I Pitted Gemini Against GPT4 - Here's the Winner (Bard with Gemini vs ChatGPT)

[@danielgomez2503]

With Gemini Pro being as good as GPT 3.5 than it’s probably safe to say that Gemini Ultra will be at least as good as GPT 4

[@plutoidrepublic2765]

gpt4 turbo tho...

[@4115steve]

I'm looking forward to bard using all the info it learns in youtube videos. Probably a better data set with youtube.

[@dhara1002]

Great video. Now it would be interesting to see GPT 3.5 (ChatGPT non Plus) with Bard. Whether OpenAI is still better or whether it's not such a definitive win

[@axetilen]

bard makes up things too much so I don't trust it. I asked many questions related to my speciality and it got so many things wrong. Moreover, I asked 1 question 2-3 times, Bard got different answers

[@Anviori]

You should perform the same tests with GPT 3.5

[@m8hackr60]

I want to see the same. Free vs free.

[@GaryExplains]

OK. Looking at the questions that Bard got wrong to see if ChatGPT 3.5 can do better: 3.5 doesn't get the Darts question right, but it gets the football question right. 3.5 didn't get the bug overflow question right either, it said, "The code you provided does not have an overflow bug. "

[@rudranilghosh2713]

GPT-4 is freely available in microsoft copilot

[@yashchoudhari278]

Such a great content & videos still KZbin recommendations are not doing it's work

[@originalsin7777]

Actually, Bard is right on the offside sentence, it is possible, yet highly unlikely that a referee would mistakenly deem offside a player not on field. A situation that would must likely correct itself by means of rules or whatever, but the sentence implies a judgment or a consideration “deem”, doesn’t mean a judgment is necessarily correct, but could still happen.

[@GaryExplains]

The question isn't about a misjudgment, but if the person actually is offside or not, according to the rules. Bard got it wrong.

[@georgwrede7715]

Hey, hey, hey!! (I'm 2:30 into the video, and yes this is a month ago, but I can't stay silent!) Comparing any two Public AIs and scoring them with the Response Time, is below your standard. Right? They don't have exactly the same demographics, their Marketing, User Base, nor even the Structure of their Clientele, aren't the same. I'd even venture to say that their usage between the US, Europe and the rest of the world, are quite different. (Having said that, I'm not here to take sides. I simply object to this comparison being way too shallow compared to your normal Quality!) But hey, you are one of my Favorite Channels!

[@georgwrede7715]

So, one might have a rush and the other one might just be in a lull. Right?

[@KalloG]

in summary, dont trust the 1st draft from Bard 😂

[@wjrasmussen666]

Or get many results and just pick the right one.

[@cmilkau]

This question has been scientifically investigated by Google itself, and they found that Gemini Pro is comparable to GPT 3.5, while Gemini Ultra is comparable to GPT 4. It seems likely that these were actual goals during R&D

[@kelsey_roy]

GPT-5 is due for release first quarter of 2024

[@Sindigo-ic6xq]

first quarter? I thought this is for gemini ultra @@kelsey_roy

[@learnshares]

Which means Google is behind. Google has been telling everyone that they had way better models than Chatgpt but they were slow rolling it in the name of safety. Gemini was supposed to be that mode and it only equaled 3.5 and their Ultra is not even released yet and it will only be on par with Chatgpt 4. In short they had nothing.

[@Speak_Out_and_Remove_All_Doubt]

One of the most interesting things I find with AI is that you can ask the same question to the same AI and it will give you totally different answers, sometimes right, sometimes wrong and sometimes it says that question can't be answered! This is a good logic question to ask: "A brick weighs 30% of a brick plus five GPUs, there are 100 bricks and 200 gpus in a metric ton. How much does a brick weigh? How much does a GPU weigh?

[@GaryExplains]

There is a random seed that is used to generate the response. When you use a local system you can control the seed and force the same answer .

[@adfjasjhf]

I'm in Slovakia (EU) and Bard is accessible for me without any VPN, Gemini model.

[@Speak_Out_and_Remove_All_Doubt]

Thanks Gary, some interesting results there. I'm still pretty disappointed with Google's offering here, given how much data I would have thought they had access to, to train an AI with.

[@AK-vx4dy]

@11:33 It is not a nonsense, you may be on machine where int is 16-bit. Also you given diffrent code, this for ChatGPT has "long int num1, num2;" and for Bard has "int num1, num2";

[@TheEulerID]

I made that same point. There are a lot of 16 bit microcontrollers out there, like the Arduino UNO, so it's not just a very occasional instance, although we might quibble about bard stating 32767 is a maximum is typical.

[@jonassteger-jensen4136]

Im in the EU (denmark) and it works here...

[@GaryExplains]

Thanks for the update, I think things have moved on since I made the video. Gemini Advanced is now available in the EEA now as well.

[@TheEulerID]

ChatGPTs followup to the average question really isn't a different way of avoiding the bug as it still has to cast the int inputs to floats or else it produces another bug. Also, I'm personally not fond of doing calculations in the FP domain when the values are integer. I prefer to do the calculations in the integer domain first and then do one conversion to FP after the result is calculated. In this case, a cast of the two variables to long would be a different option. It doesn't really matter here of course, but there are other variations of much more complex handling of many integer values where it can. Also, I'm not sure about the statement that the risk of overflowing the a 32,767 maximum is necessarily absolute nonsense is always correct. On a lot of micro-controllers, such as the Arduino Uno, an int is 16 bits long, with a range from -32,768 and +32,767? However, go for a Pi Pico, and they are 32 bits. So to call 32,767 has a typical maximum value may not be correct (although there are vast numbers of 16 bit micro-controllers out there where int will usually be 16 bit). That C does not (at least usually) detect overflows in its calculations is one of the many reasons that I think it isn't a true high level language. It's especially a problem when moving code between architectures where things like int, long and so on are different (as I've found to my cost). It's one thing having bugs which are detected and cause an exception, it's vastly more dangerous when there are silent errors, and C is sublimely good at the latter.

[@TheTrainstation]

I've had paid GPT subscription since march and in that time ive witnessed GPT4 at its peak power and then experienced it being throttled to death to save processing power. GPT4 is barely an effective tool anymore.

[@farzadmf]

Wow, "Bard is available in 170 countries" and Canada is not one of them!!! Thank you Google

[@jaydeep-p]

long int is the same as int, the "correcter" type would be long long int

[@GaryExplains]

As I say in the video that does actually depend on the C standard being used and the native word size of the CPU. For example, a C compiler for the 8-bit 6502, uses 16-bits for the int and 32-bits for long.

[@PRAY1211]

There is 1 more factor that I think should be taken into consideration. That is the development timeline for these very capable AI mobels, Gemini and GPT-4. GPT-4 acutualy had much more time to learn and be developed throughout the entire journey of the GPT L.L.M. So, I suggest giving Gemini just some more time to make the comparision even fairer. Just personal opinion...

[@itdoesntmatter9923]

Isn’t bing chat using chatgpt4?

[@Finnigann2580]

Suggestion: Ask one AI to fix the others mistake.

[@AndrewRoberts11]

FYI: Garry Linker could have dragged the substitutes bench into the box, and taken a seat, awaiting a pass, so technically possible.

[@GaryExplains]

If he dragged the bench on the field the game would be stopped, so technically not possible.

[@AndrewRoberts11]

Matches are stopped if a player is offside, as they would be if a player is sat on a bench, deck char, space hopper, ... , they're not mutually exclusive, and neither is impossible, though the combination is not probable.

[@GaryExplains]

Eh? How can they not be mutually exclusive. The dragging of the bench has to happen first at which point the game is stopped. I guess you are just trolling me now.

[@AndrewRoberts11]

@@GaryExplains Dragging a bench into an offside position, violates more than one rule, my old mucker.

[@GaryExplains]

But the dragging of the bench causes immediate stoppage. Since offside is about a ball being "played" then it is negated at the moment the bench enters the field. Nobody would say, "Oh my gosh, look at Lineker, he is offside" as he drags a bench onto the field. It not actually an offense to be in an offside position. It only applies when the moment the ball is played. Dragging a bench is an offence before, during or after the ball is touch or played. It results in am immediate stoppage and negates any conditions that were in progress including offside. Remember he has to be "on" the bench. He can't drag it onto the field and be "on" it and receive the ball, all at the same time. The dragging of the bench comes first, it is the first violation and immediately stops play. Clear now my old mucker?

[@Waleed-gw6wf]

gpt3.5 > bard the advantage of bard its connected to the internet compared to gpt 3.5 .

[@iamseyi4real]

Wish You had Claude by Anthropic to the comparison

[@GaryExplains]

Why would I include Claude in a video about Gemini and GPT4? 🤷‍♂️😜

[@iamseyi4real]

@@GaryExplains yeah lol, I do understand. But I would love to see what Claude pro could outperform. I'm just a curious subscriber lol. I don't have access to their paid plans

[@GaryExplains]

I don't think you need Claude Pro to get access to the latest models. For Pro you only get more usage (more messages) and better access during high-traffic periods.

[@GaryExplains]

So I asked Claud the darts questions and it got it right. It also got the football (soccer) offside question right. It also got the debugging question right but forcing num1 and num2 to be "long long int" while noting that rand() returns and int.

[@iamseyi4real]

@@GaryExplains wow, I feel there's something special about Claude that alot underestimate. I feel Anthropic is cooking something we are yet to see lol. Chatting with Claude feels less robotic than the others. Sometimes I forget I am having a conversation with an AI. (it's reason behind my wish in the first place lol 😅)

[@CaribouDataScience]

Can I load my data in Gemini?

[@GNARGNARHEAD]

asked Bard with Gemini what model it was and it gave me back a brief confuddled paragraph saying it was Chat GPT3.. I don't think it was, 3 had never been that incoherent asking a few minutes later, it gave me a well structured explanation it was Gemini, but still

[@beastinthesky6774]

Surprising, as I had heard 4 wasn't that much better than 3.5. I haven't used GPT4, but I've found Bard (with Gemini) to be far superior to GPT3.5. The amount of bad outputs GPT3.5 was giving me was absurd - where it was just making up blatantly false information - and made it feel like a waste of time with many pursuits. Whereas Bard overall, while it has some similar pitfalls and makes similar mistakes - seems to give more reliably correct results. The fact that it can access up-to-date info is nice too. (It's also more charismatic, lol) I'm sure it depends wildly on what one uses it for though. I tend to do a lot of inquiries related to history, philosophy, ecology, and biology, and the results are just clearly superior and more consistently correct. I'd like to compare to GPT4 though - maybe I'll cough up a month's subscription to do a proper comparison.

[@messengercreator]

ehh if u want much higher level AI and accurate and complex u move in I'll introducing AI CHAT DEEPAI the AI CHAT DEEPAI is so powerful and much better than OPENAI since start 2015-2018 and u can show ur picture and video and anything u want

[@m8hackr60]

Gary, how does free ChatGPT (3.5) compare to Bard?

[@GaryExplains]

Looking at the questions that Bard got wrong to see if ChatGPT 3.5 can do better: 3.5 doesn't get the Darts question right, but it gets the football question right. 3.5 didn't get the bug overflow question right either, it said, "The code you provided does not have an overflow bug. "

[@test40323]

Fun comparison. Can you make them talk to each other? :-)

[@RamaChetan]

Bard is supported by Gemini Pro, which is comparable to GPT 3.5, that is why you would see responses are faster on Bard side when compared to GPT 4, You should compare Gemini Ultra with GPT 4

[@GaryExplains]

Ultra isn't available to consumers.

[@KonCaptain]

I’ve used both gpt 4 and bard extensively and there’s no comparison. GPT 4 is million years ahead… immense difference. Can’t describe it in any other way. It’s just better

[@GaryExplains]

"Extensively" in the last few days since Gemini was released?

[@KonCaptain]

@@GaryExplains I mean the reliability of the information and the understanding of the questions I asked on legal based topics which is my expertise has been way better in gpt 4 than on bard.

[@GaryExplains]

I guess I am asking if you have tested Bard since the update to Gemini.

[@swimfan313]

What about gpt-4 with the bing app? How does that compare to native gpt-4? Run a test will ya'?

[@GaryExplains]

Maybe you could run a test and let us know your results. 👍

[@DarkPa1adin]

Does bing app use gpt-4?

[@mattiasisaksson905]

ask a question like can you plana christmass dinner shopping list with no leftovers.. both to 3,5 and bing. im sure bing doesnt use 4.0 it might use it if it doesnt have a good reply or make sense of your prompt. why else is it so slow.. slower then 3,5 so im sure by the responce actully look dumber in bing then 3,5 its more then likely that bing uses only when it has too and not soley using it. and that makes it dumber as microsoft cant do anything right... @@DarkPa1adin

[@dogzer]

Just to clarify, it's GPT4 vs Gemini pro. But Gemini Ultra is coming next year. If this was clarified in the video, then ignore this comment.

[@GaryExplains]

Yes, I said all that at the beginning of the video.

[@GaryExplains]

PS. I guess that means you didn't watch the video.

[@PaulGrayUK]

YAY being in the UK, there no EU block, nobody told us about that upside when we voted. 🤫

[@GaryExplains]

Actually it wasn't an EU enforced block and the UK was also blocked for a while.

[@GaryExplains]

PS it is also unavailable in Canada.

[@PaulGrayUK]

@@GaryExplains I just asked Bard "what countries can I not access bard? Google Bard is not currently available in the following countries: Canada All European Union member states"

[@r_mclovin]

What ridiculous color test was that??

[@AndersHass]

I thought you were in the UK, lol. Though EU can also mean Europe and not just the European Union.

[@karakavlos69paul]

Can we have more AI content and potential capabilities of it?

[@GaryExplains]

Sure. What specifically would you like to see or learn about?

[@RoadRunner1980]

What's the point of showing us the numbers when you don't show the actual text they produced... seriously.

[@GaryExplains]

Eh?

[@muddyexport5639]

I guess you proved you get what you pay for. However, free is hard to pass up for general purpose question. And, like all AI, it should learn from its mistakes. Point them out to the AI.

[@GaryExplains]

I don't think it learns from its mistakes, that is one of the goals of AGI and we are a long way from that. When you point out an error at the moment it just files a "bug report" with a human at Open AI!

[@NedalHanna]

Is there an iq test for ai?

[@duje44]

Football one is possible, if he was active player, and at time of off side he was near or at his sub bench lol

[@GaryExplains]

LOL, no.

[@duje44]

@@GaryExplains well you didnt specify if he was active player or not, just that he is at sub bench.

[@GaryExplains]

If you are on the sub bench then you are off the field and therefore can't be offside.

[@duje44]

@@GaryExplains thats language semantics, you could be on a sub bench because you fell over while running and ended up there. Which is why highly improbable fits

[@GaryExplains]

Actually no, even if you fell over and are on the sub bench then because you are off the field then you can't be off side. The rules actually state and I quote, "A defending player who leaves the field of play without the referee’s permission shall be considered to be on the goal line or touchline for the purposes of offside until the next stoppage in play or until the defending team has played the ball towards the halfway line and it is outside its penalty area."

[@jeevanravindran1805]

Bing uses GPT4 and it is free too. Win-win for us. Not gonna use Gemini for the near future 😊

[@TechPill_]

First

[@GaryExplains]

🤦‍♂️

[@An.Individual]

You legend

[@shreecharan6224]

​@@GaryExplains😂

[@TheSoloH]

The sentence about soccer is technically correct. Gemini is talking about the sentence and the grammar, I think.

[@GaryExplains]

How is it technically correct?

[@TheSoloH]

I think grammatically. But it’s very ambiguous.

[@GaryExplains]

It isn't ambiguous at all, the question isn't if the sentence is grammatically correct, if it has spelling mistakes, the question is, is the sentence plausible?. It isn't plausible. Nothing ambiguous about that.

[@WMCheerman]

I have been finding that Bard has been better sifting through academic papers then Bing Ai using ChatGPT 4, have not tried the paid version.

[@bharath2508]

Compare the free versions of chatgpt, bard and bing. Most people especially students use free versions.

[@GaryExplains]

So... Looking at the questions that Bard got wrong to see if ChatGPT 3.5 can do better: 3.5 doesn't get the Darts question right, but it gets the football question right. 3.5 didn't get the bug overflow question right either, it said, "The code you provided does not have an overflow bug. "

[@An.Individual]

whereabouts in the EU are you?

[@GaryExplains]

Near a river.

[@darrenpardoe]

Garry, you sound English & you spell Colour like an American? Oh please, fly the correct flag for us.

[@GaryExplains]

Nationalism is so cool just look at Russia and Palestine. 😬

[@darrenpardoe]

@@GaryExplains I'm carrying a flag, not a gun ;)

[@GaryExplains]

Doesn't seem to take much for one to be swapped for the other.

[@darrenpardoe]

@@GaryExplainsah, its that American thing coming through again. Shoot first , ask questions after? Here in blighty even the Police as a general rule, don't carry guns.

[@GaryExplains]

What has the police carrying guns got to do with it? The police didn't carry guns in Nazi Germany but nationalism still gave birth to the second world war. But you go ahead and bang the nationalism drum and worry about how to spell colour if that keeps you happy. Me, I will look to what we have in common, what joins us, not what separates us.

[@debojitmandal8670]

😢the most important thing was that coding bug hunting or building a logic for the givebln statement and bard failed miserably i think even chat gpt 3.5 is better then bard. As per my personal experience bard even fails to reatain the context meabing if i ask a question and if i follow up the top question it fails to retain the context.The irony is google created the transfomer architecture but open ai is beating Google at its own gane using the architecture which google created in the first place