WEBVTT - The AI critique system we use to improve our work 

0:00:00.640 --> 0:00:03.600
<v Speaker 1>If you've ever hit send on a document report or

0:00:03.640 --> 0:00:07.520
<v Speaker 1>slide deck and then immediately thought, hmmm, I hope that

0:00:07.680 --> 0:00:08.440
<v Speaker 1>lands okay.

0:00:08.960 --> 0:00:10.280
<v Speaker 2>This episode is.

0:00:10.280 --> 0:00:14.680
<v Speaker 1>For you because producing a high quantity of work isn't

0:00:14.680 --> 0:00:18.880
<v Speaker 1>the hard part anymore. It's producing high quality, which means

0:00:19.079 --> 0:00:22.920
<v Speaker 1>checking it, stress testing it, making sure it's clear, accurate,

0:00:23.040 --> 0:00:27.480
<v Speaker 1>and actually says what you meant before someone else points

0:00:27.520 --> 0:00:30.720
<v Speaker 1>out the gaps. Now, sure, you can enter a basic

0:00:30.800 --> 0:00:33.519
<v Speaker 1>prompt into AI to ask it to review your work.

0:00:33.720 --> 0:00:37.240
<v Speaker 1>But today Neo and I are going to level up

0:00:37.280 --> 0:00:39.880
<v Speaker 1>the way you can use AI to make your work

0:00:40.000 --> 0:00:44.159
<v Speaker 1>better and not to replace your judgment, but to sharpen it.

0:00:44.760 --> 0:00:47.000
<v Speaker 2>You're gonna learn how to get AI to.

0:00:46.960 --> 0:00:50.440
<v Speaker 1>Critique your work properly, not just polish it, ask for

0:00:50.440 --> 0:00:53.760
<v Speaker 1>feedback on clarity, logic, bias, and gaps, and we're gonna

0:00:53.800 --> 0:00:57.800
<v Speaker 1>give you one mega prompt to review emails, documents, slides,

0:00:57.880 --> 0:01:01.160
<v Speaker 1>or anything you like. Really, by the end of this episode,

0:01:01.200 --> 0:01:04.039
<v Speaker 1>you won't be guessing whether your work is good enough.

0:01:04.360 --> 0:01:07.240
<v Speaker 2>You're going to have a clear, repeatable.

0:01:06.640 --> 0:01:10.080
<v Speaker 1>Way to pressure test it before it leaves your desk,

0:01:10.680 --> 0:01:13.720
<v Speaker 1>and the confidence to decide what to change and what

0:01:13.880 --> 0:01:22.800
<v Speaker 1>to stand by. Welcome to how I AI with me

0:01:23.000 --> 0:01:27.280
<v Speaker 1>Doctor Amantha Imba and Neo Applin, head of Inventium AI.

0:01:27.880 --> 0:01:30.959
<v Speaker 1>Each episode we share one practical way to use AI

0:01:31.080 --> 0:01:35.240
<v Speaker 1>better at work and in life. No fluff, no dech jargon,

0:01:35.600 --> 0:01:40.080
<v Speaker 1>just things you can use straight away. So on previous episodes, Neo,

0:01:40.160 --> 0:01:42.120
<v Speaker 1>you've referred to how we can get AI.

0:01:42.080 --> 0:01:45.240
<v Speaker 2>To review our own work.

0:01:45.520 --> 0:01:47.600
<v Speaker 1>But I want to get into the nitty gritty of

0:01:48.040 --> 0:01:50.880
<v Speaker 1>how do we do that beyond just a simple prompt like,

0:01:51.200 --> 0:01:54.680
<v Speaker 1>you know, put your critical thinking hat on AI and

0:01:55.040 --> 0:01:56.920
<v Speaker 1>tell me what I could improve, Like how do we

0:01:56.960 --> 0:01:59.800
<v Speaker 1>get a little bit more sophisticated.

0:01:59.640 --> 0:02:01.440
<v Speaker 3>You need to tell it what you wanted to do

0:02:01.520 --> 0:02:04.360
<v Speaker 3>and also tell it exactly what you wanted to review

0:02:04.560 --> 0:02:08.720
<v Speaker 3>and check. So I've got I've called it Charles, So

0:02:08.720 --> 0:02:12.240
<v Speaker 3>I've created myself an agent to a GPT and Charles

0:02:12.280 --> 0:02:14.360
<v Speaker 3>reviews my stuff. Now I've put in a little bit

0:02:14.360 --> 0:02:17.920
<v Speaker 3>more specifics around it, but at its essence, this more

0:02:18.000 --> 0:02:20.960
<v Speaker 3>than two sentences is what you need to write. So

0:02:21.080 --> 0:02:23.600
<v Speaker 3>I've got review the last response. Of course, you could

0:02:23.639 --> 0:02:26.600
<v Speaker 3>say review the document, review this thing that I've uploaded,

0:02:26.639 --> 0:02:31.080
<v Speaker 3>whatever it is, perform a rigorous critique to find anything incorrect.

0:02:31.800 --> 0:02:39.600
<v Speaker 3>Weekly supported, unclear, incomplete, biased, impractical, verbose, wordy, or improbable.

0:02:39.680 --> 0:02:42.520
<v Speaker 3>So as you can tell, I'm going through all the bases,

0:02:42.560 --> 0:02:44.639
<v Speaker 3>I want you to review it and just go through

0:02:44.800 --> 0:02:47.679
<v Speaker 3>everything that could be wrong with it, then propose better

0:02:47.720 --> 0:02:52.680
<v Speaker 3>alternatives and provide a corrected slash revised version. So you

0:02:52.720 --> 0:02:56.880
<v Speaker 3>could just get away with that one line. So's that's

0:02:57.000 --> 0:02:58.640
<v Speaker 3>enough to be able to get it to review it.

0:02:58.639 --> 0:03:01.160
<v Speaker 3>Because what you're being here is about what you want

0:03:01.160 --> 0:03:03.280
<v Speaker 3>to review it. So you could say I want to

0:03:03.880 --> 0:03:07.000
<v Speaker 3>find instead off anything incorrect, weekly, supported, unclear, and complete.

0:03:07.200 --> 0:03:11.720
<v Speaker 3>You could say anything that is against our policies, anything

0:03:11.840 --> 0:03:16.359
<v Speaker 3>against you, how our operating process are, or anything against

0:03:16.480 --> 0:03:19.040
<v Speaker 3>how I think a CFO would want to review this paper,

0:03:19.240 --> 0:03:21.560
<v Speaker 3>those kind of things. So be specific about how you

0:03:21.639 --> 0:03:24.359
<v Speaker 3>want your agent, your chat thread to be able to

0:03:24.400 --> 0:03:27.440
<v Speaker 3>review your work. So that's an easy thing. So review

0:03:27.440 --> 0:03:29.440
<v Speaker 3>the last response and then put in all those things.

0:03:29.680 --> 0:03:32.679
<v Speaker 3>And we'll put that prompt in the show notes as well.

0:03:33.919 --> 0:03:35.880
<v Speaker 1>And in fact, we will actually do better than that, because

0:03:35.920 --> 0:03:39.160
<v Speaker 1>I think we're going to have the whole system instructions

0:03:39.200 --> 0:03:40.040
<v Speaker 1>for Charles.

0:03:40.160 --> 0:03:40.800
<v Speaker 2>Is that right?

0:03:41.120 --> 0:03:41.240
<v Speaker 1>Oh?

0:03:41.280 --> 0:03:44.120
<v Speaker 3>Okay? And if you're infratrate, okay, we're going to give

0:03:44.120 --> 0:03:47.120
<v Speaker 3>you Charles. So this is what Charles does. So I

0:03:47.200 --> 0:03:49.800
<v Speaker 3>call it Charles because I use it so often, I

0:03:49.800 --> 0:03:51.840
<v Speaker 3>may as well give it a name, So we'll give

0:03:51.880 --> 0:03:55.600
<v Speaker 3>you Charles. So Charles has that task, but Charles also

0:03:55.880 --> 0:03:58.480
<v Speaker 3>has a couple of steps after that that it follows.

0:03:58.720 --> 0:04:00.640
<v Speaker 3>So the first is it goes what do I think

0:04:00.720 --> 0:04:03.400
<v Speaker 3>Neo's really trying to do? And what do I think

0:04:03.400 --> 0:04:05.440
<v Speaker 3>He's really wanting me to check? And that's a really

0:04:05.480 --> 0:04:07.880
<v Speaker 3>important thing for you to see. Is AI actually checking

0:04:07.880 --> 0:04:09.600
<v Speaker 3>the things that I want it to check? And it

0:04:09.640 --> 0:04:13.680
<v Speaker 3>understands me. It goes to equality gates. So in other words,

0:04:13.720 --> 0:04:17.680
<v Speaker 3>what it says is have you passed these main categories

0:04:17.720 --> 0:04:24.520
<v Speaker 3>and the categories I've given it is factual accuracy, logical soundness, completeness, relevance, clarity, structure, safety,

0:04:24.760 --> 0:04:27.360
<v Speaker 3>and practicality. So it goes through those and effectively gives

0:04:27.360 --> 0:04:29.200
<v Speaker 3>me a tick or a cross, so I can see

0:04:29.240 --> 0:04:31.200
<v Speaker 3>if he goes biased, Okay, I've got to go in

0:04:31.240 --> 0:04:33.840
<v Speaker 3>there and change those kind of things, or practical sound,

0:04:34.160 --> 0:04:37.320
<v Speaker 3>factual accuracy, those kind of things. Then it goes through

0:04:37.440 --> 0:04:41.040
<v Speaker 3>and tells me some of the main points that I

0:04:41.080 --> 0:04:43.839
<v Speaker 3>could have added to or I could have changed. So

0:04:43.839 --> 0:04:47.279
<v Speaker 3>it's effectively a diagnostic audit, so it says how important

0:04:47.320 --> 0:04:51.599
<v Speaker 3>this is, how confident it is about how important that is,

0:04:51.720 --> 0:04:54.480
<v Speaker 3>or the problems there. So if you've got a high

0:04:54.520 --> 0:04:57.680
<v Speaker 3>severity like this is really important, but a low confidence

0:04:57.760 --> 0:05:00.240
<v Speaker 3>is like, well, I'm not sure i've been to to

0:05:00.360 --> 0:05:04.040
<v Speaker 3>your work neo. But if I've interpreted right, high importance

0:05:04.200 --> 0:05:08.000
<v Speaker 3>those kind of things, it'll tell you what problem it's got,

0:05:08.000 --> 0:05:11.080
<v Speaker 3>whether it's like a user experience, a structure or clarity whatever,

0:05:11.360 --> 0:05:13.360
<v Speaker 3>and it tells you what the problem is where and

0:05:13.400 --> 0:05:16.280
<v Speaker 3>what you can do to fix it. Also gives you

0:05:16.320 --> 0:05:19.320
<v Speaker 3>a bunch of alternative approaches. So for these things that

0:05:19.400 --> 0:05:21.800
<v Speaker 3>you've got in this work, the document, the email you

0:05:21.800 --> 0:05:24.280
<v Speaker 3>wanted to review, whatever it is, it will then say,

0:05:24.360 --> 0:05:28.320
<v Speaker 3>here's some alternative approaches on how you can potentially fix this.

0:05:29.600 --> 0:05:32.760
<v Speaker 1>So can I understand just day to day because I

0:05:33.279 --> 0:05:36.240
<v Speaker 1>feel like I hear you use Charles a lot, So

0:05:36.560 --> 0:05:39.480
<v Speaker 1>what are the kinds of tasks that you're bringing in

0:05:40.000 --> 0:05:43.560
<v Speaker 1>Charles for? And just to confirm, Charles is a GPT

0:05:43.720 --> 0:05:46.880
<v Speaker 1>and chat GPT for you, Is that right? Yes, that's great, yes,

0:05:47.000 --> 0:05:50.000
<v Speaker 1>But equally it could be a project in Chlaude. It

0:05:50.040 --> 0:05:52.520
<v Speaker 1>could be an agent in co Pilot, it could be

0:05:52.520 --> 0:05:54.040
<v Speaker 1>a gem in Gemini.

0:05:54.040 --> 0:05:56.000
<v Speaker 3>Yes, absolutely, or it could just be a prompt in

0:05:56.040 --> 0:05:57.920
<v Speaker 3>your prompt library you dump in whenever you need.

0:05:58.320 --> 0:06:00.719
<v Speaker 1>Okay, so tell me what are the kinds of tasks

0:06:00.720 --> 0:06:01.800
<v Speaker 1>that you're using Charles for?

0:06:03.040 --> 0:06:07.080
<v Speaker 3>Anything with analysis, I'm getting it to check my analysis.

0:06:07.360 --> 0:06:10.800
<v Speaker 3>So if I'm doing a review or some kind of

0:06:10.800 --> 0:06:13.160
<v Speaker 3>a document or whatever, then I'm getting Charles to review

0:06:13.200 --> 0:06:15.960
<v Speaker 3>that and give me their or its opinion on what

0:06:16.080 --> 0:06:19.720
<v Speaker 3>I think what it thinks can be improved in this work. Now,

0:06:19.760 --> 0:06:22.599
<v Speaker 3>by the way, sometimes Charles comes up with suggestions like, yeah,

0:06:22.680 --> 0:06:24.920
<v Speaker 3>I get that, but no, you don't know the background

0:06:25.000 --> 0:06:26.600
<v Speaker 3>like I do, and so I don't have to change it.

0:06:27.640 --> 0:06:32.160
<v Speaker 3>But those kind of things I was running through Charles.

0:06:32.240 --> 0:06:35.440
<v Speaker 3>I'd done a social post. We've got online communities for

0:06:35.480 --> 0:06:38.640
<v Speaker 3>AI people learning AI, and I had already done a

0:06:38.640 --> 0:06:40.600
<v Speaker 3>social post, and I thought, oh, I forgot to put

0:06:40.640 --> 0:06:44.719
<v Speaker 3>that through Charles, and so I posting it. I put

0:06:44.760 --> 0:06:47.240
<v Speaker 3>the social post into Charles, and it gave me a

0:06:47.240 --> 0:06:49.080
<v Speaker 3>couple of suggestions on how I could quickly go in

0:06:49.120 --> 0:06:51.119
<v Speaker 3>there and do some tweaks to make my social post

0:06:51.240 --> 0:06:54.720
<v Speaker 3>clear out so it can be anything from social posts

0:06:54.800 --> 0:06:59.000
<v Speaker 3>to emails to documentsy working on analysis as you're doing

0:06:59.360 --> 0:07:02.720
<v Speaker 3>slide you can give I've given slide decks to in

0:07:02.760 --> 0:07:05.880
<v Speaker 3>the past, a whole bunch of different things, so pretty

0:07:05.920 --> 0:07:08.240
<v Speaker 3>much anything you're doing. If you would love to get

0:07:08.279 --> 0:07:11.200
<v Speaker 3>a second opinion, then get Charles to give you that

0:07:11.240 --> 0:07:15.280
<v Speaker 3>second opinion. So I'm not quite there. I still haven't

0:07:15.280 --> 0:07:18.560
<v Speaker 3>built it into everything I do, but anything important I

0:07:18.560 --> 0:07:20.400
<v Speaker 3>get Charles to have a look at. But I think

0:07:20.400 --> 0:07:24.400
<v Speaker 3>it's almost like a spell check, Like we get word

0:07:24.480 --> 0:07:26.600
<v Speaker 3>to do a spell check and a grammar check for

0:07:26.640 --> 0:07:29.120
<v Speaker 3>anything we then do and then upload or give to

0:07:30.560 --> 0:07:33.320
<v Speaker 3>a colleague. It's a good idea to get AI to

0:07:33.400 --> 0:07:36.600
<v Speaker 3>do a review to see whether you can improve those

0:07:36.640 --> 0:07:39.000
<v Speaker 3>things before you give it to those people. Why that

0:07:39.040 --> 0:07:41.080
<v Speaker 3>review doesn't take much because of course AI is just

0:07:41.080 --> 0:07:43.400
<v Speaker 3>doing the review. You can agree or disagree with those things,

0:07:43.480 --> 0:07:45.440
<v Speaker 3>and quite often it's like, oh, that's a really good pickup,

0:07:45.440 --> 0:07:47.680
<v Speaker 3>and I'm just going to do small tweaks. So it's

0:07:47.760 --> 0:07:50.640
<v Speaker 3>like the third thing we're doing spell check, grammar check,

0:07:50.840 --> 0:07:52.560
<v Speaker 3>get it to do a your Charle's check, and of

0:07:52.560 --> 0:07:53.800
<v Speaker 3>course we're giving you the we're going to give you

0:07:53.840 --> 0:07:55.800
<v Speaker 3>the COVID Charles, so you can run that yourself.

0:07:56.520 --> 0:07:59.280
<v Speaker 1>How well does it work if AI has been primarily

0:07:59.360 --> 0:08:04.200
<v Speaker 1>responsible for creating the output, Like, let's just say, you know,

0:08:04.360 --> 0:08:07.120
<v Speaker 1>as sadly many people are, they're just kind of taking

0:08:07.120 --> 0:08:08.000
<v Speaker 1>shortcuts with AI.

0:08:08.040 --> 0:08:10.400
<v Speaker 2>They're being a bit lazy. They're producing some AI slop.

0:08:11.280 --> 0:08:13.600
<v Speaker 1>You know, maybe they've put in three bullet points and

0:08:13.600 --> 0:08:15.800
<v Speaker 1>they're like, write me a three page report that I

0:08:15.840 --> 0:08:18.680
<v Speaker 1>can give to my bus on this particular topic or

0:08:18.720 --> 0:08:22.040
<v Speaker 1>project update. Is it good at critiquing its own work

0:08:22.160 --> 0:08:26.480
<v Speaker 1>when the prompter's been lackluster and you know, maybe the

0:08:26.480 --> 0:08:28.320
<v Speaker 1>AI's output is actually not that good.

0:08:28.960 --> 0:08:31.880
<v Speaker 3>Yeah, I was a bet to joke and say it'll

0:08:31.960 --> 0:08:34.640
<v Speaker 3>recognize its own handywork and say that's awesome. No, it

0:08:34.679 --> 0:08:39.040
<v Speaker 3>doesn't do that. It'll even improve AI's output as well.

0:08:39.720 --> 0:08:42.160
<v Speaker 3>So it's a good thing to get this as a

0:08:42.160 --> 0:08:44.760
<v Speaker 3>second opinion AI. So if you've been working with A

0:08:44.920 --> 0:08:46.960
<v Speaker 3>on a document or whatever, even if you have just

0:08:47.000 --> 0:08:49.160
<v Speaker 3>been like phoning it in and saying five top points,

0:08:49.240 --> 0:08:51.520
<v Speaker 3>give me a document, it will still review that and

0:08:51.760 --> 0:08:55.440
<v Speaker 3>tell you how that could be improved. So don't take

0:08:55.520 --> 0:09:00.160
<v Speaker 3>AI's output first output as gospel ever, like I prefer,

0:09:00.280 --> 0:09:03.600
<v Speaker 3>using AI's output is to augment me, not to do

0:09:03.640 --> 0:09:06.280
<v Speaker 3>the work for me. But yes, you can get it

0:09:06.320 --> 0:09:08.200
<v Speaker 3>to review AI's output and we'll actually give you a

0:09:08.200 --> 0:09:10.640
<v Speaker 3>better response and'll actually tell you where the flaws are

0:09:10.880 --> 0:09:14.319
<v Speaker 3>where it could be improved. So yes, absolutely use it for.

0:09:14.320 --> 0:09:20.440
<v Speaker 1>That amazing So for anyone listening, we are very or

0:09:20.480 --> 0:09:25.040
<v Speaker 1>i should say NEO is very generously sharing Charles with you. All.

0:09:25.280 --> 0:09:27.439
<v Speaker 1>Just pop into the show notes and there will be

0:09:27.800 --> 0:09:31.440
<v Speaker 1>a link to get the full text in markdown to

0:09:31.600 --> 0:09:36.400
<v Speaker 1>paste into your GPT agent project of choice, or.

0:09:36.400 --> 0:09:38.800
<v Speaker 2>You can just use it as a mega prompt.

0:09:38.840 --> 0:09:43.079
<v Speaker 1>So enjoy Charles, maybe give it its own name, and

0:09:43.120 --> 0:09:47.440
<v Speaker 1>we will see you next Monday. If you found this useful,

0:09:47.480 --> 0:09:50.480
<v Speaker 1>please help us spread the AI love and share it

0:09:50.520 --> 0:09:53.240
<v Speaker 1>with someone who you think would benefit from knowing what

0:09:53.400 --> 0:09:56.520
<v Speaker 1>you now know. And if you're ready to really start

0:09:56.600 --> 0:10:01.640
<v Speaker 1>mastering AI, check out inventium dot ai. We help individuals, teams,

0:10:01.679 --> 0:10:06.959
<v Speaker 1>and organizations turn Jenai into a real work superpower, saving

0:10:07.200 --> 0:10:10.240
<v Speaker 1>ten plus hours a week and staying future already without

0:10:10.240 --> 0:10:11.880
<v Speaker 1>the jargon or overwhelmed.

0:10:12.160 --> 0:10:14.120
<v Speaker 2>Thanks so much for listening, and we'll see you next

0:10:14.160 --> 0:10:14.760
<v Speaker 2>time on.

0:10:14.679 --> 0:10:19.280
<v Speaker 1>How IAI how i AI was hosted by me Amantha

0:10:19.320 --> 0:10:20.760
<v Speaker 1>Imber and Neo Applan.

0:10:21.040 --> 0:10:23.000
<v Speaker 2>A big thank you to Martin Imber who does our

0:10:23.080 --> 0:10:24.120
<v Speaker 2>sound editing, and

0:10:24.240 --> 0:10:27.640
<v Speaker 1>Jem Rubio for production support, and thank you to John

0:10:27.720 --> 0:10:29.680
<v Speaker 1>Kilby who composed the theme music.