Venture Step

Dalton 00:02
Welcome to ventureStep podcasts where we're still figure out what we do at the show. But this week, i wanted to talk about some releases from google with their gemini product, 1.5 on gemini advance. And then open AI with their sore product. And then i also had a research paper that i read about the different approaches between startups and incumbent companies, injured assurance industry and their different approaches towards using AI to optimize their Process and customer acquisition. So, make sure this You have a simple ask that. You need. Say a logo or Just a piece of art that you would do in adobe illustrator. That maybe you don't have the time to do. And you outsource the fiber. One of those type of websites. For 20 bucks, 15 dollars or something like that and it might take you a couple days to vet the people that you want to use or you could just throw some dollars out a couple of different. Creators and then figure out which one you like. And move forward with that. But now, And, In the distance future. These things would be handled by AI. You would be able to just Type in some texts. That you want for an idea

Dalton 01:29
And

Dalton 01:30
You might not like the original that is sent back to you, but it's fine because you can iterate on it. This is huge because it cuts out the middleman. Like you, do you no longer have to worry about Having to go back and forth and waiting for these different time differences or Potentially getting something you don't want, you can just do that for like 20 bucks a month now. And you could do a lot more than just those things, but I think that is a big deal. Especially now that is becoming. More of a. Dynamic video offering with Open AI sora. But first, let's talk about google gemini. So, google originally launched with barred and then They recently switched over and kind of rebranded. Their AI offering. With this new. Gemini, gemini. So They have gemini 1.5 which was recently announced like three days ago. And what's really important about gem? And i 1.5 pro. Is that it can handle. They say they can handle 1.5 million tokens. And so let me explain what a token is. So when you're doing nlp and working with unstructured data, An interview is natural language processing by the way. I'm not gonna get too technical. I'm just gonna briefly. Explain what a token is. If you think about it as like, A book would be the full puzzle and To take all the pieces and put them back into. A full picture of the puzzle. You would need to break up. The sentence is, or The paragraphs or the pages of the book into smaller bite sizes, which would be a token. So basically, You have a sentence like, I took my dog jerry to the park. When you tokenize that, it makes each of those words. Individual in the sentence but you can tokenize by sentences. You can tokenize by words, you could tokenize by phrases. So that's why i said i'm just going to do a brief economic explanation. Because it gets fairly dark and go pretty quick, and there's different use cases for different types of tokenization. But, Long story short. You're breaking up complex things into smaller things so it's easier for the model to understand. Very simple explanation.

Dalton 04:16
So,

Dalton 04:17
Why is that a big deal? Because the average book is around. Depending how you tokenize it, but 300 thousand tokens. And gemini is 1.5 pro is 1.5 million tokens, which is Huge, and from Early access. The AI researchers are estimating from their tests. 99.7 percent accuracy. On a million tokens. Which is pretty crazy. If you ask me, i think it's insane from the stuff that they have demoed. Um, i am gonna go over some examples that Ruin chung, the founder of the Fastest growing AI newsletter. Let me see. Is it called rundown AI? Yeah. So Roland Seung was given early access, which you and i pro, and he has a couple examples that i'm just kind of read off of my notes. And just give a brief operation. So he said that, he uploaded an nba dunk contest and he requested Gemini to look through the video. Would you like? I think it was like an hour long. Look through the video and Basically saying like, who God, dive score. On their

Dalton 05:38
Dunk. Which is

Dalton 05:39
Pretty crazy because it, it takes a It takes the the model needs to have. Long concept. Understanding. And then it also needs to understand the video. And, Understand. Like the score that i was given by the dunk. And then also identify Ooh, that person was Where it was in the video and explained why it was the highest scoring dunk. This is, there's a lot going on there. That makes it. A difficult problem. Um, another one was The he had an example where he translated English into. A language called satellanic. And i had to look it up, but it is. Place in germany. I guess people speak this language and it's less than 2000 people. Speak the language. And, Gemini 1.5 was able to. Do a full into a sick. Translation, which is pretty cool. And then, They also had one where he uploaded an open AI sora video to Gemini. And, Gemini was able to not able to, but Basically highlighted things that were concerning about the video on whether or not it was Ai generator not. So, it was Able to identify issues potential issues with the video. That might give away that, it was AI generated, so, would jim and i 1.5 pro. You can upload one. Our videos 11 hours, where the audio. Over 30 thousand lines of code. And then, Over 700 thousand words, which is crazy because Chat. GBT turbo allows 128 thousand. Tokens. And then gemini, bro. 1.0 was 32 thousand. So, There's a huge jump. Crazy job. And the amount of tokens, it's a big announcement. It's probably the biggest announcement for These large language models. This year with the token increase. I guess the heroes you started, but You know what i mean? In the recent six months, i guess. So, With gemini 1.5 pro, they also announced gemini advanced studio which is like a similar offering to open air eyes. Where you can pay, i think it's 19, you know, it's basically 20 bucks, it's like 1999 a month. And it comes with one terabyte of google cloud storage. And then, gemini advanced. And in a gemini advance, you're able to Upload photos and you're supposed to be a little, like files, i'm pretty sure, but from my experience, i'm not able to upload files. Maybe i have to clear my cash or something like that, but i wasn't able to upload any files. What you're supposed to be able to But you can upload images, and it can understand like the images. And it also has a multimodal. Expert. Where it can? Creates. Text to image. So if you gave Gemini advanced. A Text prompt, it would be able to generate. And illustration or a photo. Depending on what you did. For example. The new. Venture step. Podcast. Cover art was generated by. Gemini advanced. Studio. Which is pretty cool. It allowed me to upgrade my Podcast, cover art. And a low effort manner, i still had to go through many iterations to get the right color. Hue in the right design. Because it, it seems to go either really Off of what your prompt is or way too close, to exactly what it is. And if you try to get that that sweet spot, it was, it was kind of difficult. And then, once you get there,

Dalton 10:23
You

Dalton 10:23
You get stuck into this like one more. Just one more. Let's let's see if we can get one more. This is so close. And then You wind up messing around with it for like 30 minutes. Um, but i think i think it's pretty good quality. For just a text to. A text prompt generation of the image. And then, for some reason, these Kind of models have issues when you're trying to generate an image would be a text prompt. With spelling, like if you said, I want to make a sign that says, Beware of. The aliens. For some reason. The. These models have issues generating the text onto. The. Images. And a lot of times it would generate, What you're asking for, but it'd be misspelled. Or the numbers are not the numbers. Sorry, the texts are like jambled into each other. Or that's missing or maybe half the word is there or half the phrase or what you're looking for. So i found it easier to generate what you want and ask you to put a place where you would able to Transpose, not really transposed but You're able to input the text via Photoshop or canva, or something like that. And just tell it like, hey, i want to put my text on the This area. Can you write a Can you can you write in a text blocks kind of shape to the design? So, i can put in my logo name, or Whatever you want. So we're going to move over to open AI. Sora. Opening. Eyesora is And also, the Gemini. 1.5 are both in kind of Not even beta i think but just limited release to A researchers or People who have? A creator platform. That have a. Attack falling, i guess. So there isn't much information on. On a personal in from myself. But for people on the internet that are AI researchers that are trusted. That's where i'm getting this information from. And, For open AI. They have. These videos that i'll Try to share my screen. Let's see how this works. Let's see. Share to this share.

Dalton 13:17
Sure, this Awesome.

Dalton 13:21
Okay. All right, so i'm sure my screen. You should be able to see this. This is my first time so sorry for those are aren't listening. How this looks, okay. So if you aren't viewing the video, i'll explain what's going on. So this is a text prompt about a close-up of a gray, haired man with a beard in a 60s. He's in deep thought, pondering the history of the universe. As he sits at a cafe in Paris size, focus on the people off screen as they walk. As he sits mostly motionless in dressing a wool. Cotton. Suit coat. And, It's a crazy. Realistic. Image. I mean it's really difficult to tell that this is AI generated. It's got he's got reflections. The light that hits his, his reading glasses, it reflects it. The window to his left on his glasses. I mean is this really difficult? I think the only thing that kind of looks a little weird is his ear. But other than that, i wouldn't be able to tell. I mean, if you sent this to me, Like hey this is my grandpa or something. He's so cool. I would have no idea. This is AI generated. The next one is this. The prompt says in adorable. Happy otter confidently stands on a surfboard wearing a yellow life jacket riding along Turquoise tropical waters near lush tropical islands, 3d generated, art style. Um, 16 by nine style ra And it's this otter on the surfboard as a scribe and he's like super happy. Waving his hands smiling. And it looks like something that you would see in a Disney movie. Like surfs up, maybe like a surf sub thing, but not And, You know. The arctic is, it's in a tropical island. This one's really cool. I like this one a lot. So it's a historical footage of the california. Kind of california during the gold rush 16 by 9s style raw. And it to me it took these photos, it uploaded photos of what the california gold rush looked like from historical images and then took those photos and then created a Colored video. Of, you know. Where people stayed and How they travel along and see people traveling along this. River bank with horses and people walking around This one, it's probably my favorite. This is a movie trailer. Featuring the adventures of a 30 old. Space man wearing a red wool, knitted motorcycle, helmet, blue sky. Salt. Desert. Ceramic. Oh my gosh, ceramic. Cinematic style shot on 35. Mm. Film vivid colors. Seen by nine style raw. And this is, this is like a, it looks to me. Like a high quality movie trailer. I would see something like this and i'm really into sci-fi stuff. So i would find this really interesting. And it's just the way that it looks. It's different. It's unique. The guy, when he's stepping into this salt desert, you can see his footsteps. You can see the details of the of the red hat. And he's got this other guy. I don't know if they look like different people but there's a there's another guy with a red wool hat on. One's got the silver, shiny visor? And the other one doesn't. So maybe that one's a higher rank, i'm not sure but overall, it's it's just crazy. The I, i would have absolutely. No idea. That this is AI generated blowing my mind. This is another one. That would be a lot of work to do yourself with the 3d animation. The prompt is says, three animation of a small round fluffy creature with a big expressive eyes, exploring a, vibrant enchanted force. The creature of the whimsical blend of rabbit and squirrel has soft blue fur, and a bushy, striped tail. It hops along a sparkling spring. Its eyes wide with wonder. And it looks like that. I mean, it looks like something you'd see and Kind of again, a disney a disney or kids a kid's animation show. Um, for sure. I don't. I think it should be a lot of work to do this yourself, but i think that the red wool trailer movie trailer Is. I i just think it's on another level. This one's really cool because you i i can't tell i thought this was a kind of a meme thing. Uh, where, you know, they discover this plastic chair and they know the archaeologist digging up this. Generic plastic chair. And excavating it with great care. And he's just it's just funny like they're they're being very gentle trying to get it out and They've got people around and making sure that Nothing's going on. Which is, it's just funny. This one was cool. I serve this. I showed this stuff to my nana and those were my grandma. And she was like, wow. Look at those birds. You know, they're all, they're all different colors. It's cool that it generated that. And, I was like no nana those are those are paper planes. And instead of prompt says a flock, Flutters through a dense jungle. Weaving around trees as if they were migrating birds. 16 by 9 style raw. And, Like, birds that are shaped like, Airplanes. I mean, it's really cool because they're all different colors and they kind of look like butterflies slash birds slash. Paper airplanes, but they're definitely resemble paper planes more than anything else, but they're just cool. This one's a little creepy you could, you could tell it, i think you can tell this is a generated with the hands. If you look at the bottom right it seems like she gets different hands every time on the video. But this one, this one was kind of odd because it's like A grandmother like it says a grandmother with neatly combed gray hair stand behind a colorful. Birthday cake with numerous candles at a wood, dining room. Table expression is one of pure joy and happiness Where they glow in her eye. She leans forward and blows out the candles with a gentle puff. She doesn't blew it on the candles and she makes he's like crazy facial expressions. And there's these people in the background like cheering around like that. Like they're at a rave. And it just it just it's just an odd video. There is the next one. Was this camera? Video of maybe. Like a stack of, let's see. One, two, three, four, five, that's five. And these other ones are This is five again, no. Four, five, six. So it's like a stack of like, six to four. Antique TVs, depending on the size of the tvs. And they're all playing this different kind of Old school shows and black and white somewhere in color and they're all random and the TVs are changing. This would be expensive to set up in real life. And, You could just do this on the fly. With the tax prompt was is wild. This one's crazy as well. The beautiful snowy, Tokyo city in Boston is bustling the camera moves through blessing city street. Following several people enjoying the beautiful, snowy weather and shopping. Nearby stalls. So, Just kind of the perspective is kind of weaves through, like this would be for A drone video. And people might pay a thousand dollars or, you know, 500 minimum to have a drone video of themselves or their couple walking through tokyo on their vacation. I don't think it's at the point where you would be able to, But your face in the videos, but Just the potential is. Is the pendules there for sure. This could be a marketing ad where You know, at the end of the video, it cuts to this this couple after they get the product or you get that ring, that she always wanted in. Then you guys walk off. All happy through Tokyo. I'm just spitball and ideas here. The next one is the white vintage suv. Kind of cutting through the mountainside of this dirt road and you could see the reflections and The rate tracing or ray tracing? Of the, it looks like a video game. Like, it looks like a Forza. Before the video game was really good graphics. But you can see how the different shadows hit the car, you can see the reflections and then The different. I mean it goes up through different kind of Types of soil a little bit and you can see when the cars tires kick up the dust. The dust color, slightly changes depending on the light. And What kind of swear runs over? It's very subtle changes. This is kind of cool. It's it's a cloud that It's like a man. I don't know, it's an early man because it's a cloud. But Um, it's like a This is a cloud in the shape of a man lives over earth. The cloud, man. Shoots lightning bolts down to their This one was very unique photoistic close-up of two ships battling each other. As a sale inside of a coffee cup. And it's like waves rocking the coffee cup. And There's no cannon shooting, unfortunately, in smoke, but That's definitely really cool with all the water waves, not waterways, the coffee waves. Um, aerial view. Of the Of a place in greece. And This would be something that you'd pay. Definitely a drone shot for For. Tax dollars and this is something you just generate. On your own. With a textbook. There's a crazy realistic video with. Golden retriever puppies playing in the snow. And their heads are weaving around the snow and snow skinny flipped up and That snow's hitting other puppies in the head and They're biting at it, and it's a I mean, it looks realistic to me, that's what i'll probably would probably do. That was it? Okay, so i'm gonna stop sharing. All right. So, those are the some examples from OpenAIs new? Sora product. And once again, it's still in a limited release. But then, i would also like to demonstrate where we were 10 months ago, because 10 months ago, there was this video. If it will come up. There was this video from I don't know where it's from, but It's definitely an odd video. And it's basically, Will smith. Eating spaghetti, with these crazy audios on the back. Screen. Um, around sharing. Okay, so let's play this video. I'm not going to play the sound because it's horrendous and creepy, but it's Will Smith's face distorted. Just gobbling down. Obscene amounts of spaghetti. And there's like duplicate Will Smith in their gobbling down. So spaghetti like one Will Smith holding another spaghetti. Pick a balls. We got any in this hands and he's eating it in. The other Will Smith comes in there and eats And eats the spaghetti. It's super weird. That was 10 months ago. So, in 10 months, we went from skewed. Creepy Will Smith. You need to get a to these well polished Text to. Video. Product sora. I don't know, the the progress is. Very, very I would say concerning but it's definitely something to be aware of for sure. Because I think that it will have effects on. People who live in the Marketing space. Either like a fiber or This is the marketing department in general, like, You don't need someone to go in, and Create x thing or you don't need you don't need like three people to Right up. These drafts. You could just Ask AI to write up three different drafts of these different styles. Get back to you present the different drafts, then it gets approved. And move forward from there. So i think that the level of resources that are required to do the same thing or less. Anything people will be completely replaced because That's going to be difficult and we're not there yet and I don't think that many companies would just be okay with just completely removing the human aspect of Important parts of their businesses, but i do think that If? Warranted, they would limit the Amount of capital output to. Salaries. That's a much cheaper to have. A enterprise subscription to an AI studio then. It is to pay 20 people. That was opening as sora. And, We have one more thing. We have this AI AI paper. Called. What's the name? What are my favorite name? Go. All right, i'll get back to you on the paper name but It's, you know how everything excited in the show notes and in the description and i'll definitely cite the, the authors of the paper But i wanted to go over. Real quickly just a different section instead of talked about. And i think that they did a good job of Condensing the information into like a management perspective. On just a different approaches of incumbents slash startups, so, It's, this is for my notes from the what? I Wrote down from the paper. So in the AI revolution, AI revolution and insurance dual perspective section. I emphasized in my notes that incumbents Are working on existing workflows. And enhancing customer service for new insurance. They're focused on shifts to using AI to redefine customer experience and offering personalized products, leveraging data and innovative ways to assess risking policy price policies. I think this makes sense, because For older insurance companies. Like a state farm or And progressive, isn't that old? And they're on the forefront of The forefront of the data for like an incumbent, but You know, a state, a state farm for sure. They have a very old policy administrative system that they're working on. That, you know, they're they're going to move try to move or migrate to a new system, or build build one in house and they have to deal with all these old. Data constraints for their historical data. Because it might not necessarily. Being the same format as, There are new systems. So at one of my previous companies, And like 15 years or something like that. And Through multiple acquisitions. We were on five different policy systems and they all had different formats and different constraints. Different. Data columns. Nothing was normalized. And it was very difficult to extract data. From every single one of them and then map the data into a unified. Data set to use for analytics, and it was definitely limiting for sure. And just the time that it would take to do so. Many times we would just Not go that route. Not to say that for actuarial actuarial. The way, how do i say like actual or process is that you, oh, it will do. And that's rolled apartment. We definitely have all the all the data that we needed for those processes. But as far as just getting a quick glimpse or maybe a power bi dashboard or those such things, Descriptive statistics. We weren't going to spend. Four months mapping in all the data, it's not going to happen. So that that makes sense. Okay, so the next section is strategy is an AI adoption and come to this versus new entrance. And so, mention a little bit where in comments, they finally, they've really focused on augmenting. Their, their vast data for improved risk management and operational efficiency, And then for for new insurance, they prioritize AI for You know, this customer service, you know customer first product You know where customer service is first, and they offer additional services and often. They target underserved markets and unique ways using AI. And they would to do that. They offer these like radically different Pricing models that may or may not work out. Sometimes the loss ratios for These companies are Fairly poor. So, the next section is the Comparative exploration. So, The two types of companies have different approaches and kind of emphasizing what i said and the two other sections. New entrance. They have kind of a new twist on. Our new tag on insurance and how they're going to offer it and you might be direct to consumer and And, They try to automate claims processes and have a better claims experience, and they try to use AI to take care of, you know, these these customer service journeys or just customer policy acquisition journeys. And, Allow them to be a bit more lean on. These kind of processes that are I wouldn't say time consuming, i guess there are, they are time consuming, but it's like a necessity for insurance. So it's like when you have an insurance policy slash policies, like you're going to have claims. It's like Having a E-commerce site, and then being like, oh, wow, we're having returns. If you're gonna have returns if you're selling stuff online. Let's just, there's just matter of fact. And then for in comments, it's it's more about cautiously integrating. These These changes. It seems like, you know, for the most part they want to use Robotic process automation, or Just different automations that would allow them to transfer old data. To New data platforms or, you know, data, migration or Just a little bit of more of a Cautious approach, and less front facing to customer and Obviously startups are more innovative and have less to lose. So there are front foremost. Up in the customer's face with all their AI learning from the contrast. So, For incumbents. You know, this is for case, studies, for comments, we had case studies of Of companies using AI to enhance fraud detection systems or improved customer relation through personalized offerings for new entrance. They highlight examples of levering AI for innovative risk assessment and models, or utilizing internet of things for dynamic pricing strategies. I think that in, you know, in the future AI. And this is this whole conversation about AI. Is going to be. More. Emphasized on any tech stack and on a day-to-day workflow for employees. I think right now that people think oh, if you're using AI, You might you might not be working or you're kind of treating or something like that but if People come around to using it and it keeps improving, at the rate it does. I think it's going to be integral to everyone's day-to-day lives. And especially, For. Companies that want to stay. Ahead and don't want their stock price to plummet. Today, we went over. I really got to work on the closer that was, that was bad, but we're just gonna keep it rolling, you know, we're all learning together. So, Today, we went over. I do it over a gemini 1.5 jim and i advanced Opening. I sawa and disrupting. Insurance. And i will, before i go i want to Shout out the Paper. So, we will do that right now. Papers. So it's artificial intelligence and business in a business, innovation, and insurance, and comparison of incumbent firms versus new entrants. That's probably why i didn't remember it because it's like, One, two, three, four, five, six, seven, eight, nine. You know, like 10 plus words, you know, come on. My memory, is that of a fly? And it was published by. Hawaii university. And it was talked about in Hawaii's international conference of system sciences and 2003, that was it for today, folks. I really appreciate you listening in today's podcast. I hope you found it interesting. I'm still learning what you guys like, and i need to find a way to Capture your attention and Offer value to you as well as. For myself. Have a way where i can do that sustainably. We're still figuring out our relationship together, and i hope to see you next week. See.