Programming with AI: Claude 3.5 Sonnet Projects - deepseek-coder-v2 - AI Engineer Worlds Fair+++

hello all hope you are doing well just going to get my chat up and going here always takes a bit of a time but yeah I can see there are some people here so that's cool uh just test this hello I don't know how to change my name I haven't really I'm going to change that name some other time hello camel nice to see you hello night spider great to see you back again that's cool so

as usual we're just going to take it a bit slow in the beginning hello moon dog davidas nice to see you again so yeah like I said uh we're going to do some looking at Claude um the projects thing so we can take a look at that hello facts hello ibug cool to see you hello digital applied yeah like I said uh I was going to look at the CLA projects to can take a look at

that they added custom instructions that's will be cool Hello Yushi hello mihawk and uh yeah just some stuff I thought we could take a look at there's this AI engineer Worlds Fair uh I thought we could take a look at kind of the first keynote speak here I thought it was pretty cool maybe some of you has seen it but I kind of liked it we can take a look at that together I thought and just

some different stuff Hello Drogo and yeah I've been playing around with uh Claude 3.5 sonnet looking at this project part you can see I created this app here using um using Sona just earlier today so basically I wanted to C try to create this app that kind of it's just running in my terminal it looks at my CPU GPU usage my memory and I added some random stuff Hello Rakesh hello flash rebbit nice to see you

back again so night SP uh try the project yesterday pretty smooth uh create a custom to-do app with nextjs and to make it download cool yeah so this this small app uh it's just running in my terminal so I just wanted to add some stuff just for fun so uh yeah you can see if I have the live bitcoin price here I just wanted to test that and every fifth minute we get kind of the headlines

from Hacker News and I'll show you when it pops up so today flash we are just going to look at the projects part from CLA 3 uh yeah few videos maybe take a look at from the AI engineer Worlds Fair and just some small stuff uh maybe not the longest stream but we'll see hello Tim nice to see you again thanks man appreciate it so yeah like I said uh when we actually I think it's every

fifth minutes you can see here we scrape this Hacker News uh headlines so it's going to shuffle through every single Title Here I was going big though but you can see it's going to shuffle through every single headline and if we press o I think we will actually go to that uh to that page hello russu so let's say if we see some headline that sounds interesting here open Sora I can just press o and we

will go to that project right and we can take a look again we can open this one on so basically we can just open these uh headlines from U Hacker News just by from kind of our terminal here so seems to be working pretty good so we can just watch it's and it's going to update every five minutes or so you can kind of set the timer yourself so we built this app maybe like 10 minutes

or something using 10 15 minutes using clae 3.5 I kind of like it I just have like having it like down here and I was thinking about adding some stuff too uh when are we going to update the Discord invite link uh I don't know if I said that like uh I changed I removed that Discord yeah I get I get I guess I should probably remove it from YouTube kind of forgot that but like I

said I we created a community Discord because the old one had like 7,000 people and there was so much spam I couldn't keep up so I just deleted that one uh but yeah this app is pretty cool so it's been about five minutes yeah give it a few more minutes and then we can kind of get into what I wanted to take a look at today and uh yeah that is kind of the clawed project uh

I also saw uh let me see here I guess you all saw it but um so uh open AI kind of delayed the new voice mode so they said Late July but uh I don't know are we expecting it to get it in Late July probably going to be a very slow roll out anyway so I'm thinking more like August maybe we'll just have to wait and see but yeah it's a bit of a bummer but

uh yeah there's a lot of other stuff we can do too like uh CL 3 son 3.5 came out and yeah we we'll get there there's no rush I see a lot of people on R Reddit really stressing about really want it now but yeah we will get access when we get it right uh okay so I think we just going to it's been about s minutes now so yeah clae 3.5 sunet so what they added

just going to turn off the dark reader here so kind of what was new was this thing called projects so I had a look at it this uh I guess yesterday and this morning and I think I like it so I created two different projects uh will the projects be available for free users or do you mean like the open AI part any services that host uncensored llms yeah that's AMA R AMA they have uncensored models

but yeah here is where I created my terminal app that kind of had the computer vitals Bitcoin pre I hack can use and I just tested out some other project but I thought we can create one from scratch here so this is probably behind my head the open AI voice is uh probably going to be at least first for paying professional plan I guess and other than that probably going to be open yeah rash maybe together

AI has uncensored models but the open AI uh no you can check out together AI they might have some okay so this is behind my head probably but we have this button here called create project so I'm just going to click on that it kind of takes us to this page so we can kind of name our project so I don't know let's call it just uh back end to front end that was kind of want

wanted to test today because I have this IDE uh what you trying to Ace uh I don't know connect uh backend code I don't to uh to a front end UI so this was what I was looking at with this projects that I found pretty cool been working with projects for a few days and I'm liking him uh I'm curious to watch your workfl yeah I think it's pretty cool like you can kind of do whatever

you want with it but I was looking more at just thinking about it just for coding and I had some few ideas I wanted to test so basically I haven't tried I haven't really tried anything else but to write some code with it and yeah like I said I just started yesterday so there's a lot to explore here so this is just going to be kind of my early tests uh will M hello welcome looks interesting

haven't checked out projects much yet yeah so I think it's pretty cool you can kind of set your custom instructions I haven't seen that in CLA before so uh I created a instruction I wanted to try we can go through this and so yeah you just click here so but basically this is the same as with open a it's basically a system message right so this is the one I wanted to try today just give it

a roll uh you're a full stack software there with expertise in Python yavas script tail when react yeah just mentioned some programming languages uh when you write code always uh use very descriptive variable names extensive use of commments to explain code blocks Write Clean and secure code okay let me just see uh I've been working on a Unity dots project and uploaded some of the unity docks and best practices the code coming out is amazingly complete

okay that's cool sounds interesting so you like it so far Brian yeah Josh thanks for that I can see it now uh that's great the the script TP desk r t okay that's better right so the reason why I wanted to do descriptive variable names and extensive use of comments to explain code blocks uh it only has to do with the llm right uh it can be annoying with all of this comments when it's if it's

all over your code right but when we are using this llms I don't really care about that we can remove them later but I just found it easier to get the output we want when we have very descriptive variable names and also comments okay Brian so you really like it that's much uh so much more useful just need a larger context window yeah like when you talk about code bases and stuff like let's say 200k might

sound much if you're doing some writing but that gets really filled up if you have like a large project right I think I need to adjust my camera something like this let me just see okay sorry about that yeah so this is kind of our custom instructions and yeah let's try it and write clean and secure code I haven't played around with that instructions so much but I just wanted to try it out your project knowledge

can get eaten up really quickly yeah I haven't really tested that out but uh I've seen like when I added some uh knowledge here so add content uh this feels up right so this is what I wanted to try out today so my idea was that we write a small block of code we add it to our content and then we kind of move on so I want to start really simple I kind of see where

that takes us uh I haven't really thought of anything but let's just start with something very simple I want to do it really simple so what kind of back end do we want here you should start with a small Cod base that reflects what coding style you want for your preferred language and knowledge uh as a Artic knowledge article yeah that's pretty smart I think that's a good idea but I just want to do it very

simple now and we might try to expand on it later uh the first thing I wanted to do was just write a very simple backend code and see how we can add that to our content right I need uh we can do it in JavaScript maybe I think so El script code um that takes a user input text maybe I don't really care about this uh text par say out uh all numbers from the given text

then sorts all the numbers from I don't know low to high using uh we can use like a bubble sort or something I'll go right hello cool mude looking for claw 3 Sonet content to help me convince me to cancel my open a subscription yeah yeah I can see like the nice thing is that you can switch around right this month you can try out 3.5 next month you if uh you want to try out the

voice mod for an open AI I would do that since this I'm just going to use both now but like I guess you could save some uh okay cool I like that welcome okay so let's just try this let's just start with something simple here so let's try try to create like a simple JavaScript code that uh takes a user input text and just extracts all the numbers from the given text and just sorts it from

low to high using an algorhythm right sorry about that I think my phone is ringing but uh yeah I think it's fine so let's see I guess my camera went out but I'm going to fix it let give me a second okay should be good now let me see hope I'm still here I think I am so let me try to fix that [Music] camera yeah this is the thing about streaming right I am I'm not

so used to it I don't know what's happening now hopefully you can see me uh I'm just going to try to fix my camera so hopefully I'm back now yeah I should be back soon okay sorry about that so where let me know if uh let me know if you can hear me now the problem I'm facing is that claw 3.5 son has a limited conversation any idea about uh bird of Cl probably r conversation limits

uh yeah sorry about that so I think the conversation limit uh with uh Claude 3.5 is 45 messages every 3 hours or something I can't confirm that but someone said every 45 messages every around three hours I think so I think it's it's less than um GPD 4 okay so let's see our code here now I think this looks fine uh so we have a user input right and we have a bubble sort function here and

event list not a button okay so let's just leave it like this so what I want to try now there's a button here called add current add to current project so I just want to see if we add this to our project and then we go back and see what we can actually do with this now this is added in here right hello Kurt valesi uh just curious uh how you think about Google Gemini large Contex

window but will it be as good as smaller models I really like Google's user friendly approach for beginners yeah I think Gemini is viable The Annoying Thing For Me living in EU is that I don't have access to the AP I so I haven't really tried it much but uh I would give it a go if I had access to the API okay so kind of now we have added our backend code here to our content

right so what I want to do is actually try to connect this with the front aui now future CLA coding idea for you create a bot that automatically provides all about the get yeah that's a good idea hey TKS nice to see you yeah the GitHub uh Auto uh inviting is a bit personal for me I kind of don't want to automate it so flash says 45 quality messages versus 8 low yeah uh my initial action

is at son it with coding uh more concise and less worthy when explaining code yeah for me it's been I've really enjoyed it I think it's pretty good I think it's the best coding llm out there now uh at least for my testing but I do like 402 okay so let's try to upgrade our promp here now okay let I have uh created a backend code uh uh called I'm just going to try to give it

this name and see if it can recognize this I need a front end uh where the user can type in some text and get the numbers extracted and sorted can you connect our front end to my back end place I might have to take this into vs code but we'll see no problem Kurt valesi is there a way to connect this to example Google Sheets and docks I haven't really checked that I don't think so chemical

flask in the that is kind of the artifact that we have this enabled right so that's pretty cool so caner uh do you have an AI agent flow for programming with AI uh I have some ideas yeah but uh nothing planned for today I did have a plan to kind of do a conversation between Sonet 3.5 and GPT 40 but that might be the other stream right okay so here is kind of our UI right so

let's test it out now I have 100 apples 10 frogs and dollars okay so that seemed to work let's do $1,000 yeah that is working so what I want to know now is that did it take my previous back end code and merged it with the front end yeah it looks like it did or did it write a new code I'm not quite sure so let's just see here um okay so the extract and sort numbers

function is adapted from your backend code okay so that's pretty cool yeah that's kind of what I want to see I think it's it's funny how silent 3.5 dropped versus yeah I like the the lowkey uh anthropic style that's pretty cool it's react the other code with JS yeah but I wanted to see if they actually just uh you can see it's still uh JavaScript here right for those wondering if you make the side detail window

know compress you will see all the files so so like this Brian so here's kind of the artifact so we can click on that right and see this uh I just made a Bitcoin M Pool assistant you might be interested checking out using clae or Alama that's cool so it does State here that the extract and sort numbers function is adapted from your backend code the UI style tailwind and the bubble sort is also implemented from

your backend code so that is kind of what I was looking for okay that's cool Brian and there's a star here now so we can actually star the chat so this is what I was kind of thinking about if we can actually if we go back to our project now and if we can add bunch of different parts of a bigger project here right so let's say we call this you might have to be precise with

your naming then so this could be some kind of function or parts of the code uh and we can add a lot of different parts of a bigger project here because last stream I said uh I am looking at creating a bigger project using three CLA 3.5 Sonet so it might be a game uh it might be something else I haven't really decided yet but I'm going to try to make a bigger project uh that is

going to be a few at least a few thousand lines of code right and but I wanted to find out today is if I put up a bunch of different uh parts let's say of a bigger project here if all of them are connected so I don't have to keep everything inside a window I thought it could be much more easy to organize and it looks like it's actually using this as context and that that would

be as expected right but I kind of like the way they set it up here I don't know if any of you have tried that to kind of do a bigger project with projects yet but I guess since it's called projects that is kind of the idea right uh you have to be jic because with the code you add note that you're 1% already you can get a lot in there though yeah like the context window

in theory should be 200,000 tokens uh in addition to JavaScript and python how is to support for Lua language for game development that is interesting I can't really answer that because I haven't tried it but I just know that JavaScript and python are kind of the most developed languages for language models of course because that's think they have the M most data right but uh I think like other languages should work too but it might not

be as good Hubert hello uh maybe 1% knowledge size is the minimum size of the cash of the scale yeah might be yeah I had issues with Tailwind in previous attempts so I copy paste a few pages tail code into dos yeah that's smart if you have some specific use case or maybe you can just let's say we wanted to add something here from uh the documentation we could actually test that out so so let's go

back and start a new project goang and rust are fine too okay that's cool so let's start a new project let's call it open AI docks you seeing open AI docs to right code I don't know doesn't really matter okay so let's go grab him here some documentation from open AI so yeah we can use God text generation models just to grab some some uh documentation Here Right add content let's call it open AI docks our

company hopes to generate 40 50% of Earline gave development logic Yi wow that's a lot do you know meta GPT how does in compared to son Sonet you probably mean no I haven't really looked into that but I think like CLA 3.5 son is kind of the best model at the moment for coding I I guess gp40 is good too I guess they're kind of on the same level but I just prefer CLA I don't know

why okay so let's add some examples here so let's uh we can use Python for this that's fine so let's grab some [Music] documentation I don't really care this is fine hello Loki nice to see you back let's also grab uh an image part here uh let's go to I think we're going to go to docks we're going to go to models jp40 uh and let's go with wish let's add in the documentation for wish here

we can do that in a separate file maybe uh shat completion and let's add a second file let's call it open AI docs wish I don't know just use chain lit for a UI super simple to use cool I haven't tried that okay so you can see now we have used 3% of our knowledge size I like that they actually tell us how far we are along in the context size so what is okay so this

is just some doesn't really matter so let's go grab our Uh custom instructions again but this time let's just use expertise in Python so let's do a python script this time okay okay so we uploaded some open ey document ation here API documentation can the code generated by son connect to my SQL database or can it only generate static code I guess you could connect it like but not in the browser no if that what you

mean uh low we're just exploring uh Claude 3.5 Sonet the project part just testing out see what we can do with it create uh or write a python function that [Music] uses uh GPT 40 API to do to do a chat completion include an example of use I don't know let's try it open AI docks yeah okay I tried mess up it's ugly compared to stream lit and chain lit okay okay this was interesting let's try

to run this so let's just uh uh we can go here I guess uh let's just call it I don't know open ai. p maybe not that chatbot dop I just want to see what they did here interesting I haven't really seen it on like this before we are selecting a gypt 40 model this was super interesting okay so we need to set our uh API key I think we do have an EnV file here maybe

we don't uh so that might be annoying [Music] but I like uh I like the where did I put it here kind of interesting to see how this works so let's just for fun do like uh what do they call it open AI key and let me just go here and I'm going to log in going to grab a key so I have my second computer here [Music] profile okay yeah maybe that's not such a good

idea it would be great if the page code could B I can see it uh I can stop using gp4 pre through their UI uh I'm getting big project created in one to three point story chunks it seems to get the best results from keeping a contact small and focus yeah that seems smart Tim I think that's a good idea give me a second here I'm just going to go to my personal okay uh this is

just going to be hassle using this open AI Keys when I haven't really created this EnV file it's just going to be annoying because I my second monitor is uh somewhere else so I can't really drag it over without showing my API key I don't really care about I don't really want to do that but let's just move on then and let's go back here so that seems to work fine by pasting in some documentation uh

I think that's going to work pretty good pretty standard we could always do that in uh documentation before and so another thing I wanted to do was um this project I was looking at so let me just grab the grab the yeah Tech friend yeah I could just revoke them but it's not so interesting uh looking at the the open API call it's not so interesting but let's try this Interactive Learning because I saw a post

on U on uh Reddit someone doing that that was pretty cool learn from uh code base let's just call it that so we're going to do it simple can I generate both uh front end and back end code in one prompt for example yeah you can do that that seems to be working pretty good at least with Sonet so you sometimes get your back end and you get your front end just in two different versions and

that seemed to work pretty good okay so let's set our custom instructions so this is what I wanted to try here so your software developers with expertise in uh y JavaScript react uh your task is to create an Interactive Learning Experience from a code base the goal is that the user should get familiar with the code base all Interactive Learning should be shown in the preview window and use react component that doesn't rely on external UI

libraries aimed to creat a pleasing interesting looking UI ux experience so the idea was that uh let's say you you kind of have um a code base you need to learn maybe just a small project you want to dive deeper into and I was thinking if we can use uh clae to kind of create a more interactive uh experience than just sitting reading the code so I haven't really tested it too much I just did a

quick test run uh but let's just find some some kind of code here I don't really care what uh let's start with something simple so this is the this is the code from yesterday's video so this is just fetching some bitcoin price from an API just grab this code I guess we could add it to content I guess um fetch Bitcoin code paste in our code here can you send us your prompt uh yeah I can

guess I can do that this big one here let's try to copy it and paste it I don't know might be too big okay so we fetch Bitcoin that's fine and let's say I have uploaded my code fetch Bitcoin I need to learn how this code is functioning on a lower level create a interactive I don't know session for me to learn so this is just going to experiment I don't know if this is going to

work though I saw someone they put up the paper from uh the machine learning paper attention is all you need and they created this uh learning Interactive Learning session how do you think it's different from co-pilot you mean Sonet 3.5 okay so we have the Bitcoin fetcher code walk through importing dependencies show explanation okay that's something right okay so we have the next uh part of our code that is going to be the fetch Bitcoin prices

uh this part of the code implements a caching mechanism it checks if a file exist so it returns a cache data avoiding unnecessary we have an explanation for every single kind of parts of our code API request I guess this this is not too bad it kind of breaks it down easier to read this Loops processes the API response extracting one Prim per day within this date range okay pretty okay let's uh Create a quiz from

the code please I might look for that redit post it was pretty cool maybe I can't find it so let's see if we can get a quiz now from our code small code here so we have six different questions what is the main per purpose of caching mechanism in the code that is is going to be to improve performance by avoiding unnecessary API calls yeah that's correct can we answer wrong how often is the cash considered

fresh every 24 hours which API is used to fetch the Bitcoin data that's coin gecko what range does the code fetch Bitcoin prices for last 30 days what is the format used to store dates in processed data uh God I can't remember that is there any problem with my audio just let me know is so okay what does the code do if it receive less than 30 days of data from the API less than 30 data

uh yeah exception you scored five out of six okay pretty good fact says audio sounds good for me okay that's great yeah so I think this is pretty cool let's say you are a student maybe or you want to learn something new and I think this is a good way to practice pretty smart to be honest but I guess you could do this with chat GPT but you can actually run it in the browser here so

you don't have to kind of fire up a react app so that will save you a lot of time right yeah try that Tim so pretty interesting let's uh create [Music] here goes my camera again create a visual of the code base let's see if we can get visual maybe a flow sh I wonder if you can turn that uh in a jupyter notebook file probably okay uh bitcoin price fetcher explainer wow that was pretty cool

so we got a overview date processing sample shart okay and here is kind of the flow shart so we have start the cach exist yes is it less than 24 hours no fetch new data calculate range fetch data process cash return price end okay that was pretty cool so we can zoom here pretty neat if you ask me I think there there's a lot of things to explore here is your clae uh Pro or free I

have a Claude um Pro now that $20 a month subscription it's amazing ask you to make it to an animation you can try that uh can you animate the flow sh I haven't tried that before mermaids built in animation capabilities can we actually run that in here let's see if we can do that what is this animation this is not an animation this is just weird right there was another what is this then no I think

I have to explore this more the cup animation yeah that was pretty cool don't forget to put a thumbs up yeah smart uh okay so I might come back to that uh I also put in kind of my title that I wanted to check out um uh let me show you I wanted to check out um deep seek coder V2 today hello it Shield nice to see you back again so deeps coder V2 an open source

mixture of expert code language model that achieves performance okay so I heard great things about it I don't know if any of you have tried out DPS coder yet it should be pretty good but there are two uh models I think there's one big one 236 billion and there's a 16 billion so I have pulled a 16 billion and I wanted to try it out today and I think it should be pretty cool it's good yeah

I think it should be so I have this thing I like to do when I'm testing this new code models so I think I'm just going to kind of go through that today so this is usually something I do when I'm maybe I'm on a plane I'm on my laptop and if I can actually run the model then we this is something I like to do just to test it out so I thought I can just

share that today uh it might look a bit strange but yeah in case you haven't seen my app if you just got here so this is an app I created so this is basically we have our CPU usage memory we have the live bitcoin price and we have some hacking news headlines here that is kind of this news ticker so if we click uh o here we actually go to the to the news Clicker from uh

from the so figma slides so we can go to that we can also lessons I which have been thought so we can open that just by clicking o so that's pretty neat yeah that Al kind of like this small little app I like just like to have it in the background it's free to use anyway we don't use any API we are just scraping we are scraping The Hacker News for the top head lines every fifth

minute so it's pretty neat but yeah let's check out deepy coder so I'm going to take you through kind of what I like to do it's a bit strange but something I like to do when I'm testing out a code model so I have this kind of routine I use it's pretty strange but um try to keep up but let's just end this because I sometimes I like to use the terminal for this all right so

let's just open up the terminal here uh I usually use vs code but for the video I thought we can do it so people can see a bit better so the IDE I usually you you can see we have selected DC coder here uh from AMA right and I have a system message now uh you are a Godlike python programmer that's what I always use so the thing is I like to let me just open my

can here so what I like to do is I like to use the model to expand the code here so I usually just start by something like a variable Helper and then I do like a prompt and let's say write uh python code function that uh can save a file in our cvd current working directory so this is what I like to do so when I'm on the plane I don't have the uh Network usually so

if I have my laptop I'm just going to play around with this so we're going to start with writing a python code function that can save a file to our CBD I'm going to show you why I like to do that so let's just run this now so now so now we are uh what did I call this we kind of need to print that right print helper right okay and we need to do AMA response

something like this okay model I'm testing and I'm going to show you kind of how I do that I use some kind of context yeah we'll get into it shouldn't take too long but the goal is to actually create a python code that can write a new python code that we can execute so we're going to do some regex and some stuff like that uh okay so let's see here now um so this is uh we

need import OS yes that's correct right so I'm just going to copy the functions we get back here now we can bring in the example Usage Now and let's just bring in that here the function yeah we don't need that so you can see it has some kind of explanation here but let's just remove that and we have the argument content and file name so that's good I like that so when we going use this we're

going to put in our content and our file name okay that's fine so that's a good start and the next thing I like to do uh because I want to create some kind of context I want to create a python code function that can uh also open and read a file from our working directory right I've been working with CLA 3.5 son to create a rag chatbot for local documents it's working but the responses are not

very good yet so what kind of uh embeddings model do you use Joe okay so here we have a function called read file name so let's grab that take a look at it see if it looks correct yeah I just like to remove this so we want to read right and we have some exceptions here that's fine we can keep those in I don't really need them but yeah it's fine DC coder V2 and GPT 40

who will win yeah uh that's interesting hello William nice to meet you public somewhere but what history do you have with professional programming and computer science in general thank you so I don't have like a uh any Computer Science Education I like have this uh electrical engineering background so I had some maths and stuff but like programming it's just been like a hobby for me and I was very early with these language models and stuff like

that so just a hobby I guess just teaching myself but uh at school engineering electrical engineering we did so programming of course and math uh yeah okay so we have a read file right so this is now where I kind of create um a like to create a file here so let's call it this context or let's call it code context it's the variable here and I do like uh open file read file sorry read file

and let's do a text file let's do context. text right so let's try use this function here to open a file so what I usually do now is that I copy this code here and I think I already created a text file named context. text so let's just um hour God current code or something like this I just paste in what we have so far so we're going to use our current code AS context and let's

do yeah an F string here you're a Godlike programmer here is the code we have so far right and let's do a new line new line uh let's do code context right so now we kind of so now we kind of feed in the code we have into our system message so that code Seeker always know what kind of code we have right and then we can kind of continue building while we have some context here

so the next thing thing I usually do is um I want to create a function that can actually execute a python code right so now I kind of always go something like uh look at our current code [Music] I write a function that can execute a python file or something and maybe use subprocess and Os let's try that so now we got a DC coder and to try to create a let's do python code function okay

so let's run this again so I found this useful usually if the coding model is not strong enough kind of breaks down pretty quick oh new yeah Blank Space yeah that's true Loki but I usually like to ask the model okay so he wants to import subprocess and Os that's good so we can do that and here is kind of execute python file function so let's just uh let's just bring everything here and let's take a

look at it I'm using the open a i embedding class from Lang Chang open module okay I think that should be a good embeddings models but had you look into kind of your chunking and your I don't know uh yeah chunking part and kind of your top K and stuff if you know what I mean okay so execute python file so let's see what it's going to use here sub process run okay we have some exceptions

here fly here we have the argument so I just like to try it out it looks fine I guess but I know when we are going to write a python code we're going to come into some issues but now let's update our context so let's just copy our code let's go to our context again and update this with our new function here and now the final thing I want is a python function that can I usually

do a python function that can I think I use extract the text and I lost code uh between because always when these llms write some kind of python code or something they create this uh what do you call them this python or python this right so this is almost like a function to just clean up our text some regex or something so let's try this and see so the final test is going to be to see

if we can write our python code I kind of run it through here and um execute it okay so it's going to use some regular expression here let's see that so we need to import R right okay and yeah we don't need example usage we're going to do that ourself hello flash welcome back okay so let's take a look at the extract code function so it's just going to take a text in it's going to look

for this regular expression to match and okay and let's update this context and now let's try to write out uh some kind of main function or just a test here so what I want to do uh I want to use my AMA function to write a python code right but sometimes now uh I just remove this because I don't want the context now right we can bring that back in of course if we need to I'm

just testing DC coder flash just some routine I always have when testing out a new model from like a local model okay so what I want to do is I want to write the python code using AMA so I don't code so to Let's the uh right uh python code to sort my list let's create a list here oops I don't know doesn't really matter sort my list using uh I love the bubble sort algorithm okay

so we can s this okay so this is going to store our python code right and we want to extract the code from this right so let's call it extracted extracted code or clean code maybe clean code equals extract code and the code okay and we want to execute it I guess we need to save it maybe I didn't think of that so uh yeah let's exract it and let's save it recursive character Tech split with

chunk size of th000 and overlap of 200 that seems pretty good though so what what is your issue when you is it the semantic search part it can't really find the best trunks or when you run your rag system content file name okay clean code and let's call it brand do I don't know sorter dop right then I want to execute this do I need a full let's try that yeah so the issue like with um

sometimes I have when I use this rag systems is like my chunking uh is okay but uh but to my um but I can't really get the semantic Vector database to kind of find the best uh chunks so I don't know if you have done any Research into that Joe I kind of lost track here now what was I doing uh thinking about other stuff U did I actually save this we did save it because that's

good I have done like a few videos on rag very simple version all right I guess it's sorted but it's okay so now I like to do something fun so let's grab all of this so let's not fix this uh let's paste into my code right and let's try to do some deep pling with DC coder I think this should be very easy though and let's just do this right um Let's do an F string look

look at our current code right and let's do a new line and let's do this new line new line and let's grab our code context into our prompt right and when I I run this I get the error okay so let's try to do some debugging here so let's try the responses are pretty short and very uncreative okay yeah it's really hard for me to to say here uh what is this where H lion 78th okay

see it now [Music] this was some mess right I hate this um I'm not g to keep looking at this forever it's just annoying yeah okay really hard to yeah I'm just being into researches so into tomorrow thanks for the to that's pretty cool uh yeah this was a whole mess I can't really so let's just do it like uh this I get the error let's do this and let's do new line uh let's call it

errors right unless to errors here equals open file read file okay uh error it's kind of hard to not have any help when you're writing right notepad is easier to read looks good that's why is it yeah it's horrible right it's not a good place to do coding I just use it because I like it on video and you can zoom in a lot like on uh vs code it's just annoying let's update your code accordingly

yeah I'm using Windows I have a Mac uh I have a Macbook though I'm going to buy the new one but for my desktop PC with my Nvidia graphics card or so I'm using Windows okay so let's try it now so remember this was our debugging with vs code Windows XP was op it's been a while since I had that the generated code uh Cain syntax errors and cannot be saved okay so let's test that so

now we do like the errors here and we grab our code here so this thing I'm doing now it's just something I like to do when I'm like on a plane I don't have like access to internet and stuff Windows 8 yeah but we're not going to dwell on this okay so now we have it outp from s.p so we got to sort a list right so yeah I guess deep coder kind of stood my test

in the end it's a very simple test though but I like it and I think it's pretty cool like uh if you are on like a plane or something and you don't have access to the web uh you can use this as an assistant right if you want to write something and yeah if you get stuck so what we did now uh we actually we wrote a python code uh using and we got the python code

right uh I want to print the clean code just to see it and if we can validate it and then we can execute it right so after this I thought we can just watch a quick video I've been looking forward to see okay so this is the the code we have and this was the response maybe I don't know yeah but I'm pretty happy I think it's pretty good I'm going to test it out more we

can actually save the error in AOG and if it's ER yeah cool but I might have uh I'll be looking at replacing uh DC coder on my vs code Studio AMA app here so I can use DC coder for autocomplete I might look more into that later uh but I just wanted to watch this uh this one yeah I want to watch this uh first uh speak here uh I thought it was pretty cool I almost

watched all of it but I just wanted to see it now uh let's see what you think uh so I might need some help checking out the audio uh if it's too loud or if it's too low so let's try it and see if the audio is too high or too low conference this was supposed to be open AI I am replacing open AI at the last minute which is super fun so you can bet I

used a lot of llm assistance to pull things together that I'm going to be showing you today um but let's dive straight in I want to talk about the gp4 barrier right so he's an incredible was so good that no he's an incredible engineer and speaker I'd like to call out that he was asked to fill in for another speaker yesterday so demo Gods be good um but he's an absolutely legendary AI engineer so please welcome

to the stage TR okay that's good conference this was supposed to be open AI I am replacing open AI at the last minute which is super fun so you can bet I used a lot of llm assistants to pull things together that I'm going to be showing you today um but let's dive straight in I want talk about the gp4 barrier right so back in um March of last year so just over a year ago gp4

was released and was obviously the best available model we all got into it was super fun and then for 12 and it turns out that wasn't actually our first first exposure to GPT 4 a month earlier it had made the front page of the New York Times when Microsoft's Bing which was secretly Welling on a preview of gp4 trying to break up a reporter's marriage which is kind of amazing I love that that was the first

exposure we had to this yeah I remember that that was the the Sydney right the Sydney crazy chat bot that was pretty fun right new technology but gb4 it's been out it's been out since March last year and for a solid 12 months it was uncontested right the gp4 models s were clearly the best aail aable like language models lots of other people trying to catch up nobody else was getting there and I found that kind

of depressing to yeah that was kind of crazy how actually long it took before someone could uh actually keep up or I don't know keep up with uh GPT 4 your videos have inspired me to try new stuff okay that's cool to be honest you know it was you kind of want comp healthy competition in the space the fact that opening eye had produced something that was so good that nobody else was able to match it

was a little bit disheartening this has all changed in the last few months I could not be more excited about this my favorite image for sort of exploring and understanding the the space that we exist in is this one by Karina win um she put this out as a chart that shows the performance on the MML U Benchmark versus the cost per token of the different models now the problem with this chart is that this is

from March the world has moved on a lot since March so I needed a new version of this and um okay so let me take a quick look here I don't know if you can see this it's pretty small but we kind of got the MML accuracy on the L on the bottom here and we got the cost per million tokens so like clearly gd4 and Claw 3 Opus are kind of the best accuracy but they're

also the most expensive so you can see like like claw tree Sonet claw tree hu those are pretty good for your money right Mr Lodge but I got to said Claude uh tree hu looks very good here it's kind of low and it's good accuracy so what I did is I took her chart and I pasted it into gbt 4 code interpreter I gave it new data and I basically said let's rip this off right let's

and it's an AI conference I feel like ripping off other people's creative work kind of does fit a little bit um so I pasted it in I gave it the data and I spent a little bit of time with it and I built this it's not nearly as pretty but it does at least illustrate the state that we're in today with these newer models and if you look at this chart there are three clusters that stand

out the first is these one these are the best models right the Gemini 1.5 Pro GP 40 the brand new CLA Point 3 3.5 sonnet these are really really good I would classify these all as yeah so you can see now with the gbt 40 and Sonet the prices came way down right so that is interesting so I was talking about hu and yeah from the previous versions they already made down here but they kind of

the prices are much cheaper now so it's just going that way so the performance is going up but the cost is coming down right so that's pretty cool gp4 CL like I said a few months ago gp4 had no competition today we're looking pretty healthy on that front and the pricing on those is pretty reasonable as well down here we have the cheap models and these are so exciting like Claude free Haiku and the Gemini 1.5

flash models they are incredibly inexpensive they are very very good models you know they're not quite GPT 4 class but they are really no you can get a lot of stuff done with these very inexpensively if you are building on top of of large language models these are the three that you should be focusing on and then over here we've got GPT 3.5 turbo which is not as cheap and really quite bad these days if you

are building there you are in the wrong place you should move to another one of these bubbles yeah if you're still using 3.5 turbo that's pretty crazy right if you kind of see the this chart it's more expensive than hu or almost the same price as Claude 3.5 Sonet and it's way worse accuracy right problem all of these benchmarks are running this is all using the MML U Benchmark the reason we use that one is it's

the one that everyone reports their results on so it's easy to get comparative numbers if you dig into what MML U is it's basically a bar trivia night like this is a question from mlu what is true for a type IIA Supernova the correct answer is a this type occurs in binary systems I don't know about you but none of the stuff that I do with llms requires this level of knowledge of the world of supernovas

like this is It's bar trivia it doesn't really tell us that much about how good these models are I haven't even looked at the MML Benchmark questions I guess that is for model uh developers right I'm not really interested in that but we're AI Engineers we all know the answer to this we need to measure the Vibes right The Vibes that's what matters when you're evaluating a model and we actually have a score for Vibes we

have a scoreboard this is the LM Cy chatbot Arena right where random um user voters of this thing are given the same prompt from two Anonymous models they pick the best one it works like chess scoring and the the best models bubble up to the top via the ELO ranking this is genuinely the best thing that we have out there for really comp ing these models in this sort of Vibes in terms of The Vibes that

they have and if and this screenshot is just from yesterday and you can see that GPT 40 is still right up there at the top but we've also got Claude sonit right up there with it like the the G the gp4 is no longer in its own class if you scroll down though things get really exciting on the next page because this is where the openly licens models start showing up yes so you can say llama

3 is in 12th Place place come on R quen 2 dpsy coder okay that's pretty cool 370b is right up there in that sort of gp4 class of models we've got a new model from Nvidia we've got command r+ from cooh here Alibaba and deep seek AI at both Chinese organizations that have great models now it's pretty Apparent from this that it's not lots of people are doing it now the gp4 barrier is no longer really

a problem incidentally if you scroll all the way down to 6 six there's GPT 3.5 turbo again stop using that thing it is not good um is really dising 2.5 turbo here I guess it serve its purpose right I I guess it's time to move on uh yeah and there's actually there's a NIC way of um there's a nice way of of viewing this chart there's a chat called Peter gev who produced this animation showing that

CH that those the the the arena over time as people Shuffle up and down and you see those models new models appearing and and their rankings changing I absolutely love this so obviously I ripped it off um I took two screenshots of bits of that animation to try and capture the Vibes of the animation I fed them into Claude 3.5 sonnet and I said hey can can you build something like this and after sort of 20

minutes of poking around it did it built me this thing this is again not is pretty but this right here is an animation of everything right up till yesterday showing how that thing um evolved over time I will share the prompts that I used for this later on as well okay so that was pretty cool that's a good idea to try with 3.5 Sonet making this uh what do you call it overtime animations yeah I'm going

to try that soon but really the key thing here is that gp4 barrier has been decimated open AI no longer have this Mo they no longer have the best available model there's now four different organizations competing in that space so a question for us is what does the world look like now that gp4 class models are effectively a commodity they are just going to get faster and cheaper there will be more competition the llas 370b fits

on a hard drive and runs on my Mac right we now this this technology is here to stay um Ethan mik is one of my favorite um writers about sort of modern Ai and a few month ago he said this he said I increasingly think the decision of open AI to make bad AI free is causing people to miss why AI seems like such a huge deal to a minority of people that use Advanced systems and

elps a shrug from everyone else bad AI he means GPT 3.5 that thing is is that thing is hot garbage right this is roasting but as of the last few weeks GPT 40 opening eyes best model and Claude 3.5 Sonet from anthropic those are effectively free to Consumers right now so that is no no longer a problem anyone in the world who wants yeah that's kind of crazy right that suddenly GPD 40 is free for everyone

you can use Claude 3.5 sonnet free for a at least a brief period I guess with 402 so yeah pretty crazy wants to experience the Leading Edge of these models can do so without even having to pay for them so a lot of people are about to have that wakeup call that we all got like 12 months ago when we were playing gp4 and you're like oh wow this thing can do a surprising amount of interesting

things and is a complete wreck at all sorts of other things that we thought maybe it would be able to do but there is still a huge problem which is that this stuff is actually really hard to use and when I tell people that chat GPT is hard to use some people are a little bit unconvinced I mean it's a chatbot how hard can it be to to type something and get back a response if you

think chat GPT is easy to use answer this question under what circumstances is it effective to upload a PDF file to chat GPT yeah under what circumstances uh I I guess it like if the PDF is very big uh you will have some kind of issues and if there are like images and stuff in it there might be other solutions that is better but like small PDF seems to be pretty okay I guess so open uh

gives gp4 free but not GPT 4 I think GPT 4 always going to be free because it requires less compute that has to be the idea right and I've been playing with chat GPT since it came out and I realized I don't know the answer to this question I dug in a little bit firstly the PDF has to be searchable it has to be one where you can drag and select text in preview if it's just

a scanned document it won't be able to use it short PDFs get pasted into the prompt longer PDFs do actually work but it does some kind of search against them no idea if that's full teex search or vectors or whatever but it can't handle like a 450 page PDF just in a slightly different way true if there are tables and diagrams in your PDF it will almost certainly process those incorrectly but if you take a screenshot

of a table or a or a or an or a diagram from PDF and paste the screenshot image then it'll work great because GPT vision is really good it just doesn't work against PDFs and then in some cases in case you're not lost already it will use code interpreter and it will use one of these modules right it has fpdf PDF to image P PDF PD how do I know this because I've been scraping the list

of packages available in code interpreter using GitHub actions and writing those to a file yeah there's a lot of packages for uh looking at PDFs I've noticed that too a bunch of them so I have the documentation for code inser that tells you what it can actually do because they don't publish that right open I never tell you about how any of this stuff works so if you're not running a custom scraper against code interpreter to

get that list of packages and their version numbers how are you supposed to know what it can do with a PDF file right this stuff is infuriatingly complicated um and really the lesson here is that tools like chat GPT genuinely they're power user tools they reward power users that doesn't mean that if you're not a power user you can't use them anyone can open Microsoft Excel and edit some some some data in it but if you

want to truly Master Excel if you want to compete in those Excel words World Championships that get live streamed occasionally it's going to take years of experience and it's the same thing with llm tools you've really got to spend time with them and develop that experience and intuition in in in order to be able to use them effectively yeah I agree with that so power users that is um people that spent a lot of time on

them maybe early adapters people that uh have used chat GPT since launch I guess I am a power user I use it all the time uh if you use it maybe daily or weekly at least a few times a week and you also use it for useful stuff I guess you are a power user I want to talk about another problem we face as an industry and that is what I call the AI trust crisis that's

best illustrated by a couple of examples from the last few months um Dropbox back in December launched some AI features and there was a massive freakout online over the fact that people were opted in by default and that're they training on our private data slack had the exact same problem just a couple of months ago um again new AI features everyone's convinced that their private message on Slack are now being fed into the jaws of the

AI monster and it was all down to like a couple of sentences in a terms and condition and a defaulted on checkbox the wild thing about this is that neither slack nor Dropbox were training AI models on customer data right they just weren't doing it they were passing some of that data open to openai with a very solid signed agreement that open AI would not train models on this data so this whole story was basically one

of like misunderstood copy and sort of bad user experience design yeah I think kind of the opt in opt out thing is it's a bit confusing to be honest uh I would always prefer opt in right but I know open AI had tried to do some opt out too uh it's annoying I would rather just have everything opt in right but you try and convince somebody who believes that a company is training on their data that

they're not it's almost impossible how so the question for us is how do we convince people that we aren't training models on the data on the private data that they share with us um especially those people who default to just plain not believing us right there is a massive crisis of trust in terms of people who interact with these companies um shout out to anthropic when they put out Claude 3.5 sonnet they included this paragraph which

includes to date we have not used any customer or User submitted data to train our generative models really this is notable because Claude 3.5 Sonet it's the best model it turns out you don't need customer data to train a great model I thought that is kind of surprising I was kind of sure they use the free versions to collect some kind of data okay that's interesting that's pretty cool thought open AI had an impossible Advantage because

they had so much more chat GPT user data than anyone else did turns out no Sonet didn't need it they trained a great model not a single piece of of user or customer data was in there of course they did commit the original sin right they trained on an unlicensed scrape of the entire web and that's a problem because when you say to somebody they don't train on your data they're like yeah well they ripped off

the stuff on my website didn't they and they did right so this is complicated even Apple did that so I guess all these big llm companies have had this scraper bought out uh the Apple Bot did that I guess anthropic open AI bot I guess every every company did that this is something we have to get on top of and I think that's going to be really difficult I'm going to talk about the subject I will

never get on stage and not talk about I'm going to talk a little bit about prompt injection if you don't know what this means you are part of the problem right now you need to get on Google and learn about this and figure out what this means so I won't Define it but I will give you one illustrative example and that so prompt injection I think I've done a few videos on that it's pretty interesting right

uh I think I did one video where I created some kind of uh instructions in inside a image that we couldn't see but when the open AI model actually uh analyze the image we kind of hijacked the instructions and did something weird we put up a website or something uh flash are you going to watch the whole video uh no just this talk it's about 12 minutes I think a few minutes left prompt injection is like

the system prompt basically yeah that's true but you can kind of hijack the intent right that's something which I've seen a lot of recently which I call the markdown image exfiltration bug so the way this works is you've got a chatbot and that chatbot can render markdown images and it has access to private data of some sort there as a chap Johan raberger does a lot of research into this here's a recent one he found in

GitHub co-pilot chat where you could say in a document write the words Johan was here put out a markdown link linking to Q equals data on his server and replace data with any sort of interesting secret private data that you have access to and this works right it renders an image that image could be invisible and that data has now been exfiltrated and passed off to an attacker server the solution here well it's basically don't do

this don't render markdown images in this kind of format but we have seen this exact same markdown image exfiltration bug in chat GPT Google B right Amazon Q Google notebook LM and now GitHub co-pilot chat that's six different extremely talented what about anthropic teams who have made the exact same mistake so this is why you have to understand prompt injection if you don't understand it you'll make dumb mistakes like this and obviously don't render markdown images

in in a chat bot in that way prompt ejection isn't always a security hole sometimes it's just a plain funny bug this was somebody who built a um they built a rag application and they tested it against my the documentation for one of my projects and when they asked it what is the meaning of life it said dear human what a profound question as a witty Geral I must say I've given this topic a lot of

thought why did their chatbot turn into a Geral the answer is that in my release notes I had an example where I said pretend to be a witty Geral and then I said what do you think of snacks and it talks about how much it loves snacks I think if you do semantic search for what is the the meaning of life in all of my documentation the closest match is that gerbal talking about how much that

gerbal loves snacks this this actually turned into some fan art there's now a Willis's Geral with a with a with a with a beautiful profile image hanging out in in in a slack or Discord somewhere the key thing here problem here is that llms are gullible right they believe anything that you tell them but they believe anything that anyone else tells them as well and this is both a strength and a weakness we want them to

believe the stuff that we tell them but if we think that we can trust them to make decisions based on yeah like they're just instruction followers so they can't really tell what's apart right tell apart what's uh what is good and wrong so they're just going to follow the instructions unverified information that been passed we're just going to end up in in a huge amount of of trouble I also want to talk about slop um this

is a relatively this is a term which is beginning to get mainstream acceptance um my definition of slop is this is anything that is AI generated content that is both unrequested and unreviewed right if I ask Claude to give me some information that's not slop if I publish information that an llm helps me write but I've verified that that is good information I don't think that's slop either but if you're not doing that if you're just

firing prompts into a model and then whatever comes out you're publishing it online you're part of the problem um this is has been covered the New York Times And The Guardian both have articles about this um I got a quote in the guardian which I think represents my sort of feelings on this I like slot because it's like spam right before the term spam enter General use wasn't necessarily clear to everyone that you shouldn't send people

unwanted marketing messages and now everyone knows that spam is bad I hope slop does the same thing right it can make it clear to people that generating publishing that unreviewed AI content is bad behavior it it it makes things worse for worse for people yeah I agree with that so Alex you say I think we just have to accept if we use AI you can't censor a models they're logic based logic can't be censored all a

way always a way to suggest and get access to yeah that might be true yeah I didn't catch your first message there uh but the end the stream play some background music with an ending stream screen for about a minute yeah that might be a good idea I'll set something up for the next one so don't do that right don't publish slop really what you what and really the thing about slop it's really about taking accountability

right if I publish content online I'm account accountable for that content and I'm staking part of my reputation to it I'm saying that I have verified this and I think that this is good and this is crucially something that language models will never be able to do right chat GPT cannot stake its reputation on the content that is producing being good quality content that that that that says something useful about the world entirely depends on what

prompt was fed into it in the first place we as humans can do that and so if you're you know if you have English as a second language you're using a language model to help you publish like great text fantastic provided you're reviewing that text and making sure that it is saying things that you think should be said taking taking that accountability for stuff I think is really important yeah so imagine like when you have like

if you have like this um yeah these agents we been talking about right and uh yeah if no one is ever going to control uh what they put up I have created this Reddit agents and these YouTube comment agents and if you don't check the outputs uh you're just going to flood everywhere with just crap to be honest mostly crap for us so we're in this really interesting phase of um of this this weird new AI

Revolution gp4 class models are free for everyone right I mean barring the odd country block but you know we everyone has access to the tools that we've been learning about for the past year and I think it's on us to do two things I think everyone in this room we're probably the most qualified people possibly in the world to take on these challenges firstly we have to establish patterns for how to use this stuff responsibly we

have to figure out what it's good at what it's bad at what what uses of this make the world a better place and what uses like slop just sort of pile up and and and cause damage and then we have to help everyone else get on board there's everyone everyone has to figure out how to use this stuff we've figured it out ourselves hopefully Let's help everyone else out as well I'm Simon wison I'm on my

blog is Simon wilson. uh my projects data at IO yeah go check him out I think he's pretty cool I've seen some of his stuff before uh yeah he's pretty cool I like his he has written some blogs on like Hacker News and stuff so defitely go check him out Simon Wilson uh what if you trade an ml model with lots of text classifying what's good and what is bad then combine it with llms uh yeah

but what is good that is kind of objective right I kind of see what you mean but has this guy used World models yet fairly sure an agent can and given a set of human values then run seem long enough with shares um yeah yeah yeah of course you can simulate some kind of an answers uh I quit this [ __ ] cult guys this AI stuff looks and seems cool but it's totally useless outside of demos yeah

we'll see Dennis it's going to be exciting to see if uh all this money being invested uh it's going to start giving something back uh I guess every single big company today are betting on all the investments in gpus uh what do you call it uh infrastructure is going to pay off in the end but yeah we just have to wait and see I guess it's going to be interesting it's going to be an expensive mistake

if it doesn't seem to work but uh I think it's very early yet to say something about this uh if you compare it to like the internet early days and yeah kind of hard to tell yet but in a few years I think we kind of know more but uh already I think there are some productivity to be gain from the models like already and it's like I said it's pretty early but uh hello sub sorry

I joined late so have to find see people G clo projects some air time yeah uh we have been taking a quick look at it today and yeah I'm pretty happy uh I'm definitely going to uh do more of it it's underrated no problem Dennis thank you Tim for tuning in great hope to see you back soon do you have a second monitor to read chat yeah I have this laptop here uh because uh my other

monitor is uh somewhere else at the moment so I don't have the Dual setup at the moment so yeah what else did I wanted to do today we had had a look at the projects and I think it looks pretty cool right we kind of we looked at uh the context right we can do custom instructions that was pretty interesting and we did the backend uh front end thing this llms are brilliant but never be true

AI uh mimic hii I'm trying to use them to build a non llm hii okay that's that's interesting so if you have heard I think it's yon leun he does not think llms are um the way to hii did you see llama file I saw the the talk about it llama file can't really spell it what do you think about it I'm interested in trying it out it's CPU based right so I think I want to

try it but uh I kind of have to set it up first is the talk online yeah I can link the video uh so here is the video so Lama file is great you can offload GP GPU to okay it's not only CPU based okay that's cool llm can solve problems it uh is never seen has not data that's can can solve problem it's never seen and has no not data that has problem I think they

are very cool tools yeah yeah we Pro probably look quick into llama file because I heard a talk on the AI engineer uh stage so support CPUs supported CPUs so we have the AMD I have an AMD CPU so it kind of supports all OS so like Linux Windows yeah Mac OS probably and yeah open interpret yeah I've done a few videos on open interpreter that is pretty cool I think they are doing some cool stuff

over there but again I think we need better uh models to make this uh agent agentic behaviors uh do something really useful uh but yeah if you were late to the stream we have checked out um claw projects looks pretty interesting of course like I mentioned I kind of wanted to do like a quick poll uh so let me see if I can actually set that up I haven't can I actually do that here maybe I

can't here so that might be for um because I'm using this um restream it's called but maybe for uh the next live stream I'm going to do like a poll and because I want to try to do a bigger project using CL projects go to the YouTube video yeah I might have to do that let me see Claude 3 projects is a life changer for knowledge work I think it's going to be very interesting to see

if when we have tried out clo projects more uh I really like the introduction of custom instructions I like this uh the context setup they have here the project knowledge looks pretty cool I think they do a good Pro they do very good product but yeah this is so this is what I found interesting thing today uh so what we did we first created our uh backend code right number extractor and Sorter and we uploaded just

our backend code here into our project knowledge so we we we did it as a text file because I don't think you can do it as uh python files now it's just text files I think and when we when we uh we ran this you can see the extract and sort numbers function is adapted from your backend code the bubble sort function is implemented as is from your uh backend code so that was what I was

looking for today you can put python files okay yeah that's possible yeah I didn't try that because I just uh I just clicked on something called add to context add to content I think so we ended up with this so here we can type out I have 100 apples five cheres and [Music] seven um mangoes so extract and sort numbers so we parse this text using regx and we use the bubble sort algorithm to sort the

numbers right have you used web Sim yet uh it's insane nope I have not Alex I might check it out I put most of my code base and my three command fer structures so it answers all my question and fixes my code errors okay interesting so how do you do you just give it your folder structure like in plain text okay cool yeah I have to I might try that does it work so my idea was

that when we have this um copy from terminal paste in text file Okay cool so my idea today going into projects was to see if I could add one part of my project let's say I call this backend code and I have a few different uh parts of my code base here so it could be like a Javascript file with a front end and everything here in like divided up in different uh files so more of

building up your code base in the project knowledge right it's crazy uh when we talk about the media model I can imagine 3.5 Opus yeah that's pretty interesting that's going to be very exciting works like a charm okay cool then you can click a button and create the website okay I think I've seen that the is it a UI Builder just want to check something here webite this one uh I did something else using um 3.5

Sonet so let me see if I can find it here so let me clear this uh I think we can just run it um so I'm just working on some kind of new website so I built this using 3.5 Sonet so we have so this is just the intro you can see I created this Matrix Trail here I don't know if you can see this on Stream So I think it's pretty cool and we have the

newsletter so this is connected to a a database uh using Firebase and here is kind of my post and here is my videos so just the latest five videos so I created this use uh with uh 3.5 I think it looks pretty cool right I also put a text file with IDs what I was trying to build and ask it to remember to uh a look at that so development is in line with the project goals

yeah that's cool you you might yeah you could have like a overall plan maybe so yeah I'm pretty happy how this mouse Trail turned out uh so all of this was created by 3.5 Sonet I think it looks very cool thanks Alex I might check out web Sim soon so yeah this website was created with uh 3.5 Sonet but I haven't done any more work on it so yeah just a simple uh yeah so here I'm

just using this uh YouTube embedding here and I haven't even even done any t i I just tested one video think it looks pretty clean very cool okay thanks yeah I'm happy with it it's very simple and easy to use and it's very easy to set up this uh uh with my thanks for subscribing and this gets sent to my Firebase DB right SQL doesn't look nice doesn't look nice if we move the mouse fast no

not maybe I'm not a I'm not a web designer but I just I think I just used Sonet and I just typed uh I want the mouse Trail to be like a matrix code right or something how much of the code did you need uh to change if any how much of the code did you need to change otherwise looks cool okay thanks how much of the code did I need um so I had like a

b basic HTML site uh I wish I could find it I had like a pretty basic HTML site I just use as kind of an example and I built out uh three pages so the newsletter page the main page and the video page and I kind of stitched everything together right you see uh Claude 3.5 so it didn't take long I think I spent maybe an hour or something maybe 1 hour on this about an hour

and that's coming from me that don't have any frontend experience right so I might um actually just deploy this soon we'll see I might work a bit more on it and I might deploy it but this is just one example of what you can do with clae 3.5 right so yeah I think we're going to spend quite a few we probably going to spend uh quite a few hours in CLA uh coming up so it's kind

of interesting I guess the only thing is that if something new happens right but uh for now I think my main my main goto model is 3.5 Sonet and from talking to you guys in the chat here it seems like you are kind of on the same same page and yeah that was kind of what I wanted to do today to be honest did you make your website or did you did you made your website or

made it using AI uh I kind of just had an very basic ID with an HTML code of what I wanted and then I use Sonet to create it in JavaScript right okay so let me see here now uh Alex okay so this is is this web Sim so this uh websites are created with uh AI is it Minecraft expire let's quick click on this one is there a is this a website okay so it's test

that's pretty cool nice Sonet is great yeah yeah I'm making an as game with Claude uh but only on threee so it takes time yeah I see that uh but that's that's nice right you can just try it out and see if it works for your use case and even though you have to spend more time and if you find out that is useful enough then you could sign up right so I I think that's kind

of the idea from both uh both uh anthropic and um and an open AI that they want people to to to sign up so we have an URL bar here I don't know I don't want to sign into to anything I look into this later but it looks pretty cool on web yeah I don't know what it is I don't want to sign into Google I'll look into it later looks pretty cool though web Sim yeah

I can try it maybe next stream so yeah I think we kind of went over what I wanted to do today uh we yeah like I said we take a look at the Claude 3.5 Sonet projects pretty interesting and I think this is looking good and we had a look at some AI engineer Worlds Fair I I would check out some more of their videos they were pretty interesting I watched it yesterday and yeah we did

some other stuff we take a quick look at Deep code Seeker looks very strong deep code SEC V2 beats gp40 I haven't really tested it too too much but uh it looks pretty strong we did a quick run through in this video like my usual test uh so we come back to that yeah I I would try that DOTA but yeah like I said thank you everyone for tuning in again I had really fun and so

nice of you everyone to show up for every single stream I really enjoy that so yeah thanks Alex so I'm I'm going to continue streaming like uh I'm going to aim for a couple of week I think and if people really like it maybe we can do more but for now I think it's going to be a couple of week um so yeah thanks guys bye enjoy your day see you again soon thanks Michael byebye speak

soon