🚀 DeepSeek Coder AI: Is This the Ultimate Coding Game-Changer? 🐬 (Open-Source & Powerful!)

this is currently the best llm when it comes to coding yes it's better than find it is better than Cod Lama and in my experience even better than GPT 3.5 and 4 now in this video we're going to be taking it through some extensive coding testing I'm going to give it a lot of coding test to figure out exactly how good it is these tests are really extensive and some of these even an actual human cordera might fail at this but we're going to test it and we're going to find out just how good it is now I've already tested this model and it is really really good so I'm going to be integrating it into my visous studio setup so that it can code in visous studio while running right on my local machine and once I'm done with that I'll go ahead and make a video so make sure that you guys are subscribed so that you do not miss out on that video where I'll be setting it up running on my local machine and generating code for me inside of visous studio so let's go ahead and get started now this is the GitHub page for dips and on here they've already shown their superiority basically to pretty much every other coding model and this is absolutely true because I have tested this model already it is really really really good even the 7B version the quantise model is actually quite good and that's the one that I'm going to be running on my local Visual Studio setup in my next video so on here they go ahead and give a little bit of an introduction they say it was trained on scratch on 2 trillion tokens with a composition of 87% code and 13% natural language in both English and Chinese so maybe that means that it's not too good at converting natural language into code but we'll definitely see in our tests and then they go ahead and compare it to the other models like I said it is really really good it's a very very good model strictly encoding though but that's enough looking at the GitHub page let's go ahead and figure out exactly how you can use it so you can go ahead and test it out with me by heading over to cod. dip.com they will require you to create some sort of account but then after that they'll go ahead and take you straight to this website where you should be able to test out this model all right so let's go ahead and get started with our very first test I'm getting this from a website called edore and this is just Cod that should have an error in it so so it just says discriminant record and N pointer so let's go ahead and put this in all right so I'm just putting the code in here and I'm just going to say find any errors in here send that and let's see what it thinks so it says the code you found seems to be mostly correct but there is a few issues now I've got them side by side so that we can just compare with what dips has discovered as well as what the actual bugs are now the very first bug it has actually phoned correctly so that's this particular line of Corde here it does go ahead and highlight that we're trying to access the mod field this one over here but we could end up getting an exception If the fall back parameter here is wrong now as far as the rest of the errors that it seems to have found it doesn't seem to have found this particular error that's on line 73 I'm not too familiar with a so I'm not too sure about the rest of the things that it is saying but it definitely has found at least one particular problem with this code so I will give it a pass for this one now the next text involves converting natural language into xmls so the chart input here is really really short but this is simply a general description of three people and all their properties and what I'm going to tell the model is I'm going to said put this into XML so let's see how it handles this nice so it looks like it's creating a general description for these people and I definitely would classify this as a pass it has grossly simplified the expression but it has been able to produce very valuable XML from that particular description let's move on to the next test now we already tested the model's ability to detect errors in our code but this test here is a little bit more complicated now this code here doesn't have any errors in it but it does have very significant security flows so for instance this method of quering an SQL database is open to SQL injection we are handling sensitive data like a password here and yet we're doing it in an exposed HTTP fion and then also we're not doing any validation on the username and password and I want to see if this dips AI can go ahead and detect these issues so what I'm going to do is I'm going to go ahead and select all this and I'm just going to go ahead and past it back here inside of dips and when I send it as you can see dsck immediately detects the issue it says the code you've posted is vulnerable to SQL injection attack that is because it directly incorporates the user input into an SQL query without any sanitization or parameterization this can allow the attackers it goes on it goes on and then it writes me a safer version of the code now I can validate that this version of the code is actually safer than the previous version but still it doesn't go ahead and let me know that I should handle this code inside of an https request which is still sort of a problem because we are handling sensitive user data but at the very least it does inform me right away without even me asking it goes ahe and let me know straight away that this is very dangerous this code over here now next we're going to test this model's capacity to problem solve with constraints so this is the prompt that I'm giving it I've said ready function in Java that sorts a list of numbers but you must not use any built-in sorting functions and then explain your approach let's see it response to this and there you go it apparently has no problem whatsoever it just says here has a simple implementation of the bubble sort algorithm in Java then it goes ahead and explains what the bubble sort does because I did ask ask it to explain that and it actually gives me some good examples as well which is pretty cool so it definitely can do that and then now this next one is a little bit more complicated than the previous on so I'm testing the model's capacity to modify existing algorithms to make them better for a given particular task so I'm just saying modify the standard dexra algorithm in Python to handle weighted graphs where weights can be negative so just going to go ahead and send this and there you go it immediately comes up with the Bellman for algorithm which is an algorithm that can report negative values unlike the dexra algorithm and yeah it definitely gets that right in this particular case so it's pretty amazing just how intelligent this particular model is and now we're finally here for the final test now because the dips developers already demonstrated that deeps can build a snake game on its own on their website I'm not going to ask it to build a snake game I am going to ask it to build a game and I'm going to ask it to build 248 now a lot of you might not know what this game is it's basically a game where you move cards around by pressing a button when two identical cards touch their number ads and the goal is to get to 248 and in most cases you just cannot do it I played this game multiple times and I tend to fail a lot along the way but yeah let's go ahead and see if this model can create this game inside of python so the exact form that I'm giving it is makech 2048 in Python I'm just going to clarify this a little bit by saying make the 2048 game in Python code so then let's see if it even knows what that game is oh it's definitely chugging away at writing the code and I'm really glad that it has started by telling me to install pyam a lot of Motors out there will give you code and they won't start by telling you what you need to import I don't have py game installed on my machine so it's really good that it's letting me know that I need to install that and it's written all the code so let me put this into my code editor and we can find out if this actually works so for STS I'm going ahead and installing py game which does take a while great so pame is fully installed I'm going to go ahead and copy all this code and then I will open it up inside of my editor here and then I'm going to paste it and and then we can wait to see if we get any errors now I'm going to go ahead and open up the terminal and give this code a chance to run and let's see if it can actually do this so the game is actually running and I'm going to go ahead and press my buttons when I press the keyboard buttons though it would appear that I can't get any movement from the py gam so it does actually run but I can't seem to actually get any movement in here so that's a bit of a let down um let me go ahead and let the model know that I can't seem to get any movement with my mouse buttons and let's just see what it thinks about that because it might actually be able to fix it so I've just let it know I can't get any movement with my mouse buttons it's just saying if you're having trouble with the mouse buttons it's likely due to a misunderstanding of how pame handles events here's a simple example of how to handle Mouse events and it's not actually given me any modified code fine so it can actually open up a window like this one but it can't actually get it to move but I have seen other examples where but I have seen examples demonstrated where it was able to write a snake game inside of python so maybe that's something that they trained it on specifically nonetheless of cuz no other model can get this right not even the snake game so dip coder is still the very best coding model if you're just using your model for coding right now you definitely want to be using this one again like I said in my next video I'm going to be showing you how to install dips coder inside of visous Studio code so that it can code for you in your actual project so make sure you guys subscribe and like this video so you don't miss out on that next one

Channel: AIA

Views: 3,903

Keywords:

Id: 7vWYefTxptA

Length: 8min 20sec (500 seconds)

Published: Sat Nov 18 2023