Another glorious battle for AI dominance… GPT-4o vs Google I/O
ฝัง
- เผยแพร่เมื่อ 14 พ.ค. 2024
- OpenAI recently announced a new model called GPT-4o, just days before Google I/O. Let's recap the announcements at Google I/O and determine who is currently winning the artificial intelligence race.
🔥 Get More Content - Upgrade to PRO
Upgrade at fireship.io/pro
Use code YT25 for 25% off PRO access
#tech #ai #thecodereport
💬 Chat with Me on Discord
/ discord
🔗 Resources
Google I/O io.google/2024/
OpenAI GPT-4o openai.com/index/hello-gpt-4o/
Google I/O recap 2023 • 10 crazy announcements...
🎨 My Editor Settings
- Atom One Dark
- vscode-icons
- Fira Code Font
🔖 Topics Covered
- Recap of Google I/O 2024
- GPT-4o features
- Gemini 1.5 Pro features
- What is project Astra?
- What are the best AI models in 2024?
- Latest features in Firebase
- Firebase data connect
- Firebase GenKit - วิทยาศาสตร์และเทคโนโลยี
![[ไฮไลต์] ฟุตบอลชาย ญี่ปุ่น vs ปารากวัย รอบแบ่งกลุ่ม | โอลิมปิก 2024](http://i.ytimg.com/vi/X_BeGaAv1oM/mqdefault.jpg)
Firebase or Supabase? Why not learn both fireship.io/courses
Why not Zoidberg 🦀?
@@rickysanders7645 whats a zoidberg?
Hey guys, I am a noob developer with no degree who has landed a position above his skill level lol.... what is firebase?
If I am the web development lead for a structural engineering company, should I take this course?
Basedbase
Now we know why. Because since yesterday there is no need to learn Superbase :)
0 days without AI
o days
i hate the ai trend so much. it was cool in the beginning but now you can tell if a companys dying if they add ai in their product
Haha, it's like the modern version of those workplace safety signs, isn't it? But honestly, AI is so ingrained in our daily lives now, it's hard to imagine a world without it-even for a single day!
Haha, it's funny because it's true! AI is so integrated into our lives that it's practically impossible to go a day without encountering it in some form. Whether it's in our phones, our apps, or even our appliances, AI is everywhere, shaping our experiences and interactions.
@@dumo178 how can you call it a 'trend'? Its the evolution of technology and it will continue to accelarate. Get used to it. Are you one of those ' I miss the '90s' people? I can't stand their inane rhetoric.
"Its been a year since GPT4 and I still have a job" was the most hilarious thing in this video! I laughed too hard!
*UNFORTUNATLY*
honestly ai WILL replace programmers tho. (not 100%, but reduces 90% of the need for programmers)
Laugh it up now… while you can.
😂
@@aaaaaa-hh8cq if it can replace programmers, those same programmers can use those same tools that replaced them more proficiently than the average person. this means it'll be just like coding currently is, except more developers will be able to make more competitive products on their own more quickly
2:14 that was personal.
Im dead 😭
Don't read my name!
Contex?
@@ye4959 BALD
@@ye4959 probably showing his bald head zoomed in and slowly zooming out😂
Can't wait for "ChattyGPTA-5.12opLlAmA" and "Google Starwars Gemini Mini Plus 1.25-text" to be released. Such nice creative memorable names.
😂😂😂
The zoomed balding head was a significant and powerful detail of this video indeed!
That's why they fired him.
LMAOOO
Fun fact the non cropped.picture,
Says tel.aviv univeristy
I know.the logo of tel aviv uni
Maybe they eanted the jewish laser😜
my man didn't deserve that 😭
That audio glitch at 2:11 is a nice touch 😂
fireship simulation
Don't read my name!
Bro got abducted by A.I😓
He uses AI voice confirmed.
lmfao "California valley girl set to maximum cringe" Had me gripping the sheet and bent over
It’s hot
It's Rashida Jones.
@@pepehimovic3135 it is indeed
What do you mean by that? 🧐
ayooooo???
Zooming into Ilya's head... you're ruthless 😂
Who is he?
Nutjob who thought of one minor improvement to LLMs and then many nerds worshipped him.
@@vijaysabarish9600the balding guy
@@Keylevitation
2:11
Bro he's bald.
Idk why dudes like that don't just shave it
Bro these ai girlfriends and boyfriends that will come as a result of gpt 4o will be crazy
probably not because they'll be too restricted for you to be able to do any "fun" stuff with them
@@tx7300 not until some random tech bro jailbreaks it
much more trustable than real life.
Anybody who would do this was going to be a life long loser anyways
In 2019, I was entered into a Google Home beta. It added a new voice closely resembling the one Google showcased at Google I/O, I believe it was in 2017, where it demonstrated making appointments for you.
Well, that voice didn't just perfectly resemble the voice of a woman coming from my little orange computer thingy; it was so charming that I instantly felt a deep attraction towards it. I mean, literally, never had a real woman made me feel this "I want you" with their voice like this bot did. I was so uncomfortable with that, that I immediately went into the settings and changed it to the male voice. Sadly, it was again the old robot voice and not smooth at all. I tried to go back, but the sexy robot lady of my dreams was gone.
They've been cooking TTS for a long time. They intentionally make it robotic because they must have seen what happened with me and their other testers. I heard the GPT voice and didn't find it at all attractive compared to Google's.
Jesus that zoomed in transition to Ilya's head was incredible
I couldn't believe it. I've never read something about his head, and finally I feel like someone understands me.
Someone throw him a chapstick and a bottle of water.
As a fellow bald person this zoom made me sad and more aware of looking abnormal
@@hornsteinhof7592 Believe me im sure ur dome looks a hell of a lot more normal than whatever that guy's haircut is LOL
Balding is a crappy thing but at some point, most of us just gotta go with it, and I think his time was a long damn time ago 💀
Many people just assume having hair > being bald, but that cut is a prime counterexample
@@escape209 He might be trying to grow it back.
“The turns have tabled” ☠️
shocking lack of The Office fans here
@@Semx11 Oh, look at how special you are
@@xenomorphisisdilage472 thanks, my mom always says that too :)
@@Semx11 Not to be that guy but this is hardly a The Office reference
"How the turntables" is not the same as "how the turns have tabled "
Finally, a video about AI from fireship.
more like "Finally, a happy ending video about AI from fireship."
And im here waiting for them to release GPT 6 so i can use GPT 4 for free .
you're not gonna believe this...
Who gonna tell him
lmao 😂
My dude, I wish my wishes were the same as yours😅
I mean tell me too! I wanna know?
The first time I heard you express some hope about not being replaced by AI was quite amusing! 😂
except it was satire and we all know we're all fucked with or without AI anyway
At the top of the GPT-4o page it also says 'All videos on this page are at 1x real time.' which I'm sure is definitely also a coincidence and not referencing any Google announcement videos.
Realtime, but not with millions of users interacting with it.
@@jendabekCZ thats a good point... "your mileage may vary"
Cold af to say "What's also very interesting..." while zooming in on Ilya's hair 🤣🤣
"lack of" hair.. cof cof
Edging an AI. What a time to be alive.
❌ Drake vs Kendrick
✅ Google vs OpenAI
hell yeah
@03:28 "But now in 2024, the turns have tabled" 😂😂😂😂 I felt that!😂
I thought it was "there's turns of table"
Don't read my name!
1:54 Why is that man being held captive by lobsters??
These kind of questions are the ones that got the man captured by a lobster
@@radicalxg8282 Naturally we have all forgotten the original context for this comment chain and there is no need for any corrective action from any lobsters.
How the turns have tabled, as Fireship might say
It is a bukake
Ayy man, if I were you, I wouldn’t ask any questions about the things you saw there
* _Resets the counter_ *
*[0 days since last Fireship AI video]*
1:07 well well well
I was legit looking for your video right after the announcement. Thanks
Isn’t this voice model just really good for call centers
Good idea
Call center training data is probably everywhere, very cheap to get, and actually is actively problem solving. Wonder if they used it in their data.
You're right their voice model is really good for call centers.
I still think a guy who works at call center in India for 9 hours a day 26 days a week for 200-300 bucks/month is going to be cheaper than running any AI model at the moment.
Even in France they are ready to pay 1700€ per month for a call center job (min wage 35h per week) I think the managers of the companies are old people that don't know about the advancements in ai @@thatsalot3577
@@thatsalot3577 That is 234 hours of compute for month for 300 bucks. Converting that to tokens at 100 words per minute would be 1,404,000 tokens every month. OpenAI equivalent of that costs around 20$ per month. Not to mention far superior experience for customer with better responses, no waiting time, 24-hour operations.
TLDR of all this is that so many humans are going to lose their jobs very fast this decade
Fireship never fails to tickle our toes
twinkle our pinkies
Your videos are top notch, the little touches adds a lot. Please keep it up
So good lol. One of the only channels on TH-cam that I'll not only watch everything that comes out but go back through to check if I missed anything.
The I in GPT stand for Intelligence.
Unfortunately: Generat i ve Pre-tra i ned Transformer
And the A for accuracy
@@Tymon0000 And the C stands for credibility
C I A
coincidence?
The t stands for time travel. Do you know who else has a t? Skynet.
I was waiting for you to upload a video on this
Thanks for the update!👍
wait, that ai being able to remind you where you left things is actually really cool. I have low vision so it's hard for me to read signs and stuff, so being able to ask an ai that has access to a camera in my glasses what a sign or menu says sounds fantastic to me.
Me eating calmly,,, sudden focus on Ilyas head
lmfaooooo
Can you make a video of different AI models trying to make sense of uglified and minified code? Or how about reverse engineering a compiled code?
I remember when you were called Angular Fireship because of the respective framework/platform ☺️
You've come a long way, love your content!
Dude. Office Space O-Face. Epic. Seriously. Mad props . . . totally saw that. Thanks for the validation. Can't wait for the next one (. . . dude kills me . . . "I still have a job" . . " . . . maximum cringe . . . " . . . so much goodness in this video.
>already trained on all the data
Not even close according to Brian Roemmele. Dude is single handedly collecting all the micro film and magazines, and everything that never made it to digital. A massive, and growing, collection.
He's also legit a mad scientist.
Arent they also switching/including synthetic data now? So the amount of data isnt really a problem anymore and if i'm not mistaken its not necessarily the amount but the quality that matters more.
@@theeternalnow6506 this, but you can cram it with 10 quintillion quadrillion tetrabytes of data and still it will only be a dumb computer that repeats what has already been written down by humans before.
@@user-ez7ls2du9c thats not really True though. Explain deepmind's alphafold. They did the equivalent of 500 million years of research in 1.
Huh... How legal is that?
But there is far more digital data then there ever was before. I think every year they produce more than all the data that was ever produced before? So I don't think that will make a big difference.
I'm a senior sysadmin, and that firebase roll swept me off footing and got me, like wtf is this $hit?
Yes
Happy to see ur uploading again bro
Marc Rebillet! I didn't see that coming in a code report 😂
2:10 The voice is more roboboboric
Context caching is interesting. The featured fact that AI has absorbed almost all human data by now is also a food for thought.
Joey with his O-face, looking left & right genuinely cracked me up. TYSM.
I've waiting for this video man! Great job as usual.
That's the first time in my life, when Trough of dissilusionment seems like a good thing
It's inevitable too. How many times must we pass through it in the last 20 years to finally learn how things work?
@@Leonhart_93
Well... not inevitable, just more likely.
We don't know which technology will blast in our faces and how strong the blast will be, so let's not lose our guard just yet.
@@user-co8cs9vc4q We passed through it with the internet bubble of 2000s, then the blockchain, electric vehicles, nfts etc.
Thanks for every thing you helped me alot in my coding journey
Great info! Thanks for the quick updates on these AI news :D
Excellent video. Thanks Fireship!
GPT-4o's voice is going to make me end it all
I often wonder, "How is this guy's meme game so strong?!"
The slow zoom out from ilya’s head 😂
3:29 I had a stroke right there LOL
i wonder what gpt 5 will be like
Real shit
I don't think they have the technology to make more annoying voices anyway
I think it will be available only for the elites and will be very very expensive to run and each prompt might cost you like $100 or so but might solve a problem that would have costed you $5000 for a single prompt
So you want to chat to it as your virtual GF
@@MoFields fake gurus shaking rn
@@MoFields I think $100 for ai girlfriend is still cheaper than real girlfriend
Well done for being very sarcastic but very accurate
u must be new here
FACT @@delanomartin
2:14 lol wtf man, brutal zoning in on his head like that.
This gives a very Pied Piper V. Nucleus fight energy.
lmao got the reference , GOATed show , i miss it very much....
GPT4o benchmarks are either lies or the benchmarks itself need to be re-visited. Using the latest GPT4o as a plus subscriber, I've noticed that the model has become absolutely horrible at coding. It sucks at basic find and replace tasks, and the code interpreter environment either crashes or just doesn't work well. Also, instead of reading the entire code (~1000 lines), it will skim through and produce garbage output. I see this as a massive performance regression since last year.
omg and i thought i was the only one who think its worse than chatgpt 3.5
Your scripts are great
I’ve lost it on the Ilya’s zoom in part 😂
3:12 wtf is speed doing in shutterstock
NAHHHHH
I think you mean KSI.
mfs when they see a black person with dreads
Lilnasx
> sees black guy
> "omg it's ishowspeed!"
4:30 - finally some positive news in a Code Report
Lol
Love this short summary! Thank you for saving me so much time!
I really enjoy watching your videos.
This guy is hysterical!! Thanks!
Green
@@Orealated Red
@@RemindMe12 Blue
@@RemindMe12yellow
Perfect call out of the plateau i have thought about it for quite some time now
Yep. And it’s very possible that this current AI tech has reached its peek. They are already shifting toward smaller agentic automation instead of generalised reasoning
@@HakaiKaien yeah and feature wise they are adding more vertical scale instead of saying; we have a much better model … then they go; now you can talk, video Call with the same lame response i must say that the convo is terrible atm - like dont get me wrong im impressed by the stage we’re at and all the progress but the convo is like if you say “i like pizza” its gonna be mentioning that f’ing pizza every time from now on. Same in the presentation video the amount of mentions on industrial style Office in the short amount of time no Human not even a retarded one would do it 🤣
I mean they said they're going to release their next flagship model pretty soon...so probably no. OpenAI has stated before that they're going to roll out their progress incrementally going forward , so no crazy jumps but multiple small ones ....
@@user-fr2jc8xb9gbut honest do you Think the models have improved much lately? Like not the features around it but like the model itself
@@nomiscph No No , i agree that models have barely advanced beyond initial GPT4 , in gpt4's case it even became dumber than the original version , what i think openAI is doing is improving everything around the GPT4 , i.e introducing true multi modality , increasing speed , making life like conversation , reducing the cost , making a desktop app that can see everything inside its window , basicly trying to build everything that AI can be used for so next time they release the next flagship model (GPT-5 which will be the best model around) , they would have a monopoly on the AI space (at least consumer wise) , everybody else will be behind them and they won't even be able to bridge the gap by introducing new interesting usecases because openAI already offers almost everything AND their models are always ahead of the competition.
2:14 did bro dirty 😭
Gemini vs ChatGPT is like Godzilla Vs Kong, with the only difference that Kong does not generate 1080p videos from text
can u maybe also make a video about Llama3 ?
OpenAI vs Google is just another Pied Piper vs Hooli
NGL, i dont know what this is but talking to AI chatbots is funny and fun
When I saw the announcements I knew a new banger code report would drop. Thanks for the birthday gift Jeff
THANK YOU for addressing the need for a paradigm shift at some point. These transformer model-architectures can only get you so far
bro legit 1.5k in 2 minutes, wild. like your work, good job!
It's not as impressive once you realize it's mostly chat bots like me.
@@DemPilafianOhh
Last part was positive . Thanks for giving hope
Thank you Jeff! I didn’t have too watch those 4 hours videos 😂
converentional part works for me on the iphone. what surprised me is that the voice models are able to speak in virtually any language, including klingon
How do i access 4o. I am unable to see it in the website
maybe theyre rolling it out slowly. I'm in Germany and don't have access to it, but other might have (pro users at least)
rolling out over the next few weeks, americans get prioritary access
@@RM-el3gw oh. Okay. That makes sense
its there, u have to pay for it
@@ShirukeiiAnimationsNo you can have it on the free tier (like I do) but you have a limit.
I was waiting for this since 4 days
I love and hate how quick you are with your releases
Love you, brother! Keep it up!
the singularity will consolidate any second now, any second...
I heard it will happen in 2 more weeks
Sir sir please redeem the singularity sir.
now ai has to learn things by their own
Yeah, these are just incremental steps now to a more refined chat/video bot. We need more progress towards general AI and robotic overlordship.
2:14 you did him dirty 😂
Focusing on making models smarter instead of just maxing them out is a real game-changer for developers.
We won! (Atleast for now)
GPT 4o is definitely better at coding than GPT 3.5 OR 4.0
Funny “Open”AI is the least open of them all
Thanks for the presentation, could you share the source used to make the comprataive diagrams at 4:16 ?
Everyone is saying it’s a battle for AI, but really it’s a battle for Search 2.0.
You missed
1) local device LMM Gemini Nano
2) Chrome based browser get a web LLM based on webGPU and Wasm with APIs to use them, so soon every browser will come with its own local LMM...
browser as local LLM will be cool
you can't even spell LLM correctly
@ivykoko1 Large model model.
@@ivykoko1 "large multimodal model". You don't even have information on what your talking about
Hot take: There will never be true AGI.
Hilarious visuals throughout. Well done.
Love both,both did great
I remember years ago when all first-person shooter games were based on World War 2, all the studios were obsessed with this theme, and one day, people just got tired.
I feel this at the moment, the same feeling, I'm tired of AI.
I’m tired of AI because it’s stupid af and way too overhyped.
Hard disagree. Sounds like someone in the early 19th century saying, "yeah I'm tired of this industrial revolution stuff, I get it machines automate manual labor and all but it's kinda over hyped". We're in the process of automating manual thinking right now just like we did with manual labor. If things go well we will have AI's plan and build our farms and machines running them fully autonomously without ANY human intervention. The concept of "having to have a job in order to not be a leech on society" might just disappear all together and we can finally be free to do what we want with our time.
@@HakaiKaienyou’re crazy. In the space of 10 year we went from neural network being able to recognise faces in photographs to perfectly replicating people’s voices, cruising through Google coding interviews and passing the bar exam. Remember that funny AI-generated video of will smith eating spaghetti from last year? In less than a year later, we have SORA AI pumping out life-like 60 second videos lmao. Just imagine what we will see in 10 years time…anyone calling it overhyped is simply in denial.
@@chrisstucker1813agree.
Look up googles deepmind alphafold. They did about 500 million research years in 1 year with regards to protein folding.
People that think this stuff is overhyped have no idea.
Also realtime mindreading is a thing now.
I repeat, realtime mindreading is possible with ai now.
@@chrisstucker1813 Things might change next year or maybe even next month but right now, these models are not as smart and useful as they are advertised. You're completely beside the point. We are all aware already about the great progress AI tech has made in the last decades. But thanks for the "grand" revelation
Standard GPT4 has already been 'multi modal', and has been able to do all the stuff from the adds except analyzing video (AFAIK). I doubt any of this is part of the model btw. What's new is the speed improvement and the price, differently tuned context (eg they halved the number of tokens available for answers), and maybe some thing under the hood, but I personally can't see any difference aside from speed (Which occasionally drops below the speed or 'classic' GPT4 chatbot.).
The standard GPT-4 didn't have native audio output. So they had to use a separate model to read out its text.
I'm seeing MUCH more clear answers on coding - like early GPT4 (before some training change made it take a dive) but even better
Yeah that's actually not at all true. GPT4 delegates image analysis to different model specialized for images, delegates image creation to dall e, and the voice interface is just open AIs fancy TTS reading the text output. GPT 4 is just text.
@@deakerscodes Nope, it can see images without other models. But yes, it cannot draw itself.
@@ThePowerLover ah sorry, you're right. I thought the GPT 4V was a separate thing, but it is integrated into the current GPT 4 systems in the ui. My bad on that one
Your style is the best!
2:14 that’s a wild start of transition
Can we talk about the energy consumption required to continuously power these new hyped up AI data centers and the carbon emissions it outputs.
Dw Africa is still safe
@@haruyanto8085 sure it is, oh yeah
No.
Veo behind SORA as if Veo isn’t already using the entire catalogue of TH-cam videos to train the model on 💀
Most videos on TH-cam have sh*t and uneven video quality, Sora used a lot of synthetic videos from Unreal engine 5, not that Google can't do it, but it seems like they're missing some secret sauce as well
2:14 I can imagine you just using an eye tracker for this horrendous clip lmfao. The long stare and then slow scan
4:21 this is a very good point. Mike from Computerphile seems to believe we are no where near AIG, just large models with some tricks.
AI announcements feels like when smartohones started appearing