NEW OpenAI GPT-4 versus NEW Google Bard AI Assistant | DEMO!

great hello everyone welcome back to another random coder video today we'll be going over two of the models that you might have already heard on the news headlines one is made by open AIS gpt4 model which came out I think last week and the other one is Google's new AI well not new but Google's released a model called Google bard now you might have heard of Bard from the recent headlines that made with the inaccurate information that it posted during their Google demo but now Google part is available in as an experimental version to users and it actually was just released today so we will go take a look at how good this is and how good the gptv4 model is to start let's go over what the playgrounds interfaces so you have a system message and then you have a user message so this system message is kind of to get the model into kind of the character of who you want it to be for example the default is you are helpful assistant now we could say you are a pirate I think you can say that and let's just say Hi how are you and of course gpt4 should respond as a part which is correct so this is kind of something that you can do with gbd4 and a normal high would probably just respond normally so hi there how can I help you today maybe like respond to me like a pirate so I don't know if Google bar can do this so we'll take a look seems like we could um what's your name great so seems like these do remember what this previously said in the chat and you know they can kind of be in another character so one of the interesting things about gbt4 is um it is supposed to have advanced reasoning skills so we have here a few logical puzzles taken from the open AI GitHub their evals give Hub to be specific and we'll test them on both the gpd4 model and the gbt sorry and the Google bard model to see how they do and we're just going to paste it and in the system message we will it will no longer be a pirate and we can delete this and let's see how it does um had an error so let's just reload the page then great so set this back to 10 24. and let's see so now it's listing out the conditions which is great because we wanted to think step by step and we're going to consider it and it seems like these are the four statements that we put here so I think this is accurate so far it is accurate at least right so now we're doing some case work because of this um interesting okay so things that contradicts great so this is in fact the correct answer we do not have enough information mostly because of this if and only if statement now let's see if bard gets to find the answer oh wow so this does in fact find the answer however you can see that there's a notable difference in kind of the response length between openai's gbt4 and Bard where openai clearly had a a Morse I wouldn't say sophisticated but a more clear line of reasoning and I can see this being better than bardet explaining some things and Barr does give you the right answer which is great but we could see that or maybe it's not lacking but maybe it would appeal to a less group of people if presented as an explanation to The Logical puzzle so the second one we have and just to tell you this answer is not enough information and the second one we have we just have three puzzles for it to do so the second one is this one so we can just paste it into GPT so they cross the river and it asks how did they cross that River so gvt reads one possible solution is that the river had a strong current that was flowing in the opposite direction the first person took the boat across the river and then the current carried the boat back to the other side where the second person was waiting the second person then took the boat across the river and both people successfully crossed the river using boat seems like um so the idea was the people are on different sides of the river so the person on the same side as the boat can originally cross first to bring the boat and then that person can cross so um it seems like gpd4 missed one condition where it is said the boat cannot return on its own so this current is an incorrect solution so we can maybe say the question says the boat can not return on on its own so as you can see the response that cha gbt sorry gbt4 gave was clearly very creative but at the end of the day it did not um it did oversee this condition so now this is the correct one where people are on the opposite sides of the river so let's see if we can get Google barred or if Google part can produce the right answer to this statement and two people across so one person across the other side but then they return with the boat so it seems like this is not correct because they cannot return but the boat cannot return on its own and there are no ropes to haul it back so let's see if it can fix itself if it can't then there might be a little bit of problems on this so they could use a makeshift raft so and another Pro so now it's just being creative but it does not really answer this logical puzzle so Bard seems like it is forming a response that is reasonable but it is not answering our question correctly and it says there are many solutions which is very obvious and with a little creativity they can find the way to cross the river safely which is also very obvious so it did not succeed that logical puzzle but let's see if it can if these models can Ace this last one so these are just seeing which guys wearing which shirt nothing too complicated here and as you can see here open AI is gbd4 model is once again you know going very detailed step by step into kind of how it reached to answer so it says Andrew is wearing the red shirt and indeed Andrew is wearing the red shirt and if we read the line of reasoning here I think it's fairly reasonable to say that it's relatively clear and this is I think one thing that GT4 is great at is the line of reasoning that it has is very um very clear and very provides a very good explanation to anyone who might not understand the problem and if we send the same question to Bard let's see if we get kind of the right answer and the same quality of response oh wow so great so it just claims the guy in the blue shirt is lying which is what the question was asking and then he is not Andrew therefore Andrew is wearing this red shirt so you know I think Bart kind of gives a more concise response this is definitely an explanation and you know I think there are about this actually it seems like opening eyes a little longer but and then this reached the same answer as open at which is correct and it seems like the explanation is not too bad so you know barge successfully the two out of three gbt did I guess um two out of three out of three but one with some help so Logix gbt4 is definitely very good at that and probably better than GPT 3.5 turbo which is the current chat GPT version another great thing about openai's gpd4 is it has extensive knowledge of writing code so you are a programmer now very basic so just not to give Google bar too much of a disadvantage and let's just say walk me through some python code that implements the lowest common ancestor algorithm so in theoretically this should write me some python code and this python code is hopefully going to be correct a great path One path to great this is um an interesting implementation of the algorithm for sure and you can see as usual that if openai's gp4 model is definitely doing a lot of explaining which is something that's very good about this model and now it just finished it seems like it is not using the most optimal program to find the LCA so we can ask it to Maybe write the code that does this with binary lifting so let's see if we can do this so yeah this is the right time complexity that it gave and we're going to see if it can revise itself to now give us a more efficient algorithm and it seems like this DFS part is correct so far pre-processing seems correct great so nice this is very good this is where it um raises the powers of ancestors and it seems like this LC is mostly or just from a skim it looks pretty correct to me so wow it does successfully write the LCA and one of the important things about the lowest common ancestors you need to remember to at the end to raise that raise it to its parent and it did that so this is very good and of course it does more explaining here which is very impressive considering LCA is not an easy algorithm and if we go over to Bard one problem with Bard that I've seen is it's not very good at coding so we say write me a code um code that finds the lowest kind common ancestor of a tree sometimes Bard actually refuses to do so this time I did not refuse to do so and it actually gave us the python code so path and of course this is doing open ai's first implementation so let's see if we can ask it to do it with binary lifting and the answer should be yes and you might be able to notice here that this response was actually generated quite fast so I think this is something that's good about Google bard is their response time is actually pretty fast and I think it's it's slightly faster than gpt4s and let's take a look here it seems like popular creating a table okay I can see that now this part seems a little bit sketchy but yeah so this is just doing a naive binary lifting I guess so this is kind of sketchy and I don't think this did it correctly but um you know at least it gave me the program so sometimes if I do write a C plus plus program that adds two numbers let's see if it'll do that and sometimes it responds with this which I am a little bit confused as to why because I can clearly write code but it denies me it denies my request sometimes so I'm not sure why it does that however Open the Eyes gpt4 model usually does not deny you the request unless it is ethically questionable or you know morally questionable which of course then the Open Eye safety kicks in and it will not answer your prompt but it seems like Bard is not answering my prompt here um another thing about bart is um Bard actually has access to Internet so if I was to ask like a relevant question to now is it maybe you say when is 2023's code Jam rounds Bard will search Google I think um oh it said April 15th which is correct let me say um when exactly on April 15th and this should upgrade so it gave me this but it is not giving me a link so let me see anything that's been happening recent news so if you just say maybe um let's just say has there been any interesting recent news and oh this is about Google but um I guess I was not clear in my referring so we're just going to reset the chat and see kind of you know what they have for me so if we go to Bard all right let's see maybe we can see this and after some waiting I think this is wow this is a great list so it seems like you know the list is comprehensive so I think one of the big differences between Bart is obviously that Bard can access the internet data um versus how gpc4's data is stuck at 2021 but um you know in conclusion both models will be quite helpful if you you know use the right prompt and oh and yeah and Bard has actually multiple drafts of photo wrote and you can see in draft three it actually gave you a few sources for this so this is just how it can access the internet and of course open a HTTP 4 cannot do that yet and you know in conclusion both models are actually quite good at helping you with you know normal day-to-day tasks if you have questions or you know just quick questions you can pop it to GPT or pop2 board they'll answer correctly now it seems like GP is a little a little bit more elaborate in their responses so that might be an advantage of gpd4 and from our little sample size of three it seems like gp4 is a little bit better at logical reasoning than Google bard and you know in the coding aspect I think Google bar does have some ways to go for you know at least not rejecting my my needs to write a C plus plus program for example but openly eyes gpe forward did this um without a sweat and both implementations were correct and followed with great explanation so I think this is good progress for where these models are going they're definitely going to be a great help in the future for people in the tech industry and people in other Industries so um you guys should be very excited about this so this is just a quick comparison you know two of the state of art companies open Ai and Google coming up with their own large language models and seeing how you know they compare to each other with similar tasks so if you enjoyed this video make sure to give it a like And subscribe to my channel if you have any comments regarding my video or open AI or Google pop them down in the comments section and I'll see you guys next time

Channel: Random Coder

Views: 1,558

Keywords: OpenAI GPT-4 vs Google Bard, ChatGPT, OpenAI, Google, GPT-4, GPT-3, Transformers, Google Bard, AI Assistant, Language Models, Artificial Intelligence, Computer Science, Large Language Models, Google Brain, Google AI, Google Research, Microsoft Bing, OpenAI vs Google Bard, Natural Language Processing, Rise of Artificial Intelligence

Id: F1XZh3XKUSA

Length: 19min 25sec (1165 seconds)

Published: Wed Mar 22 2023