00:04:24,240 Speaker 2: It's a true agentic flow and I'm doing like air 75 00:04:24,320 --> 00:04:27,320 Speaker 2: quotes here agentic, so gentic is where the AI is 76 00:04:27,360 --> 00:04:30,039 Speaker 2: going off and doing things on its own. So it's 77 00:04:30,080 --> 00:04:33,159 Speaker 2: great where the interface actually shows you what the AI 78 00:04:33,320 --> 00:04:36,320 Speaker 2: is thinking about in order to do that deep research. 79 00:04:36,680 --> 00:04:38,440 Speaker 2: So it will say these are the sources that they 80 00:04:38,520 --> 00:04:40,520 Speaker 2: need to look at. Then I need to look at this, 81 00:04:40,720 --> 00:04:43,359 Speaker 2: Then I need to research this particular thing, and it 82 00:04:43,360 --> 00:04:46,400 Speaker 2: will then plan how it's going to approach this problem 83 00:04:46,480 --> 00:04:48,760 Speaker 2: and all the different types of pages it's going to search. 84 00:04:49,560 --> 00:04:52,440 Speaker 2: So behind the scenes, it's doing the plan first. Now 85 00:04:52,480 --> 00:04:54,400 Speaker 2: before you get it to do the work. You can 86 00:04:54,440 --> 00:04:56,200 Speaker 2: agree with that plan or not, and you could say 87 00:04:56,320 --> 00:04:57,919 Speaker 2: I need to do more of this than that. So 88 00:04:58,320 --> 00:05:01,479 Speaker 2: work with it first the plan, and then when you 89 00:05:01,480 --> 00:05:04,000 Speaker 2: hear the go button on the deep research button, then 90 00:05:04,040 --> 00:05:06,480 Speaker 2: it does exactly what it planned to. So it'll go 91 00:05:06,520 --> 00:05:09,840 Speaker 2: and research many different websites. So I've had ones where 92 00:05:09,880 --> 00:05:12,080 Speaker 2: it's like one hundred and fifty websites kind of common. 93 00:05:12,440 --> 00:05:15,120 Speaker 2: I've had some up to four hundred and ninety different 94 00:05:15,120 --> 00:05:17,599 Speaker 2: websites that it's looking at in order for it to 95 00:05:17,640 --> 00:05:21,359 Speaker 2: go through the plan. So it gives you so much 96 00:05:21,480 --> 00:05:24,240 Speaker 2: information from those sites. And then what it does is 97 00:05:24,279 --> 00:05:28,320 Speaker 2: it builds you a report. And the report is very detailed. 98 00:05:28,320 --> 00:05:31,200 Speaker 2: We're talking on it five ten pages worth of information. 99 00:05:31,640 --> 00:05:34,520 Speaker 2: Plus it also sites where it got the information for 100 00:05:34,880 --> 00:05:37,120 Speaker 2: so you can click on the links and learn more 101 00:05:37,160 --> 00:05:38,920 Speaker 2: and just make sure that it's got it right. So 102 00:05:39,320 --> 00:05:41,800 Speaker 2: in summary, it plans the search, work with it on 103 00:05:41,839 --> 00:05:45,000 Speaker 2: the plan. It then does the search, and it produces 104 00:05:45,040 --> 00:05:47,279 Speaker 2: you an excellent report, so you've got an outcome. 105 00:05:47,880 --> 00:05:52,000 Speaker 1: Okay, So something I have wondered about is how exactly 106 00:05:52,480 --> 00:05:55,880 Speaker 1: should I best write the prompt to get the best 107 00:05:55,920 --> 00:05:58,640 Speaker 1: result from deep research? Because I know that in the 108 00:05:58,680 --> 00:06:01,320 Speaker 1: past I've sometimes been quite lazy. I've just written a 109 00:06:01,360 --> 00:06:05,000 Speaker 1: quick one sentence prompt. I mean, the results have been 110 00:06:05,040 --> 00:06:08,040 Speaker 1: pretty good, But ideally, what should I be doing with 111 00:06:08,240 --> 00:06:08,960 Speaker 1: the prompt? 112 00:06:09,920 --> 00:06:12,320 Speaker 2: Same as any other prompt, If it knows what you're 113 00:06:12,360 --> 00:06:14,520 Speaker 2: trying to achieve, and it knows what the goals are 114 00:06:14,560 --> 00:06:16,800 Speaker 2: and the context and things like that, you will get 115 00:06:16,800 --> 00:06:20,040 Speaker 2: a much better result. So instead of saying what's the 116 00:06:20,200 --> 00:06:24,600 Speaker 2: modern strategy for banking or something like that, very generic, 117 00:06:25,320 --> 00:06:28,000 Speaker 2: you need to say what specific strategy you're looking at. 118 00:06:28,320 --> 00:06:30,719 Speaker 2: So this is the actual area I'm looking at, and 119 00:06:30,760 --> 00:06:32,719 Speaker 2: these are the types of things I'm trying to solve. 120 00:06:32,839 --> 00:06:36,040 Speaker 2: So work with the AI on this. You can even 121 00:06:36,080 --> 00:06:38,919 Speaker 2: do what they call meta prompting, which is working with 122 00:06:38,960 --> 00:06:41,040 Speaker 2: the AI giving it all the context and what you're 123 00:06:41,040 --> 00:06:43,800 Speaker 2: trying to achieve and then get it to produce you 124 00:06:43,839 --> 00:06:46,240 Speaker 2: a prompt, which is a nice little cheats way. But 125 00:06:46,360 --> 00:06:49,640 Speaker 2: the prompt is not the most important part. It is 126 00:06:50,240 --> 00:06:54,280 Speaker 2: equally as important as working with AI on that search strategy, 127 00:06:54,440 --> 00:06:56,360 Speaker 2: So making sure it's going to search the right kind 128 00:06:56,360 --> 00:06:58,080 Speaker 2: of pages and trying to get the right kind of 129 00:06:58,080 --> 00:07:01,159 Speaker 2: outcomes in the two of them together equally is important 130 00:07:01,320 --> 00:07:02,320 Speaker 2: for a great result. 131 00:07:03,120 --> 00:07:05,440 Speaker 1: And what's your advice on working with the AI to 132 00:07:05,680 --> 00:07:09,240 Speaker 1: design a good approach to doing the deep research? 133 00:07:10,960 --> 00:07:14,040 Speaker 2: Saying my goals are and here's the context, like it's 134 00:07:14,040 --> 00:07:16,760 Speaker 2: the business, here's what I'm trying to achieve. These are 135 00:07:16,760 --> 00:07:19,120 Speaker 2: the two main things that it really needs to know 136 00:07:19,200 --> 00:07:22,440 Speaker 2: to be able to nail a great deep research prompt 137 00:07:22,600 --> 00:07:25,600 Speaker 2: or a deep research process. So once you've got those 138 00:07:25,680 --> 00:07:28,720 Speaker 2: things absolutely agreed with the AI, then you'll find you're 139 00:07:28,720 --> 00:07:31,520 Speaker 2: going to get a much better result. So better background 140 00:07:31,520 --> 00:07:33,840 Speaker 2: and better context you can give the better the results. 141 00:07:34,400 --> 00:07:37,320 Speaker 1: Okay, Now, a question that I'm sure a lot of 142 00:07:37,640 --> 00:07:42,080 Speaker 1: listeners are asking in their brains is can we trust 143 00:07:42,080 --> 00:07:45,720 Speaker 1: the results? I mean, we all know about hallucinations, where 144 00:07:46,080 --> 00:07:50,000 Speaker 1: the AI just makes stuff up, it creates sources out 145 00:07:50,040 --> 00:07:53,640 Speaker 1: of nothing. Is deep research different in that sense? 146 00:07:54,640 --> 00:07:58,520 Speaker 2: It isn't in that you can get hallucinations. However, I 147 00:07:58,600 --> 00:08:01,440 Speaker 2: found that there are few hallucinations, and the reason for 148 00:08:01,520 --> 00:08:05,040 Speaker 2: that is it's effectively summarizing web pages that it has 149 00:08:05,120 --> 00:08:08,880 Speaker 2: already just read. It's not hallucinating out of its own knowledge. 150 00:08:08,920 --> 00:08:11,160 Speaker 2: And if there's a knowledge gap, there's a problem. And 151 00:08:11,240 --> 00:08:13,440 Speaker 2: so out of the box you will find that there 152 00:08:13,880 --> 00:08:17,720 Speaker 2: are fewer hallucinations. The worst that I've found is it's 153 00:08:17,960 --> 00:08:21,800 Speaker 2: interpreted a page slightly differently than I would have interpreted it. 154 00:08:22,160 --> 00:08:25,440 Speaker 2: And the page may have been for a tangentially related 155 00:08:25,920 --> 00:08:28,360 Speaker 2: problem or industry, and I can see how it's got 156 00:08:28,360 --> 00:08:31,600 Speaker 2: the things kind of together, but in reality, I would 157 00:08:31,600 --> 00:08:33,760 Speaker 2: have said, yeah, probably that page wouldn't have been something 158 00:08:33,840 --> 00:08:37,000 Speaker 2: that I would have included on my search. So it's 159 00:08:37,440 --> 00:08:40,560 Speaker 2: very accurate. It's more of an interpretation issue that I 160 00:08:40,600 --> 00:08:43,480 Speaker 2: have found, and even then it's really rare for it 161 00:08:43,520 --> 00:08:45,240 Speaker 2: of me to have found it after all the deep 162 00:08:45,280 --> 00:08:49,960 Speaker 2: researches that I've done, so it is a very thorough search. 163 00:08:50,200 --> 00:08:52,720 Speaker 2: It is really good at summarizing the pages that it 164 00:08:52,760 --> 00:08:55,200 Speaker 2: has read, and even if it has read fo one 165 00:08:55,240 --> 00:08:57,839 Speaker 2: hundred and eighty pages do notes, so just because it's 166 00:08:57,880 --> 00:09:00,640 Speaker 2: done a deep research that it isn't the world's expert 167 00:09:00,679 --> 00:09:04,240 Speaker 2: on those things. So it's just summarized the four hundred 168 00:09:04,240 --> 00:09:06,880 Speaker 2: odd pages that it's read, and it's done an awesome 169 00:09:06,960 --> 00:09:10,360 Speaker 2: job at the summarization. However, you might need to still 170 00:09:10,400 --> 00:09:12,560 Speaker 2: do some more research on top of it so know 171 00:09:12,679 --> 00:09:16,440 Speaker 2: that it could potentially hallucinate. It also might not have 172 00:09:16,520 --> 00:09:19,360 Speaker 2: searched a really important page that's out there, just as 173 00:09:19,400 --> 00:09:21,080 Speaker 2: you and I when we do a Google search, we 174 00:09:21,160 --> 00:09:24,000 Speaker 2: may not have searched every paper out there. So the 175 00:09:24,200 --> 00:09:28,080 Speaker 2: probably the biggest risk, biggest takeaway is review everything. Click 176 00:09:28,120 --> 00:09:30,160 Speaker 2: on it, and if you're wanting to know more about it, 177 00:09:30,200 --> 00:09:32,400 Speaker 2: click on those links so you can be sure that 178 00:09:32,480 --> 00:09:33,600 Speaker 2: everything in there is correct. 179 00:09:34,679 --> 00:09:39,640 Speaker 1: Neo, thank you so much for taking us through deep research. 180 00:09:39,840 --> 00:09:43,280 Speaker 1: It is hopefully a function in people's GENAI that they 181 00:09:43,360 --> 00:09:48,640 Speaker 1: are going to now use a lot more frequently. If 182 00:09:48,640 --> 00:09:51,760 Speaker 1: you like today's jow, make sure you follow on your 183 00:09:51,760 --> 00:09:55,559 Speaker 1: podcast app to be alerted when new episodes drop. How 184 00:09:55,640 --> 00:09:58,000 Speaker 1: I Work was recorded on the traditional land of the 185 00:09:58,000 --> 00:10:01,000 Speaker 1: Warrangery People, part of the Cool and Nake. A big 186 00:10:01,040 --> 00:10:03,360 Speaker 1: thank you to Martin Imma for doing the sound mix.