Venture Step

Dalton Anderson (00:01.75)
Welcome to Venture Step Podcasts where we discuss entrepreneurship, industry trends, and the occasional book review. Ever get tired of spending hours researching a topic? Well, maybe you do, but you want to research more or maybe you don't and you just want to move on with your life. That opportunity is there now for just $20 a month. You can access your own AI analyst.

That's right, folks. You can initiate a prompt to Gemini and utilize their deep research feature, which will allow you to set and forget, and then the analyst will come back once it has its findings.

In the future, this will become more advanced and you could research complex topics hours at a time. Currently, I think if you're just on the advanced, which is like the paid tier, but non-enterprise advance, they limit it to five minutes. And then for enterprise, I'm not completely sure.

on the limitations of how long I can think for, but they did talk about enhancing the length of the thinking time. So the longer the thinking time theoretically means you could research or have requests, more complex research results and topics and just in general requests that would take longer. So if it takes AI 30 minutes, it might take you five hours, four hours to do.

So that being said, the episode is going to cover a live demo. What has been my experiences with deep research and.

Dalton Anderson (02:07.168)
I have three examples and I've used it on prior episodes. So what I did with the Grayball episode, I was doing some research and then I was like, well, I don't know where to go. Right. Like I could research the internet, but what's some interesting stuff that I need to look at? And normally he'd just go to Wikipedia. But I was like, let me try this deep research thing because I tried it before two episodes ago, three, guess now for who knows?

But the previous technical episode that I did with the CoTracker, I wrote my own notes about the research paper and they're very detailed. And then I asked the deep researcher to go, or the AI analyst, I like to say AI analyst, the AI analyst to go and research about this CoTracker 3, read the research paper, look online, look at the code, whatever.

and then come back to us. And I told it to just look for just general information and summarize and make sure you comprehend it and talk about it as if you had to present it to a group of researchers at a panel. And your level of expertise is recent PhD grad. And it came back with some pretty good results and

they were up to par, similar to what I was thinking. Not as long, but general sense, yeah, it was pretty nice. And I was like, wow, this is great. And then I tried the same thing for a gray ball where I wrote maybe half my notes, maybe 70%. And then I was like, let me have the AI analysts research this other 30 or something.

other 40%. And it did a pretty good job, because I still researched it myself, because just because the AI analyst does the research for me, I still have to talk about it on the podcast episode. So it's not very relevant for me in the area of podcasting, but it is relevant in just answering curiosities and or.

Dalton Anderson (04:30.638)
just figuring out things that you don't necessarily have time to research. Like what would be cool is if they could increase it to 30 minutes or an hour and then you could, you could like stop somewhere. You're at a roadblock, right? And instead of you getting frustrated and taking time out of your day to like try to figure out and troubleshoot, which is good. It's a good learning experience. Don't take that away from the kids. But as an adult, you're paid to get stuff done.

And so no one really cares how you get your stuff done, as long as it's done accurately. And you know what you're talking about when you're asked questions. So it would be great if you could increase the thinking time to an hour. You're frustrated, whatever the case is, you go work out or you go pick up your, your son or daughter, or get lunch with your partner. And then you come back to work to the office or your home office and

the AI analyst has done a whole bunch of research for you and troubleshooted different things. That would be phenomenal. Like what a lift that would be. And you wouldn't necessarily do less work because eventually everyone would start doing it and then you just have to do more work. But the work that you're doing is...

Dalton Anderson (05:50.318)
potentially more valuable. I don't know. But before we dive in, I'm your host Dalton Anderson. I've got a bit of a mix of background in programming, data science and insurance. Offline, you can find me running, my side business or lost in a good book. You can find this podcast on video or you can find it on YouTube in video and audio format. If audio is more your thing, then you can find the podcast on Apple podcasts, Spotify.

or wherever else you get your podcasts. Okay, so recently, Gemini has a new feature or model that you can access as a user, an advanced user that allows you to request deep research. And you give it a prompt, it will confirm that it understands what you're asking for, and it will walk you through the steps that they're gonna execute.

And then it will just go off on its own and it'll go start browsing the web. shows you what websites it goes to. And then it creates a report for you that is very detailed. And then it can be exported to your Google doc or you can just copy and paste it yourself. I've used it four times now, four or three times. It works pretty well and aligns with what I was looking for.

either when I had already done the work and I'm reviewing what they wrote, or if I am close to completing my work and it was bringing me to the finish line. Those are the two scenarios that I've experienced. I've done twice where I'm checking the actual work where I've already done the work a hundred percent. And then the other time I was pushed to the finish line and those scenarios, it worked pretty well. I thought it was great. And

I wanna share it with everyone. So before I do that, let me share my screen and we are going to do some prompts that are already created for us. why I them before the episode, but yes, they were pre-generated before the episode inception.

Dalton Anderson (08:15.82)
Okay, so I am going to pull up the prompts here and then on my other screen, and then I'm sharing my main screen. Just give me one moment. And I'll do that.

Dalton Anderson (08:29.856)
Okay, so let me show my screen.

Dalton Anderson (08:41.998)
Think if they share the window.

Dalton Anderson (08:50.272)
All right, so we should be seeing.

this screen. I'm sharing this window. If I want to switch this window, will it tell me to

Dalton Anderson (09:08.654)
All right, cool. So now I get to see that I'm actually sharing. Cool, cool, cool. All right, so let's put you back over here. All right, so this first thing is going to be e-comm. So I have an e-commerce prompt that I want to put in here that I helped create in conjunction of AI enhancing it. So this is a long prompt, but basically it's asking the general sense is we're setting the stage for

e-commerce trends, like what are the current trends? Like would it be VR, social commerce, social brand building, or when you build the brand together, like where you walk the audience through, like how you built the product and step by step, like building the business. Looking at the emerging product sectors. And then the last thing was speculation and predictions. And we'll see how goes. So I'm kick this off.

no, this is the wrong, sorry guys. It needs to be deep research is what I need to do, sorry. Okay, so when you kick this off, it's gonna read through your prompt.

And then it's going to ask you if you agree with their research proposal.

works, what's going on here?

Dalton Anderson (10:41.934)
thinking, it's thinking.

Dalton Anderson (10:49.806)
Okay, here we go. Took long enough. Okay, so it says, here's the research plan we put together. Let me know if we need to make any changes before we start researching. So it laid out this whole plan for us, which is quite long, right? I mean, if you're looking at the screen, it's extensive. So then it's saying that the time to complete all this is gonna be a few minutes. So I'm gonna kick this off.

This was the more interesting one. So I want to kick this one off first and then we could try the other ones. I have one related to EV. I've got one related to yo-yoing. So just different things that we could read through together. So I'm going to start this research. So it's going to start researching.

And so you could leave the chat, but it's gonna show what it's looking at. So now it's looking at 31 websites and it shows you all the websites it's looking through.

which I think is pretty cool.

Dalton Anderson (11:57.07)
It's at 45, and it's gonna keep going, and there's a little progress bar. It's at 61 websites it's researching. After that, it will notify you when it's done, or you can just come back. I think it sends you an email. Down with it, 71 websites. So then now we're going to kick off another research. So let's do yo-yoing.

let's do EVs actually.

So this is EVs. I'm thinking with EVs, I wanted to know, what are the pros and the cons of mass EV adoption, the environmental impacts, some of the stuff that people already know and that is quite common. It's not necessarily like a difficult topic, which is easier to know if it did okay or not.

So these are just low ball.

Dalton Anderson (13:13.102)
All right, so we put on the model, we're kicking that off. Let's see what it proposes. Let's check on another model. And so we're cooking this all up and then I'll start the episode, but I want to make sure I get these things cooking. So when dinner's ready, it could be served, you know?

Dalton Anderson (13:31.874)
and I and then I'll do the yo-yo-yoing. So this one's pretty short.

That's fine, let's kick this off. I wonder if you could do two at a time. I've never done that before, so let's try it. Start research, starting research. All right. This is at 119 websites, quite extensive. This is at eight. That's yo-yoing and just the history and evolution of competitive yo-yoing. And then I've got one related to

dinosaurs.

Dalton Anderson (14:15.31)
Wait, did I? Sorry, yeah. That was actually not yo-yoing. That was EVs. I don't know. I got confused. I already had yo-yoing copied and pasted. And it was ready. I was ready to just copy and paste yo-yoing in. So I thought that's what I copied and pasted into the the chat. OK, so for the right model, and now let's transition and.

Copy and yo-yoing. So we'll have three ongoing research projects with our AI analyst. This is getting there. This is quarter. This is going pretty quickly. It's 35 websites thinking about EVs. And this is about yo-yoing. Let's see here. Start research. So we've got three research projects going on at the same time.

which I think is pretty cool.

That's pretty cool.

Dalton Anderson (15:23.662)
So we'll come back to that in a second, I'm pretty sure.

and then we will look at the results together, right?

So we got those kicked off. Let's do a brief overview of what's going on there. with this, there is, I'm gonna stop sharing so it doesn't mess up the videos that I.

Dalton Anderson (15:51.08)
With this is a.

one million token context window. And there is some confusion about that. the context window includes the prompt that you send, the prompt that it creates, and then the information that it has to process to create the prompt. So it's not like it could process exactly a million tokens and or output a million tokens. It's not how it works, but. And more token, like.

your context window doesn't necessarily mean that you have additional context on answering a good prompt or understanding what the user is asking for. So you gotta have a context window, but then also the comprehension to understand what is in the context window. So that's quite important. And if you don't have that, then you don't have much. And just wanted to throw out there that

The context window is cool that it is a lot and it's still the most with their ultra model and this 2.0. I think the context window went up. That stuff is great and all, but at the same time, doesn't mean that it's going to be way better than everything else. So just throwing that out there. Why?

while we are waiting for.

Dalton Anderson (17:28.888)
our research to be completed.

Dalton Anderson (17:37.56)
So you have a one million context window. You have the ability to kick off these deep researches. And I just demoed the starting of the AI analysts doing their research. Doesn't take too long, maybe five minutes or so. It says a couple minutes on that original. It's still not done yet, so. They're liars. I don't know. It definitely takes more than two minutes for sure, or a couple minutes, which.

It was two minutes, but I would say under 10, and that's fine. If it could save you hours for 10 minutes of waiting, it is what it is. Like it's not a big deal. I would love if it, if you could hook up the AI researcher to your code, like, okay, I need to build out this feature. Here's the code base. Do your thing.

and then make a commit to our repository and our view it. Like what if you could have 10 of these AI analysts just doing stuff for you all the time and they were.

semi like a semi agent, like not as like full independent agent, but they're just doing certain tasks. That would be cool. And I think that this AI researcher thing is one step closer to having a legitimate agent that acts on your behalf and saves you time. Right. So they have this custom gems, I think they call them. One thing that Google is not good at is the naming. So I've got Gemini.

gems, some other thing. then they had, before they changed the names to Flash and all these things, everything was with a G. They had Gema, Jimmy, I don't know. There is so many gems. So everything had a G in it and it was very confusing and ended with an A. or yeah, Gema and didn't end with an A, but the other stuff did.

Dalton Anderson (19:56.246)
Jim and I has the flash, the advanced, the ultra and this other thing called custom gems and custom gems is a way that you can have reoccurring tasks be performed for you. I use it to generate my podcast outlines and so I'll write up my notes. It'll be on organized, it'll be a mess.

and now throw it in there and AI does this thing. It makes a beautiful outline every time. Every time. I've never had an outline that I was like, no, that's not the one. Every time it's great. And.

What is cool about that is it saves me a lot of time, but also I think it's a fundamental difference between patients when you're training a human. Think about you're training a colleague at work, new recruit, they're coming in and you've got to onboard them. How much time do you spend with them to train them on tasks? Onboarding, HR spends a week with them. Then you spend a couple of weeks

and then you have another colleague train them. And so you're investing all this time and this employee to execute tasks independently, but people are so impatient when it comes to training these AI.

chat bots or agents or whatever is to come in the future, the level of patience that is afforded to the AI is so minimal. And I would think about it is what if for a human you had to invest 10 hours to train on a task? I think you should at minimum

Dalton Anderson (22:06.606)
probably invest 10 % of what you would do for human. Because AI can remember your stuff and knows a lot more than a human.

Dalton Anderson (22:18.626)
these other factors that help AI learn a little bit faster than a human does.

Dalton Anderson (22:27.138)
So when I made that outline, I took some time on it.

And I'm only pointing these things out because...

If this deep research thing becomes something where you could have an agent that's for coding and this agent, you kick it off and it starts doing coding all day for you and then you have a compliance researcher that does compliance stuff. People have to invest the right amount of time into training these agents and giving them the right prompts. And then there's a whole thing regarding prompt engineering. Even the researcher, the AI researcher,

reiterates your prompt that you provide and engineers it into a computerized version of what you're saying and makes it more detailed and format it the way it's most optimal for the model to compute.

Dalton Anderson (23:32.502)
So there's one, training your agent to be, two, writing the right prompt. Both are essential. And so I was just pointing that out there because people get very impatient with these AI things because one, they're like, I saw this YouTube video and it could do XYZ and then you just throw in a video and ask it to do something and it doesn't do it. And you're like, this thing sucks.

That's not how it works. You gotta train it, you gotta give it some love and affection and then before you know it, it's all grown up and ready to do its own thing. So if you take that same approach when you're training your chats or your agents, you could get good results but you just gotta put in the time. Like I spent a couple hours on the outline but it would take me a couple hours every week to make sure the outline was perfect and

the way that I need it. And when I say perfect, I say that loosely. It needs to be good enough where the outline is familiar and I can just glance over. And a lot of times I'm just looking at the camera to you, but sometimes, hey, I want to see where I'm at and I just do a quick glance and that's all I need and I'm good to go.

because I already know the format. I've been doing that same format for a year or so. So I know what to look for right away and I could just jot my memory and I could recall what I wanted to talk about. So just want to point out that you've got to be patient. They don't know what they don't know, but they could learn if you tell them what they need to know. Okay.

Some of the key benefits of this AI agent is it's got up-to-date information, it travels the internet, researches, and then it creates a detailed report. I've had times where it creates a table for me. It creates a table, inserts it into the Word doc, or it's Google doc, and it's great. And then has all its references at the bottom.

Dalton Anderson (25:49.036)
I don't see any issue with it. And I think it will save time if you want to research like maybe edge topics or niche things, or you're just generally curious or it save business time. When they scale this up and it becomes an enterprise product, it's going to be cool. It's going to save people a lot of time. If you could create a compliance analyst that researches law all the time, and that's all it does is like, okay, now I'm going to go research and

go on to every state website and look through their meeting notes and draw up analysis on region stuff. the stuff in the East, so all the states in the East, Southeast and Northeast, these are...

the current meetings, the things that are in the house, and here's what the meetings are saying, I think this might pass, let's prepare, or this is gonna get shot down, look at the transcript. Those things would save people a lot of time and money, because they wouldn't have to go through the minutiae and or pay companies to aggregate that information.

And there's a lot of companies that make all their money really off of just data aggregation of like government data that's publicly available. That's all they do. They just hire people to aggregate data and they license it. That's it. Like the data's there, it's freely available. If you had an AI agent, the AI agent can do it for you when it becomes advanced enough that you don't have to worry about those things. So I think

I think that'd be pretty cool. I'm excited about that. I could enjoy that as a potential outcome. Wouldn't mind it too much. So let me go back to sharing my screen as the demo is completed. Let's see, share, window.

Dalton Anderson (28:10.798)
Okay, so you should be able to see. Okay, so it created this e-commerce, current trends in emerging sectors. I'm gonna read the headers. Current e-commerce trends, mobile commerce, and then it made a table for us. Mobile wallets and contactless payments, augmented reality and virtual reality, personalized recommendations and AI-driven chatbots, adapting.

to mobile commerce, social commerce, successful social commerce strategies, personalization, how businesses use personalization, technologies enabling personalization, sustainability, how retailers respond to sustainability demands, AR, VR, and e-commerce. I'm not gonna read all the headers, this is crazy long.

I mean, this is extensive. Research 119 websites to find out what I was looking for. This would have taken me forever. I how long would it take me to read through 119 websites? And I don't have time in this one episode to read through an extensive report like this. But if we just skim through, if I just scroll and just do a random scroll and then just read a section, not out loud, but you know what I mean?

Dalton Anderson (29:41.07)
emerging market technology or emerging technologies in product categories.

Dalton Anderson (30:01.038)
That's pretty cool. So it's talking about 3D printing, like making personalized products, like on the fly, like whatever you request. And so I could see that as maybe you could 3D print a wedding photo into a table. Like say you make the same table every time and then you can have a wedding photo or something related to your dog, something like that. Biotechnology, they're saying,

that you could have personalized supplements and customized skin care routines per person, which I think would be pretty cool.

Ahem.

There's a shift in customer behavior.

growing interests and experiences over material possessions. I agree with that. I've seen that being trending. It's been trending for years now. And I talked about this a while ago with, I gave an example, Diplo and some other folks and different events, run clubs, all sorts of stuff. And I talked about how there's such a,

Dalton Anderson (31:17.16)
lack of connection in the world that people are wanting and yearning for real world experiences versus virtual experiences like social media because everyone is connected quote unquote but we're not really connected. We don't really know these people. We don't really care. A lot of times especially if you've had social media your whole life like if you grew up on Instagram or TikTok or whatever generation you are if you had an Instagram like

when you were in elementary school. Like you're following all sorts of people that you don't even care about when you grow up and you're seeing their stuff. Yeah. And I unfold crazy amount of people. So now I don't have that problem, but for the most time people just would either create a new Instagram or they just leave it as is.

That being said.

Dalton Anderson (32:14.114)
people are gravitoring towards experiences versus possessions. So this looks pretty good. I don't have time to read it, but I will export it to my doc because I want to read it later. So this is something I was generally interested, like this is what I wanted. And then I could probably build on top of this and add, okay, now that we have this, let's dive a little deeper.

what are some products, potential products that we could look for. So it's created the doc for me. Let's go over here.

This is the EV. This one made a table of contents for us. It's nice. Introduction, battery technology advancements, charging infrastructure challenges, environmental impact, economic impact, leading countries and policies, consumer trends.

Dalton Anderson (33:22.51)
I really like these.

So pretty good.

Dalton Anderson (33:31.8)
Don't think that it's that far off.

Dalton Anderson (33:40.166)
I who the leading countries are. Several leading countries, there is Norway, Iceland, Sweden, Netherlands, and China. That makes sense. I wish it went and broke down their policies though. Let's see if we wanted to see that. Tax incentives, if we went to tax incentives, you could drop this down. It has the link of where it's getting this information from. Open this up.

Let's see.

These countries are adopting electric vehicles the fastest. So this is one year old.

Dalton Anderson (34:22.574)
and

Dalton Anderson (34:28.962)
Norway, here's a graph.

All right, pretty neat. So then the last one, let's look at Yo-Yo, which is a little bit less complicated.

Dalton Anderson (34:45.216)
So Yo-Yo at one point I thought was some kind of self-defense weapon and then eventually became playful. I don't know. Maybe that was just fake news. Maybe I got fake news a long time ago.

Dalton Anderson (35:05.26)
Yo-yo's are made out of wood and-

Dalton Anderson (35:13.966)
paying homage to the gods.

Dalton Anderson (35:24.738)
I don't see anything about using a yo-yo as defense, but I don't know where I got that from, but at one point I thought that it was. So maybe I did in fact get fake news and we're finding out live here.

Dalton Anderson (35:44.888)
Yo-yo style and tricks. It's got a list of tricks. It made a table for us for the list of tricks. Yo-yo manufacturers, some key famous yo-yo players and what they did.

Dalton Anderson (36:05.6)
I don't know. I don't know anybody who's done yo-yo.

Dalton Anderson (36:11.822)
Yo-yo.

Dalton Anderson (36:31.438)
fence.

Dalton Anderson (36:35.278)
There is no historical evidence that yo-yos were ever used as weapons, but the idea was used as a marking tactic by the Duncan Yo-Yo Company.

Dalton Anderson (36:50.648)
Dukan, yo-yo company. However, there is some evidence that yo-yos were used as weapons in the Philippines. Yeah, so I did get fake news, but it wasn't necessarily super fake news, but they got me. That's so funny. So they marketed as weapons, but weren't really weapons, even though there was no history that they were weapons, but they still marketed as weapons, just till they were a little bit cooler.

Interesting, interesting, Hmm. Not sure how we got there, but we're there. So that was that was the results of our little research projects that we kicked off. I think the topic of the product research and what are the emerging trends and and what kind of things I should look out for was very extensive.

That was probably the most complicated prompt I've ever requested Jim and I to provide. When I request to the deep researcher, this AI analyst, that's definitely the most complicated for sure. And it did a good job, I feel. I didn't read it extensively, but from the skimming that I did do, it seemed...

not too far off of what I would expect, like some of the stuff it was saying seemed legitimate. And then if you don't like what it's saying, you just go to the link that it's linked to. So if you're on the UI for the Gemini after the research prompt, it will have all of the references right below it. So you just click on it and then it pops out the link where it got that information. And so it's basically just aggregating information for free. I mean,

For what it is, it's pretty much free. 20 bucks a month, if you could run that. If you had a use case where you ran that once a week, that should pay for it. If you could run that once a week and you got something useful once a week, and it saved you two hours once a week, yeah, it pays for itself.

Dalton Anderson (39:13.932)
I already talked about what I used it for before. We went over this demo and found that some of the results were pretty good. And I also got fake news ish. I'll change it from fake news to fake news ish. And I talked about the future of what this could look like, where you could have your own AI analyst and how that would work. And I also talked about nurturing the AI analysts.

to make sure that, you're in the right spot here. You can't just throw it to the wolves. You've gotta train the analyst to do his thing. And if you put in the time, it should give you the right result that you're looking for.

Dalton Anderson (40:02.782)
After that, we kicked off some of our own research projects. We looked at them. They're pretty decent. In a general sense, I'm pretty excited about this product. This product is interesting. I don't know if there's something like it in comparison on the market right now. I know other companies are gonna be coming out with something similar soon, but this...

right here is two things why I think it's cool. One, it shows you the websites it goes to and it links all your information and the output of the report. Two, Google announced this in the summer and they said this is gonna happen and releasing it later this year. And it came close to the end of the year.

looked like there was some late nights involved to get this shipped out to the public.

But that is what Google has done. They've said that they're going to do it. They announced it. They did it with these other AI companies. They make announcements. They don't do live demos. They do pre-recorded live demos, and then they do a live session of the demo to make it seem like it's live, but it's not. Google does live demos. Meta does live demos. These other folks like Anthropic and OpenAI, they're

aren't live demos, they are pre-recorded. And OpenAI said they are going to ship some stuff when they never did. And then I've lost track on the stuff that they said they're going to ship and they just never shipped it. in my mind, I don't trust the things that they're saying and I don't trust the demos that they're providing because everything is pre-recorded and it's very polished. And if you aren't

Dalton Anderson (42:06.616)
confident enough to live demo it, then is it ready? Like, do you really believe in what you're building if you can't live demo? That's a big part of announcing your product to the public. If you're announcing your product that the public is going to be using, why would you not live demo? If something goes wrong, it's live demo. It's fine. Like whatever. I people get upset.

It is what it is. But if you're blatantly not confident enough to live demo consistently, then what's what's really going on here? But I digress. I hope that you found this episode interesting. And useful. And if you have used these AI analysts before, let me know what kind of things that you've done with them. And if you haven't.

Let me know what you want to do and what you want the AI analysts to be able to do. I would love it to just kick off in the morning and I have some set agents and they do their AI analyst thing and then they come back to me with reports and things that I need to approve and where I can delegate almost a virtual team of these AI analysts to do its own thing and come back to me with the results and save me time. That's what I would like.

Not there yet, but baby steps. But of course, wherever you are in this world, good morning, good afternoon, good evening. Have a great day. I hope that you enjoyed listening to this episode and hope that you listen in next week. See ya. Goodbye.