Demoing the most innovative new ChatGPT features with Sunny Madra and Vinny Lingham | E1732

0:00Hey, everybody, welcome back to this week in startups. We're doing another AI roundtable and this is the best one ever. Vinny and Sonny join me again to demo chat GPT's new code interpreter. This was just released on Friday. We're playing with it over the weekend, and we're going to play with it here on the show. We take a random couple of CSVs that we grabbed off government websites, we upload it to ChatGPT, and it takes this and acts like a data scientist, and it starts doing analysis of these documents. It's incredible magic. Make sure you listen to this episode with your teams, because at your startup, you're probably wasting tens of thousands of dollars that this new tool is going to remove from your expenses.

0:39These rapid innovations AI are going to change the world. I've been talking about it multiple times per week here on This Week in Startups and on the All In podcast. I think people are going to become 30 % more efficient this year. But, but Sonny thinks I'm wrong. He thinks it's 300 % or more. We get into it. I show you a bunch of details of some GPT stuff I did over the weekend and some stuff I'm doing in Python on Replit. It's going to be a great show. It might even blow your mind. Stick with us. This Week in Startups is brought to you by OpenPhone brings your team's business calls, texts, and contacts into one delightful app that works anywhere.

1:20Get 20 % off your first six months at openphone.com slash twist. Coda is the all-in-one doc for teams. If you've got a stack of niche workflow tools, or if you're buried in docs and spreadsheets, Coda is the doc that brings it all together. Get a$1 ,000 startup credit by signing up at kota.io slash twist. And release. Large enterprises pose unique challenges for SaaS startups. Unlock customers with unique needs for private and single-tenant hosting without the toil of DIY with release delivery. Get your first month free at release.com slash twist. Hey, everybody. Welcome to another episode of This Week in Startups.

2:10With me again, Vinny Lingham and Sunny Sandeep Madra. We were doing a crypto roundtable, boys, and AI has taken over all of our lives. Crypto still seems like an important technology, but it does feel like the amount of energy being put into AI startups, language models, is 100x or 1 ,000x what's happening in crypto. So we'll skate to where the puck is going and continue our discussions about AI here. So this is our weekly AI roundtable. If you have ideas for the producers here, producers at thisweekinstartups.com. If you see something interesting, say something, email producers at thisweekinstartups.com.

2:51All right. So let's get right into it. you shared a link with us uh sunny on the group chat that uh some chat gpt users now have access to a code execution or code interpreter plugin uh what is this and why is it important yeah so this is really really uh big and what it what chat gpt has enabled open ai has enabled is the ability for the interface to run code. And what's interesting, and you can now input data via an upload feature. So one of the really cool examples that people are doing this weekend, this was just released on Friday, just to show you the pace, is that you can take a spreadsheet.

3:45That spreadsheet can have data in it. You can upload it. And then you can basically have a chat GPT do some basic data science for you. And so it's really, you know, the process to do that before would have been to, you know, go get a data scientist or write a Python program. And so it does all of this in line in a very similar way to how we saw the plugins work. We're seeing that now for, you know, running code. and that code interpreter if you were to just do a google search right now for if you do a google search for chat gpt and you go into chat gpt on the drop down you see especially if you're paying the default which is 3.5 version of chat gpt gpt4 and then you'll see some other things like gpt 3.5 with browsing which is in alpha gpt4 with browsing that's an alpha and then code interpreter which is marked as alpha and you see this all in the drop down menu and if you happen to have applied to the plugins which i applied to and i've been using and i got my team on you'll see plugins alpha i think paying for chat gpt the 20 bucks a month we'll get it there so is code interpreter available to everybody do you know i think it's only available to those folks that have plugins enabled which means that they've been allowed into this very limited beta or alpha group that are kind of developer centric or people that are you know real um you know publishing stuff to the community to help educate everyone so it's not widely available yet got it and so an example of this might be what and this is stuff you might ask a data scientist to do in google sheets or excel previously or to query an sql database or something exactly that that's normally how someone would deal with it yeah so inside your organization uh vinnie people are like oh we got this google sheet oh we exported our google analytics oh we downloaded some data we got some you know client data we've got we exported something from salesforce or whatever tool we're using now the team has to go find somebody smart who is either in the accounting department the data science department or just happens to be good at hacking this stuff together and this is something that civilians the other 80 % of people who work at a company just don't know how to do.

6:07It would be too hard for them to do. You have that experience, I guess, in your startups as well, Vinny? Yeah, I mean, it's definitely a lot easier to, I mean, the barriers to using data science right now is coming down by the day. It's democratizing data science. I've got a friend who's a data scientist, and I invested in his company, and he's been using data science models for years. And like, it's just, I think it's a game changer for them. I mean, some of the data science companies out there right now, they charge ridiculous amounts of money. I mean, we're talking like millions and millions of dollars to do data science for companies.

6:49And there's some big businesses out there. I think one's Datadog, I think. And there's a couple others. you know and and uh you know open ai and chat gpt is basically you know reduce the ability to do this to you know smes enterprise individuals can do it what i think is interesting though um on a slight deviation here is google has got access to so much company data to the google suite so if you if you like run a startup and you're on google google drive uh you know google docs Google Sheets, everything. That information is incredibly powerful. So now Google just needs to take BARD and say, would you like to activate BARD on your company documents?

7:35And then, you know, create like, you obviously have to figure out the privacy stuff and, you know, rights. But basically, if you have access to documents. That's already been done in an organization, right? Like generally speaking, the organization should have set their permission. So. Well, so just keep this in mind, right? If BARD starts learning across the company, it needs to be able to partition the knowledge and not infer information that only you have access to if i'm the hr department and i've got a bunch of documents that only the hr departments are and then somebody in sales does a query hey how much do we pay our people internally and what's their compensation you don't want that coming up in the results exactly that is an important permission issue yes but but but but if you're the ceo you should you know do you have access to everything do you have access to everything or do you have access to and what about like if jkel's got a private doc uh sheets in there that no one else actually are you allowed to see that or of course i mean the the organization owns it this is like a fallacy that some employees have that i'm on my like corporate account yeah i agree with you yeah if it's personal if it's personal information you shouldn't have it on the company's servers anyway i am amazed by that it should not be if it's company information it should be available to your manager your manager so that's an important issue to flag um but you know just as a fair warning to everybody there who works at a company everything you say on your email is saved for all attorney your documents for slack for all eternity do not expect anything phone calls as well a lot of companies record all calls incoming in i mean some of it's compliance and some of it's just the default when you leave a company you assign the documents to the next person or to the ceo so if you wrote your diary or your journal in your corporate account.

9:16I mean, wake up people, it's 2023. Don't do that because it's going to be indexed and then somebody's gonna be able to pick it up. So poor initiative flag. Stop using your personal phone for your startup in 2023. You have to stop doing this. It's such a common mistake that founders make. Open phone has totally rethought every detail of what a business phone should look like in 2023. And it's so affordable, you have no excuse. They make it super easy to get a business phone number for everybody on your team. It works through a beautiful web app on your phone or your desktop. And I can tell you, it's amazing because our sales team and our ops teams use it daily.

9:53Recently found so much value using OpenPhone for our Angel Summit communications. OpenPhone is the number one rated business phone on G2 for customer satisfaction. And Twist listeners are going to love it. Brian Jagger, he's the co-founder of startup called athlete he tweeted the following i'm literally cash flow positive from listening to this week in startups for listener deals and he explains that he previously got open phone money from this incredible discount that they give to this week in startup founders and he says i'm not paid to say that i don't know jason pure honest feedback and appreciation and you know what i love to hear this because there's so many people who listen to this podcast we're founders and you need to use these tools but hey listen you you might be cash constrained or you might want to put that cash into your product open phone is already affordable at a starting price of only$13 per user per month but twist listeners can get 20 % off any plan for your first six months at open phone.com slash twist and if you have existing numbers with another service no problem open phone will port them over at no extra cost so head to open phone.com twist to start your free trial and get 20 % off do you have an example to show here sonny if people watching at youtube.com this weekend or on spotify or the video feed like let's just open up gpt4 here and i have something i was playing with this weekend that got interesting too that i'll share okay so i'm gonna share here give me a second all right and we're doing this live because we just got the data set from our producer okay so we're inside of uh chat gpt here and we're going to upload this electric vehicle data set and that when you said send a message there's a link on the right there and if you on the left of send a message and that's where you upload from yeah right here there's like a little like a there was a yeah see this little plus icon and normally the uh so you can see the first thing i didn't know that is that only for that is only for the code interpreter got it and so show just so people can see the interface here because we have never done this but could just hit a new chat there and let me just show people the interface and then just describe that for folks um you create a new chat you click new chat in the top left you hit this down arrow key now you can see all the different um items plugins default etc so you got a sportscast is a little bit so people see it and then it gives you a little description of what it is um and how good it is and they get a sort of internal rating of what it does but you picked called out code interpreter interpreter correct all right and then you hit that and now this is about uh i think a 29 meg file and so it's going to take uh you know a few seconds to upload here i see that yeah and so now what it's going to do and none of us have really seen this file yet which is fascinating it is fascinating i'm doing this alongside of you Yeah, so this is the code.

12:49So it's generated this code. This is Python code here, J. Cal, you were asking about this weekend to read that file. And it's still generating and it's understanding. Now it's now you can see here, it's starting to tell us, hey, the data has been rolled into a data frame. And from the first few rows, we can understand that this is the data. So we're going to let this just let this complete. And I'll tell you the next, the next piece, which what Vinny was talking about a second ago was like, you know, where you normally have to go get a data scientist and so to do something like that. this and so and it you know throws some things up here and it says okay so it's done so then my next question is going to be this well let's describe what it showed there it's loaded the data and it says oh it looks like the data contains vin location model year make vehicle type msrp and department of licensing vehicle id some locations utility and some census tracking producer nick gave us was the electric vehicle population data and it figured out what's in there and it's reflecting that back to you in plain english correct it is and it's saying hey i'm ready to do something it's loaded it what i'm showing here is the prompt where it's loaded it into uh like a python library called pandas which is what a lot of data scientists would use to start analyzing data so there was a little carrot there that said show the work so after it uploaded it when it finished work it asked you to do that and fascinating when it did yours for me it did a different response to the same data which is really interesting like chat gp4 told me the data set contains information about electric vehicles with each row representing a specific electric vehicle the columns in the data set are as follows and it did it one through ten it actually gave me a list of them which is really like a totally more helpful response that's very fascinating that and we had two different experiences sort of the nature of llms that can happen but this next question which i'm putting down in the prompt so i'll read to everyone says can you conduct whatever visualizations and descriptive analysis you think would help me understand the data because i have this producer nick sent us this file and so now let's see what it does in this next phase here and so i don't know what it's starting to tell us is we'll look at the following aspects of the data distribution of electric vehicle types um you know uh battery electric vehicles versus plug-in electric vehicles that's dev versus phev top 10 most popular electric vehicle makes and models, distribution of the vehicles by year, geographic summary of the vehicles, and summary statistics of the range and base MSRP.

15:24And that's all, it's doing all of that just based on this question, which was, can you conduct whatever visualizations and descriptive analysis you think would be helpful to understand this data? And so now it's doing the work to basically do those five things for us. And again, you can imagine - What's very interesting about that is that you did a very generic question, which is you asked the ceo question all right thanks for the data yes data scientists in a meeting uh why do i care just get to the point what what did you learn by studying the data and it it's basically just starting with some general ideas here to get you started and you could pick one to double click on yes correct and so it's now doing the work and what you can see here oh again like you know yeah and so what you're seeing remember imagine people are listening sunny so sportscast okay so So it gave us five examples of things to look at the data.

16:18So the first is the distribution. A chart of the distribution. Yeah, exactly. Of a chart that shows us the distribution between battery electric vehicles and plug-in hybrid electric vehicles. And this is a visualization. It would have taken someone a few minutes to, you know, maybe 30 minutes to generate this chart in PowerPoint. And it's been generated for us automatically. And it shows us that the distribution is almost five to one here, right? Maybe four to one in terms of there's way more battery electric vehicles and plug-in electric vehicles, according to the data set that we were given.

16:51Okay. The next chart is we're going to look at the 10 most popular electric vehicle makes. And we see here that Tesla is a clear leader with Nissan at number two, then Chevrolet, then Ford. And we see a visualization of a chart there. next we're going to look at not by make but we're going to look by model and we can see here that the most popular model is the model three then the model y then the leaf and so forth if you look at this chart and then when we look at by year and obviously you know this we're only part way into 2023 we can see that the by year the distribution of electric vehicles has generally been increasing with a little bit of a slowdown in 2019 and 2020 and a pickup back in 21 and a huge jump back in 2022 and we're only you know a quarter a little bit more than a quarter away 2023 would be my interpretation but what's interesting here is now that you start to see some of these things you could actually ask chat gpt why is there a spike uh but you could just do that in another window at chat gpt4 what's your takeaway here vinnie just to bring you in on the conversation i mean i'm i'm going to start using this to analyze my wine collection fantastic you have a csv upload it what do you tell me about that's exactly what i'm gonna do i'm gonna go to pull it right now and see if i can go you know come up with some some strange stats you know recommend other wines for me let's see what it comes up you had do you have plugins go do it we'll show it on the air if you're comfortable what's interesting here also is based on the visualization and summary statistics here are some key insights from the data it actually uh wrote some of these and it said top 10 most popular electric vehicles is 0.3 tesla model 3 is the most popular electric vehicle model followed by nissan leaf etc so you start getting into some really interesting concepts here and for mine i let me um share mine this will be very interesting to do if i may uh or did you have another one you wanted to do um sunny no no no that's what you know i wanted to just show that capability because that's the new feature the unlock is uploading the data set which i know you've been thinking about a little bit jay cow because you have a lot of spreadsheets i know in your business i got can you see my uh screen now okay so i did the same thing i uploaded the same file but what you'll see here is that um if you're seeing it remember i said it gave me just a list of what are the columns so it gave me the list of columns and then i asked a slightly different question what are the three most interesting trends in this data and it said to identify interesting trends in electric field population a we need to analyze various aspects of the data set pretty generic let's explore the following three trends electric vehicle vehicle adoption over time most popular electric vehicles make some models distribution of electric vehicle types like yours and then it gave me a couple charts it did a different design style uh which is weird but electric vehicle adoption over time instead of using a histogram it did a a line chart it did the same thing most popular electric vehicles and then the same thing the distribution and it too gave me uh some highlights here and uh what i could do here is uh an interesting one let's see if this works please give me the same analysis but take out all tesla models and if it gets this right that's like game over right because this is something you might ask and you're like okay we know tesla is running the table on everything but i don't care that the i mean we all know model three out sells everything because it's you know the greatest model y i think it's the greatest car ever made but those two but let's just take out all teslas and see if it does that right so now you're starting to be able to do things with data i mean it's just this is just stunning uh what could be done here um i was over the weekend uh trying to do things here inside of it i'll show well i can't leave the screen it's one of the problems with chat gpt4 i think if you leave the screen it will uh it can stop yeah sometimes yeah i guess they're trying to get people to not do this but all of these little blocking and tackling things will be worked out over time uh like doing multiple queries simultaneously like just for the love of god uh greg and and uh give me a corporate account here let me put all my people into chat gpt4 let all of this data be shared in a common repository i need multiplayer mode for chat gpt4 and i would pay 200 a person per month i would pay four thousand dollars a month fifty thousand dollars a year right now i'm paying 20 across everybody in my organization and hopefully everybody in my company is actually doing this now if you hear my voice i've been like tweeting about just oh wow here we go uh let's see electric vehicles over time without tesla that's interesting uh and then the models uh yeah wow it nailed it most popular electric vehicles makes without tesla models and you see a very more even distribution in the chart nissan chevrolet ford bmw are one two and three but it's not as spiky because you're taking out and then you see here that actually the hybrid since i guess tesla doesn't produce a hybrid versus battery electric vehicles becomes much more normalized so here peak sales in 2022 it looks like is 14 000 23 is not complete yet right so So that's why it's last complete year.

22:08It was 25 ,000 over 25. I'm sorry. Number of EVs. Would that be 25 million? What is the left hand here? No, it can't be 25 million. It would be 2.5 million. Maybe a million. Yeah. Probably makes more sense. So it's, it is like, yeah, it says 14 ,000, but it actually means add, uh, probably two zero. So 1.4 million. So you're just taking out a lot of vehicles. Probably. Yeah. Tesla sold. what looks like 500 ,000 is that right no 50 ,000 is this maybe in america close to a million yeah this might be u.s because it was uh it had state ed state vehicle and other information what's the time frame what's the time frame for this like a year a month since 2000 back a few years yeah i mean this is just incredible i mean you just see like we're lifelong technologists we know how much time this kind of takes uh to do this kind of stuff and but imagine you take your website information or your podcast data and then you start slicing and dicing that now yeah and imagine the work and the number of people it took and the time it took you jason would want that answer right away where are the listeners from you know which ones you know all the different you know you're going to go after this jason and download all your data and going to be uploading it immediately is my guess right yeah and i mean well i mean does it it doesn't need to have a developer account for that or like what do you need to have to be able to use this right now you have to be um have a developer account and you need to be uh let in by open ai this is the year you need to perform you need to be focused and i want your startup firing on all cylinders and how you're going to do that you're going to use coda coda helps you do more with less In Koda, your team can work on entire projects from start to finish.

24:03That's right. One product. You have everything you need in one place. We're in the efficiency revolution. You have to do more with less. And right now is the perfect time for you to jump in and learn about all the amazing features that Koda has. Koda is the doc that brings it all together. And it's efficient and it's fast. We use Koda at this very podcast to track my Jtrades. If you just go to jtrading.com, it'll take you to a gorgeous Coda page of all my J trades. What an amazing product. It's always advancing. The templates are next level. But here's the important call to action. You can operate and collaborate in one place to get your projects done faster.

24:41Take advantage of the special limited time offer just for startups. Sign up today at coda.io slash twist and you will get the$1 ,000 startup credit on your first statement. That's right. CODA.io slash twist for a$1 ,000 signup credit. And this offer is so generous. I want you to take advantage of it right now because I don't know how long this absurdly generous offer from Coda will exist. CODA.io slash twist for$1 ,000 in signup credits right now. There is a waitlist for plugins and... But this is... It's compute intensive, Jay. Jayco, this goes back to the whole GPU shortage problem, right? this is compute intensive if they gave everyone 100 million users plus access to this it would just fry the system they don't have the capacity for it honestly like i i think we're getting to the point where this is so valuable for organizations that azure and aws should just start offering your own uh what is it a100 is the nvidia amazon's working on it but it's not that simple just to like spin these things up it's going to take a couple of years to get oh i mean Just racking them is going to take time producing them.

25:51You have to basically. Jekyll, there's a shortage. There's a chip shortage out there. I absolutely understand. But what I'm saying eventually was the word I use, Vinny. Sure. Eventually, I think organizations are going to start provisioning their own GPUs for this because it's so valuable. And if you told me right now, an A100, you know, costs$10 ,000, would you like me to sell you one for$20 ,000 to have it in your organization today to start doing this? I mean, it's a de minimis amount of money compared to the value created. I just answered another question. And I was like, which dates had the most growth in 2021 and 2022?

26:28And it said, based on this, the electric vehicles dates 2021. Here are the two top states with the most growth. Do you want to take a guess? Which states had the most growth percentage wise? Without California? No, it included California. California and New York. No, I said percentage growth though. A percentage. percentage i said which states had the most growth in 2021 and 22 interpreted that as percentage not raw numbers okay so it is in fact i'd say texas okay which are okay keep going i'm not going to say that right and texas and probably maybe florida maybe what do you know washington yeah i'd say well i say yeah like washington smart sorry uh what a number of minutes in f-bomb um washington is number one they grew from 18 to 27 000 a growth of 9 000 evs 50 growth and texas was number two uh yeah actually i got that wrong it says number of evs in 2021 three number in 2022 four it's washington state data specifically it's in washington state oh this data set yes this data sets washington state data from washingtonstate.gov oh sorry okay so what we're looking at has nothing to do uh with by state okay that's why the numbers were low okay great yeah what i'm looking i'm looking at the csv though it does have uh all kinds of counties and cities like i see san diego we just took by the way for the folks listening we just took a random data set the producers found and just uploaded it so we found this data set doesn't have perfect information and so just understand like the where this is kind of an interesting use case somebody sends you a csv you don't know what it is and it starts interpreting it for you all right well producer nick who is an exceptional producer uh you hear people talk about producer nick on all in and here at this week in startups did a wonderful job producing today and he said uh explain to the audience what you found and what you did while we were live on air yeah so i found a website where they have a bunch of csv files from government data one of which was the one that you just saw previously, the Washington State EV data.

28:33I also found one which has something to do with the topic that we're covering today about FDIC bank failures, which was from the actual FDIC.gov website, which you can see right here. Pretty amazing. I uploaded it. It found a formatting error on the CSV file, and I was about to look up how to fix it, and ChatGPT just fixed it itself. Found a Unicode error. Okay, yeah, that's common. And then just fixed it. Pretty crazy. Perfect. And then I asked it, what are the most interesting ways to visualize this data? Gave me some examples. I said, okay, do that. And here you go. Okay, so let's take a look here.

29:11It said bank closures by year, bank closures by state, top acquiring institutions. Fascinating. Heat map of bank closures, timeline of bank closures. This is fascinating. So let's scroll down here and see what charts it came up with. Again, finding errors and fixing them. Scroll down. now let's proceed with creating visualizations i'll start with bank closures by year bank closures by state type acquiring top acquiring institutions is fascinating and obviously we see by year scroll down 160 or so in uh the financial crisis and then it slowly went down but what's interesting about that vinnie if you look at it do you notice that the bank closures that started in 2008 peaked in 2010 so it was a full two plus year process of peaking and then trailing off you're going to have some per year but it still took it was basically four years of bank closures well well so just remember so a lot of this was back in the days we had a we we've had a lot of like the smaller banks being consolidated up and then they passed the laws on the bigger banks as well so yeah it's unlikely for us to see the same sort of tail right now because all the small banks have been cleaned up however if you look at the latest data and just the the amount of money that's in the banking sector has blown up in the past two three months out of like five banks we've had um more AUM blow up i think in 2023 then in 28 and 9 and 10 11 like the whole banking crisis in the past three months we've had like by dollar amount by dollar amount yeah yeah like like like washington mutual uh versus silica valley bank versus first republic etc like the scale is so different right now because these banks are so big it's interesting also about what nick found here is like you could see some of these didn't have acquirers they just shut down some of them uh you know were acquired by state bank and trust company first citizens bank america bank u.s bank NA.

31:09So just fascinating ways to look at data. If you're listening to this, in your organization, there's going to be two possibilities of what happened. This is what I've been trying to explain to people. Maybe I have to go back to BaseCal and start using all caps on Twitter. But I am finding that 30 % of what I do can be done inside of ChatBTE4 today. I'm finding my producers and you saw nick pull up his thing there and i saw questions in his thing that were questions i was asking during live this week in startup so when i'm doing the show the producers are looking up data they're using chat gpt4 all day long um and even during shows so this to me is what i would implore people to try to understand right now smart people who are using this are taking i would say between 10 and 50 of their job and automating it and then they're quiet quitting or they're doing more work and they're going to be more effective in their organizations or their boss is going to figure this out and everybody's going to get more work done.

32:11And instead of hiring, people are going to start firing and getting more done. So just think about gains. 30 % gains across an organization of, let's take my investment firm, about 20 people. That's the equivalent of having 26 people. So one of two things is either going to happen. If you had 20 people, you're either going to go down to 14 and save that money or you're going to act like a 26 person organization or something in between. That's how management thinks. Now for my team, we're doing a great job. I just want you to become 30 % more efficient so we don't have to hire more people, but other people are going to look at this Vinny and they're going to take a different approach, which is, okay, we have how many data scientists?

32:51Great. Half their requests are not necessary. They're going to be done by chat GPT four. People are not gonna need them. So we just get rid of half the data scientists. Now take a moment to, think about what I just said. There's been a competition for data scientists. Some organizations say, how many of these data scientists do we need? Well, I'd say right now, J.K.L., we probably don't have enough on a global basis. So I don't think there's going to be a shortage of data scientists anywhere. So they may be reallocated from companies that have seven down to three, and then those four go elsewhere that's needed.

33:29so i think you probably need fewer data scientists per company but there's still companies out there that's going to need that never thought of having data scientists because they just didn't have the you know but i mean like you still have to pay for the licenses right for the software that they use which is like millions of dollars a year so now the cost of the software has come down dramatically you still need the people to operate it because you know some people just need to be focused on the stuff and a lot of companies are data and multiple databases and spreadsheets and it's all very disparate you still have to build data warehouses that have all information etc et cetera.

33:58So it's not as simple as that. I think that - Is it not as simple as that? No, I don't think so. I think in a world where everything was highly efficient and everything was run properly, yeah, maybe. But we're so, I mean, the gap right now between the haves and the have-nots in data science is very, very big. I don't know, Sonny. I might disagree. This weekend, I started learning Python. You already called me Sonny right now. Vinny. No, I was going to Sonny. I was going to throw to Sonny. I thought you did too. i was like oh my god i was gonna throw up sunny listen you guys are sunny and vinny you're two of my best friends the names are different by one letter sunny and vinny two letters sunny vinny two letters oh right yeah yeah sorry sorry i'm i had a long weekend i had the kids alone anyway i am going to disagree vinny and sunny i want you to reflect on this you and i were chatting we're trying to get together over the weekend to do a little code jam but you know kids whatever got in the way but i started on warriors game nick's loss warriors one incredible uh shout out steph curry replit uh is like a coding environment so i just signed up and i started taking their python course i was like oh my god this takes so much concentration i'm never going to be able to do this like this is not going to be my chosen career but i do want to see how far i can take it because they have a bounties thing on replit and i put a bounty up and then i explained in details i'd like an auto gpt agent that checks our database of already contacted companies by url so these are startups we've talked to so we say hey com.com and uber.com are in the database we don't need to call them then finds new startups on crunch based products hunt and linkedin and sends them a semi-automated email from one of our researchers introducing our venture fund acceptance criteria app is able to find a recently updated crunch based profile within a specific criteria geography investment stage and sends an email to that founder pretty simple right and i put this up for 27 000 cycles i guess they call them on replet shout out to the team at replet that emailed me immediately after i talked about it on the pod um and i put it up for 270 i got four applications and as you can see here one person says jason have built this in the past and building for a few funds so i'm not the only one thinking like this would love to chat more about you can check my github linkedin for resources and he's done three bounties uh cripes jake out i'm a fan of the pods i've read your book dumb luck i'm poking around replet and see what all the fuss about i stumped you about regarding your bounty i'd like to help uh ask you to flesh out your criteria yada yada yada i do either of these free as long as we can take pretty much time to you coaching me on my personal churn i don't like taking free stuff but anyway my point here vinny yeah and then i'll go to sunny um is i am the ceo of the company i'm the gp the general partner of the fund i'm looking at this and i'm like i wonder how long it is between when i can describe something to a bounty program and have code sent to me and then i run it myself just like i am using chat gpt4 and i feel like I'm on a collision course sunny between using chat GPT for with plugins and uploading stuff myself and then working with the developer community to write tiny little scripts for $270 that add a$50 salary or$40 salary or$60 salary for let's say an operations person in our organization you know that would take five hours I can basically take what is 50 hours a week of work in our company two researchers doing 50 hours a week of work 1500 a week maybe i don't know two thousand dollars a week fully baked with benefits a hundred thousand dollars a year of work and i can just automate it for 270 bucks am i crazy or is this going to change the world no i mean you're you're we're 90 days away jaycal 90 days away at the pace we're going at right now because you know what you put in here is mostly just doable and um and it's like i said we're entering a world where um the core framework is being absorbed by open ai and so if you just saw what we did um that they're gonna open like they're taking their they're taking their time right now from a safety perspective that um the code interpreter that we were just playing with jcal doesn't reach out to the internet just yet but we know that they have browsing capabilities because there's other plugins that can browse yes as soon as they allow code to go out to the internet yes which you know they've controlled that it's not like they don't know how to do it then you have that problem solved right inside code interpreter it's crazy because you would describe your problem inside code interpreter and say here's my spreadsheet go to crunch pace And so the same thing you did in the Replay, you'll do inside there.

38:55Developer talent is the most precious resource for B2B startups. You know that. And you want your developers focused on product, not on compliance, right? When you're selling B2B software to large enterprises, you need to jump through a ton of security and compliance hoops. And one of those hoops is large customers need you to host your software on their cloud. and you need to build that out on a per customer basis. Think about that. So B2B startup companies constantly face this dilemma. Do you keep developers focused on infrastructure, which could hurt your product velocity, or do you keep them focused on the product velocity, which would then delay your ability to close large customers?

39:37Well, I have a solution for you, and it's called release delivery. What release delivery does is it automates the creation of enterprise class app delivery for private clouds and single tenant applications. Basically, this lets you deliver your software seamlessly into any customer environment. This will unlock a ton of revenue potential for you and release delivery will put all the tedious stuff on autopilot for you. So you can turn your ideas into apps and deploy those apps quickly and flexibly into their clouds. So here's your call to action. Let release show you the power of release delivery and get your first month free at release.com slash twist.

40:17What a domain name. R-E-L-E-A-S-E dot com slash twist. It's up to$10 ,000 in value at release.com slash twist. I would agree with Sonny on this. I mean, guys, this is the fifth generation language. Like, you know, we never really got to it. This is natural language programming. Like, everyone's a programmer now. You just need to be able to speak English at this point to be able to do it. Not even English, other languages as well. activity you can translate for you so as long as you can i mean if you think about it like you know language is code you know like natural language is code and we just we had to create this layer where with digital uh you know digital machine you know software programs machines could interpret what we're saying accurately and because the human brain is so complex so language is a very complex thing for us but machines that we've had to instruct machines based on a very limited number of words, functions that we have that was written.

41:14And now it's fully expansive. Now you have the entire English vocabulary that you can use and the machine understands what you mean. You can be extremely precise in what you're saying to it as well. Whereas in the past, you'd have to write functions to do certain things. It basically now understands every single word in the English dictionary to a very, very deep level. And every single word becomes effectively somewhat of a function or a describer or something. so like you know i posted a tweet i think yesterday we'll pull it up nick i think this is a very important point we should we should probably touch on um today and get your your views on this um i think that that you know in the next cycle so we're in a we're in a bear cycle right now right we're only heading to one or whatever you want to call it like obviously we you know we may not be in a recession i think we are in a recession for what i'm seeing i'm seeing the signs of a recession already.

42:06The next cycle that we go through is either depression or it's a recovery and a boom, right? So whatever you want to, you know, however you want to define the next cycle. Regardless, I think we're heading for deflation in a big way. And I think that this will become the number one driver of deflation. I think you're exactly correct. What's going to happen is massive efficiency will come to the companies that get on this early. then what will and you know the if you're running a company right now you should just give everybody the tool ask them to show you what they did with it and if you have 10 people in your department if seven people use the tool and three people don't you should fire the three people who don't use the tool i know this sounds crazy but this is exactly what i saw happen in the early 90s we put pcs on people's desks some people literally did not want a pc on their desk they wanted their secretary to have the pc and those people lasted i think you know less than a decade in corporate america and that was back then when you know you got to keep your job for a long time there wasn't as much turnover for boomers but there were boomers who were like literally when i was installing computers in the early 90s who were like yeah just i don't want the computer you can don't put it on my desk put it on this like little cubby over here in my in my law office and my assistant will do it They never logged in.

43:25And those people got phased out. They were relationship people. If you're not using this every day, you're literally a dinosaur. You're literally a dinosaur. That's my belief. So you're exactly correct. This will make every company 30, 40, 50 % more efficient. And then what you have to ask yourself is, are there enough problems in the world that your company addresses for you to solve to generate revenue in a capitalist society? I believe there are decades of problems left. i don't think that this is going to result in uh a ubi universal basic income where all the jobs are done i think humans are going to be creative and find more things to do but i literally believe efficiency of five percent gains per year for humans let's say if everybody got maybe let's say everybody got ten percent and every year every seven years people doubled their efficiency i think what we're going to see is everybody's going to become ten percent more efficient like a month or let's say a quarter which means every seven quarters every year in nine months people are going to be twice as efficient what do you say sunny well i think there's a great example jacal and i've seen it but nick if we can pull it up in terms of efficiency so um this is someone who's working on a do not pay plugin oh josh brown program yeah yeah oh there you go so maybe just you know josh browder is bill browder who wrote the book red notice his son he's an entrepreneur and he has do not pay is the name of company he's been on the podcast and his whole thing was to help you like um get out of like um reoccurring subscriptions etc but he's also so let's do a reaction thing jay cal why don't you read this because you've seen it so go for it i haven't seen this so what did it what did it say okay so so this is you know do not pay it's an app on top of a chat gpt leveraging it it goes to ask, how can I help you?

45:14He says, find me money. Is it connect? The app says, connect your bank account. Uh, it connects account. And then it, uh, finds the subscriptions that this person is paying. It obviously learns that ever. And then it says, what do you want to do?

45:31Yep. Okay. So let's go to the next spot. Incredible. Okay. And in this, he says first using do not pay at plaid connection. I had, uh, it scanned all about 10 ,000 bank transactions. So it found$80 and 86 cents leaving his account every single month and offered to cancel, offered to cancel those. Great. Okay. Let's keep scrolling. Okay. Then the bots basically got working mailing letters in the case of gyms. Right. And it used a USPS, um, API, um, and chatted with the agents to basically start working on the cancellation. And so we can scan through this and we'll maybe drop the link in the notes.

46:16But the beauty here is going back to efficiency. Think about the time and effort. There's one last example, if you can go back there, Nick, where it actually found a bill for a Wi-Fi connection. And it turned around and asked, hey, did the Wi-Fi work properly? When he said no, it drafted a letter to send to GoGo or whoever the Wi-Fi company was asking for a refund for that. And we've all experienced that where we pay for it and it doesn't work or it's bad. And basically, yeah. And so very similarly - We let it go start a negotiation process to cancel that and get a refund. Yeah. And then similarly, it started a negotiation process with Comcast.

47:01it's just that's what i'm saying jake how we're these are apps that are being built on top of the technology so we are almost where you're talking about as i said less than 90 days away from incredible things happening for us which then aligns the deflationary argument it's definitely going to be super deflationary if you hear my voice you know like and you're not using this and you're not getting up to speed on it man um yeah you're not really following how fast this is I started playing with, I'm giving a speaking gig on Wednesday in Laguna down in Orange County, doing my paid speaking gig thing.

47:40It's a corporate gig, and I'm talking about travel. And so I started testing some, I was like, you know, in this luxury hotel kind of situation, I won't say which one, but let me share my screen here. So I started using the GPT-4 with browsing. Browsing? Web browsing. I don't know if you've played with this, but it doesn't work very well. I had said on All In and Chamath and Sachs laughed about this, that, hey, you're going to need to start citing your sources and then getting permission from them, etc. Or else this thing is going to become gnarly. And all these lawsuits have already been filed.

48:14But when you hear, I said, what are the major trends in luxury hotel travel? And it started to browse. And I guess it did a search. And it said, search major trends in luxury hotels 2023. 23 it found this link 20 from a website ehl and then it read the content a bunch of failures it's not working very well their web crawler is terrible or it's really taxed i don't know what's going on my team today has been playing with the web crawler but it only found this one and then it basically just cribbed it so now you can kind of see what's happening with chat gpt4 it is cribbing a lot of data and just rewriting it um and then it does some thinking on top of it i want to i want to clarify something.

48:56So in the case when you're without the plugin, you're asking for something, then the cribbing is not occurring. And I think that's a discussion that's happened before. In this particular case, you're asking ChatGPT to go look for something with the browser plugin, so then it will crib. It's two very different use cases that we have to be aware of here. So anyway, this EHL Insights had written this and, you know, you can see it basically took look what they had on their website and it summarized it a little bit better and then way down here it gave a citation you see that 12 it gave a little tiny citation and then i said which hotel chains are known for having the best hotel workspaces none of them offer dedicated work desk and high-speed internet over ethernet connections and it started browsing the web it's actually doing it right now because it failed so many times but i want to show you another uh one i did here and this one was a fascinating and i said what are the major trends in luxury hotels and it gave me um up to september 2021 this is without doing web searching personalization sustainability wellness authenticated experience smart technology blending home blending work and leisure unique design and architecture multi-generational appeal privacy and exclusivity partnerships so i said which three of these are the most important for maximizing a hotel's loyalty and revenue so i'm asking it to think you know a bit here yep and uh it said personalization smart technology and authentic experiences and i was like huh the first two definitely authentic experiences i don't know if that's actually like culturally immersive activities genuine connection to the destination i was like i don't know this feels a little woke to me i was exactly my time i was about to say that it's what gpt so i was like please give me 10 examples of how a luxury hotel might personalize a hotel guest's experience so i just went after the personalization and this was incredible like and i don't know if where it's getting all this from like is it from its web crawl you know but it said pre-arrival communication customized welcome and amenities like a favorite drink or snack tailored room setup um oh like temperature preferred lighting curated experiences personalized dining options customized spa treatments dedicated concierge service flexible room configurations tailored in room entertainment uh personalized turndown service i said you know what expand that list to 25 ideas and it just went to town you know and customized minibar i'm like well that's a great idea i've never experienced a customized minibar i've had that idea before personally personalized business and wellness programs uh customized transportation options customized bedding and linens i've heard about that actually chamoff has that at uh the peninsula where they have cp pillowcases with his initials on them uh so he had talked about that pet friendly personalization uh scent experiences that's dumb but interesting um personalized communication they do that in vegas yeah communicate with guests through their preferred channels such as text email or phone that's actually a really important one some people everybody's got a different one um so then i was like okay well let's take this to the next level And I said, let's see if we can set up a scenario where we tell it to pretend it's something.