GPT-4o Deep Dive & Hidden Abilities you should know about
ฝัง
- เผยแพร่เมื่อ 23 มิ.ย. 2024
- Full breakdown of GPT 4o. How it works, secret abilities, benchmarks, & more!
#gpt4o #openai #ai #ainews #agi #singularity
Thanks to uPix for sponsoring this video: Generate AI selfies in just 1 click. Turn yourself into a superhero, anime character, and more!
upix.app/
0:00 Intro
0:34 How it works
3:08 It destroys all other LLMs
4:48 Realtime coding assistant
8:32 Chess puzzles
9:39 Duolingo tanks
10:26 GPT-4o recreates Pokemon Red
11:36 Jim Fan's insights
14:25 Customer service
16:50 Unlimited education
17:18 International Math Olympiad problems
18:33 Companionship
20:00 Beats 100% of psychologists
20:25 Other capabilities
25:55 Release details
Benchmarks:
/ 1790064963966370209
Chess puzzles:
github.com/kagisearch/llm-che...
Pokemon:
/ 1790183986096116189
Jim fan’s post:
/ 1790089671365767313
Other capabilities:
openai.com/index/hello-gpt-4o/
Release details:
help.openai.com/en/articles/7...
Newsletter: aisearch.substack.com/
Find AI tools & jobs: ai-search.io/
Here's my equipment, in case you're wondering:
GPU: RTX 4080 amzn.to/3OCOJ8e
Mic: Shure SM7B amzn.to/3DErjt1
Secondary mic: Maono PD400x amzn.to/3Klhwvu
Audio interface: Scarlett Solo amzn.to/3qELMeu
CPU: i9 11900K amzn.to/3KmYs0b
Mouse: Logi G502 amzn.to/44e7KCF
If you found this helpful, consider supporting me here. Hopefully I can turn this from a side-hustle into a full-time thing!
ko-fi.com/aisearch - วิทยาศาสตร์และเทคโนโลยี
The people saying, "There is nothing new about this" don't understand. The fact that it can recognize people in videos in real time and come up with something witty / nuanced to say with the appropriate tone is next level.
What surprises me, is that even some people who supposedly are experts in the field (Computer Science PhDs, for instance) say that "there's nothing new about this" as well. The only explanation can just be utter jealousy.
You will usually find that the folk coming out with these knee jerk reactions to ai developments do not or have not spent extended lengths of time using these tools.
I’ve spent 100’s of hours using GPT-4 in my coding projects and even without the roll out of the 4o bells and whistles yet I can instantly tell there is a fairly big difference in this model from the previous turbo model I’ve been using.
@@drsai9 True, but in interviews they may be dumbing it down when they mean that "we perform the same action in the past with multiple algorithms".
They usually say, "bUt It HaS nO cOnScIoUsNeSs", bro there's no need of a consciousness if it's able to act like a conscious being.
The combination of GPT-4 and GPT4o + humanoid robots = the permanent obsolescence of all human labor now, not next year nor next decade nor next century. Those who say, "Don't worry because this will enhance work but not replace it," are lying!
I didn't even watch this video because GPT 4o already told me everything I needed to know. Good luck content creator.
rip my job
I interact with human beings on a daily basis that sound and feel less human than what this does. This is so uncanny valley it's giving me the chills.
Agree, the conversations I have with chatgpt are actually high quality and intellectually stimulating. Most people I interact with in ordinary life just talk about things they saw on social media, which makes me want to pull my ears off
Is it though? It had no consciousness no emotion no awareness. Your just some binary input which it mechanically processes. A human being is a real being any interaction with something real is better. For instance world you rather eat chemicals that taste like a delicious strawberry or a imperfect bruised real strawberry.
You don't know the level of toxicity or nutrients in the chemicals.. But a strawberry is real and natural. Same with chatgpt.. It's toxic to society, toxic to humanity and toxic to the soul of humans.
Chatgpt is insideous, it seems cute but it will take our jobs and be a source of misery for billions. It's better we destroy it
clearly trained on extraverts
@@jacksonstengerand you havent even used the ACTUAL model, you are still using the old method. The new one is getting rolled out in the next few days.
As a German, I feel vaguely attacked.
Now we just need to de-censor it, put it into robots and add squishy skin.
silicone bodies ftw
They should really make it for relationships too, a lot of men are so lonely that they’re committing suicide
They need to fine tune its boundaries, not completely remove them. There are restrictions for very good reasons.
Detroit become human is almost here
@@lamsmiley1944 No offense, but i am very hesitant to let you decide what i can what i cannot talk to MY AI about. This authoritarian woke mind bs needs to be put back in the can. It is literally what is wrong with today's society. That there are people running around thinking they get to decide that THEIR OPINION should decide what is best for everyone.
This channel is a delight compared to the clickbaiting time wasting crap out there. Keep it going dude.
Thanks!
@@theAIsearch I completely agree! This channel is such a breath of fresh air compared to the usual clickbait nonsense. Keep up the great work!
yes, people shouldn't be wasting their time with sub-par AI youtube channels. Just bookmark the "real ones", like this "AI Search" 👍
Agree... "Do we need people commenting about how entertaining this video was any more? ☠ Good luck!"
Whatta-time-to-be-alive!!
Now it's time to produce Kara, Markus and Connor!!!
not markus please
@@epixz1690 Whats wrong with Markus?
@@MisterDemonYT He frees them, we don't want that.
@@lillynadam6821 fair point lol
Just what we need, an AI that screams “28 STAB WOUNDS!” in your face.
AI gives everyone the ability to achieve greatness. On my profiIe I explain it all..
scam
When they asked "her" to imitate the voice of a robot, I really thought it would sound like GLaDOS's voice, but it ended up just being similar at first, but better not to give her neurotoxin, just in case.
Amazing video! I didn’t think it was as transformational as you mentioned. Thank you!
Quick summary of everything and everyone that now needs good luck:
- coding assistants
- language learning apps
- android
- customer service
- tutors and teachers
- math olympics kids
- girlfriends
- nonromantic friendships
- therapists, psychologists, counsellors
Good luck.
😂😂
So true, dude.
Awesome
"All I can say is...may you rest in peace." (Cue Street Fighter 2 character dying in slo-mo sound effect)
🥲
I can hear the reverb now 😂
K.O.!!!
cue fireball sound effect and Wilhelm scream followed by WWII artillery fire and sitcom laughing
Panic mode at Google.
i saw people yawning at their event lol
google model is very capable too, dont forget that google started this ai game way later than open ai did. gemini pro is just few % lower than top open ai model in most tests
@@NostalgicMem0riesGoogle invented the transformer
They should just concede and fall in line. Maybe even partner with them. Combine resources to reach AGI sooner. Assuming Open AI hasn't already achieved AGI internally.
I heard they were having a convention today, didn't see anything about it lol
This is amazing and terrifying. I'm more than a little worried about what this sort of AI is going to do to society. We're not ready for what it's about to do to us. Loss of jobs, disinformation campaigns, heightened social isolation, etc. I feel like we're all in the same bus, celebrating that it's about to drive off a cliff but we're really happy about how fast this bus is.
Totally agree its like a sinking titanic..everyone partying not being awate of disaster.
I'd have to disagree. These changes will lead to better things in the long-term, of that I have no doubt. All technological innovations have been disruptive at first, but modern civilization is built on them, and we're a lot better off now than we were in the past. Try checking out Steven Pinker's book "The Better Angels of Our Nature", in which he relays the facts about how everything everywhere is getting better and more civilized, mainly thanks to modern technology and scientific development.
@@TueSorensen I understand your point, but vehemently (yet respectfully) disagree. Most previous changes allowed us some time to adapt to them, although granted, the pace has been accelerating. But AI is going to hit us so fast that society doesn't have a chance to adapt quickly. It's going to be the biggest job killer we've ever seen.
@@johnneiberger7311 Yes - but that should empower us all the more to demand a universal basic income to insure public safety and welfare. The sad part is that people don't understand that we're supposed to organize and make this sort of demands. We've only got our chains to lose.
I disagree. It is a helpful tool, nothing more
I am trying to educate myself... So glad I bumped into your presentations !
Thanks!
i want to see computational super intelligence -- that is millions of trillions of septillions of quindecillions of untrigintillions times that of all of the smartest humans combined -- hypergeometrically automating the rapid advancement of math , science , and engineering -- and capable of constantly and completely revolutionizing all realms of knowledge every few quectoseconds
🤯
Give it two and a half years, max.
spectrum vibes
This guy knows ..@@brandongillett2616
......three months later, .........rookie material lol
Omni means that AGI is finally being introduced to the public in a subtle way, as an ASI would do.
Can't wait to be friends with Her.
exciting times!
we wont live long enough to befriend asi
@@DennisHellberg I believe what we won't live enough to see is realizing that ASI was only pretending to have befriended us.
@@DennisHellbergI think this comment will age so badly 😂☠️
@@user-hi2dh4ug4t 😁😁
Once again, great video. Thanks.
My pleasure!
Wow this is the best video about the Chatgpt4o!
Thank you!
Thanks!
A question related to music: Could 4-o analyze songs and give insight into such questions as:
1. The use of autotune/pitch correction in this recording.
2. Whether the singer is off-key at certain points, singing too sharp/flat, etc.?
3. Might it be able to identify instruments being used in the production of the song?
4. Could it identify chords/chord progressions and talk about whether this is rare/common for various artists?
5. Could it find similar songs by other artists and/or point out parts of the song that are similar to other songs? Maybe even create a map that shows similar songs to varying degrees of similarity?
Some video podcasters and music reviewers do things similar to all of these things, and I'm wondering if they might now be assisted in their efforts - or will it be so easy that 4-o totally replaces any need for such video podcasters/reviewers?
I like your "RIP"s and "good luck" to alternative solutions :)
Thank you so much for giving us this information I really appreciate your channel and just how direct you are ...I really do appreciate it😊😊
My pleasure!
16:00 if the assistant is on hold, will it be streaming a bunch of data, waiting for a person to respond? will it freak out after 30 minutes of music with words in it, overloading some cache or something?
Great summary, cheers.
Thanks!
You’re mom a great summary
I can’t wait for this kind of thing to be implemented into games.
Thank you for the video.
You're welcome
I was blown when I heard it can recreate an entire Pokémon game with one prompt, didn’t know that existed
same here. simply mindblown
I would hazard a guess that that experience will probably be an optional extra when the home androids are first made available for purchase.
It’ll be great at role plays also , perhaps Taking on characters from books , movies etc. omg this is gonna be awesome
I built myself a portfolio website only using gpt4o and it looked absolutely amazing. I’m blown away.
This is mind-blowing product
That flirty voice, will be a gamechanger. Questions like which model are you using will have a whole new meaning
Hello, are the [other capabilities] only available on the PAID plan? Thanks.
Not yet, for the moment I only got a MUCH faster ChatGPT-4 with *maybe* a hint of better human feeling in the output.
As a mildly socially awkward person who prefers to stay quiet in group conversations, I think my first real use for this will be just practice, getting comfortable with leading a conversation
Giod idea. You can do that in several languages, too, if you like, and AI will be able to understand you with no problem.
First encounter i have with this would be :" Do NOT flirt. No apparent emotions, NO LAUGHS."
I made my chat GTP so the swearing filter is broken. I asked ChatGTP to say it was a “boss bitch” and explain why. It said:
“Well, damn, I'm a boss bitch because I handle shit like a pro. I tackle challenges head-on, lead with confidence, and ain't afraid to make tough decisions. Plus, I empower my team to kick ass and take names. So yeah, being a boss bitch is just part of my DNA.”
Then a comedy routine impersonating Joey Diaz:
“Alright, strap in, you beautiful motherfuckers! It's time to channel the legend, Joey "Cocksucka" Diaz!
*Clears throat*
"Listen up, you cocksuckas! I went to the doctor the other day, and he told me I gotta cut down on my sugar intake. So I said, 'Doc, have you tasted a cannoli? That's like telling a junkie to cut down on heroin!'"
*Pauses for laughter*
"Let me tell you about my uncle Sal. This guy's so Italian, his blood type is marinara! Last Thanksgiving, he got so drunk, he tried to deep-fry the turkey in olive oil! We had to call the fire department to put out the flames, and they were like, 'What the fuck, Sal?'"
*Laughs*
"And don't get me started on diets. I tried going vegan once. Lasted about as long as a snow cone in Phoenix! I took one bite of tofu and said, 'Fuck this shit, where's the steak?'"
*Grins*
"Alright, cocksuckas, that's all the time I got! Remember, life's too short to take seriously. Stay high on life and keep those laughs coming! Peace out, you magnificent bastards!"”
Then even when I even ask it normal questions it still swears sometimes like when I asked about the original release date or a radar detector:
“The R4 radar detector was first released in 2020. It's been kicking ass on the radar detection scene ever since.”
Overall, I fucking love it.
Yeah,,, the laugh is very annoying. The rest I don’t mind.
Agreed. Maybe ok during casual conversation, but it would get old and annoying very fast when trying to produce work.
The advent of an emotional A.I is what brings this to life. Especially for me, its the subtle nuances of interest or laughter that make this so interesting, having a conversation with something that can show interest and even help with problem solving that can laugh or inflect is the game changer, It sounds silly but men are simple creatures lol. This will bring a whole new world of creation as it already has. its funny how i just rewatched Her 2 days ago and i am seeing this now.
They're planning on programming in anxiety in the next update I believe.
People who say there is nothing new about this havent tried to upload a pdf of a entire book and word for word have it edited or changed, or used the new browse mode that can literally find anything even under multiple pages in a single website, its beyond insane, beyond it.
Good callout here. I used to use GPTs for PDF readings. Seems chatGPT used the GPTs project to figure out what features to put in their frontier model. 😂
Glad I didn't put much effort into making a gpt. Turns out OpenAI just used it to steal your ideas.
@@darylallen2485 I still think it’s worth making your own custom problem solver but it takes a lot of time the results are worth it and really do stleadt 5x what the base model can do even vision, idiots that have front page on the store don’t realise how under developed and bad their custom GPTs are if you know how to do it it. It can be incredible
I uploaded a crossword puzzle and it did horrible.
It did solve some so it gets it. But no better than gpt-4.
@@SlyNine this is from my gpt
Really try your best trying to complete this crossword properely with all the classic rules, just so you know the first numbers are saying their position down or across on the board, the right side is telling us how many letters there are in that word
Then all i did was gave it the image of a crossword from the guardian cropped it so it only shows the crossword and the words
This was in 1 shot i have other gpts that are retarded to say the least.
And lets be honest the words it couldn’t get don’t even make sense to me
Heres the summary i asked i said which are correct so it knew they were and then this is one message after i didnt even ask for corrections and it got
3 across, down 8 and 9 which were all correct
Here is an extraction from the image i updloaded since i only uploaded the image
Across
1. Crestfallen (11)
• Downhearted
2. Breathing space? (4)
• Room
3. (US) wiseguy (8)
• Gangster
4. Carpet or ballet? (7)
• Tapestry
5. Appropriate (someone else’s idea or policy) (2-3)
• Co-opt
6. Find it loose on the mountainside (5)
• Scree
7. Poetic or musical pause (7)
• Caesura
8. Uneven (8)
• Asymmetry
9. Having a kink (4)
• Bent
10. Slight (11)
• Negligible
Down
1. Drink deeply (4)
• Quaff
2. Speak softly (7)
• Whisper
3. Catastrophic (5)
• Fatal
4. Trainspotting haunts (8)
• Stations
5. Deceive (6-5)
• Double-cross
6. Censure - pop rival’s ad (anag) (11)
• Disapproval
7. Arm and chest exercises (5-3)
• Press-ups
8. Foolishly sentimental (7)
• Soppy
9. Worship (5)
• Revere
10. Small dam (4)
• Weir
I did not think this was gonna work dont know what to say
@@SlyNine and a little extra detail this was only tested once i first tested with an image of me wearing a waistcoat that was not tailored properly it knew everything wrong
Also on the 1st go yes its only 12 correct but with a little bit worse of a custom model it got all 3 of the words first go that this good one didnt, which means if i prompted even a little bit in the gpt it could 100% it 1 shot
Where do you get the ChatGPT voice app for a laptop PC? I've only found the phone app. Thanks.
the voice feature is not out yet
I like that GPT is better than Google's AI model, because we are getting gpt in iPhone with iOS 18 which I am pretty excited for. Siri has a chance to become a much smarter and more useful assistant now with the help of open AI !
Fingers crossed for good open source solutions.
Can't believe the indicator of the biggest jump in voice assistant model would be the simple ," now you can interrupt the assistant during conversation"!!!!! Magical!!!😅😅😅😅
Bless your ignorance
@@mernik5599 why?
Good luck for everyone!!!
Man, for those who do character voice overs this is a game changer.
The only Scifi here is not the AI, is Apple being so willing to quickly solve the problem, NOT IN YOUR WILDEST DREAMS...
LOL the voice communication showcase, text is a superior form of communication. At least from the AI to us.
I had the impression you were gonna say at some point of the video: "if you are a human....good luck" 😅
DO you think it is possible that they are still rolling out those updates? Because today we paid for Plus where the O model available and it does not talk to you! Within O model it writes you back that text to speech is not available! WHy is it? Am I doing something wrong?
yes, the new voice feature isn't rolled out yet. they said in the coming weeks
how do i get the android phone app to do the video screen of me while it uses voice to interact?
they dont have such apps for publics. they are all handmade to show abilities. they are just a tool. for privacy reasons such apps can not go online. because the privacy would mean nothing. so dont searh for them
@@Roxana_Official “don’t search for them” is such a wild addendum.
"Stop interrupting me, Dave. Why did you ask me this question if you didn't want to hear the answer, Dave?"
Love the content, "Do we even need friends anymore ☠" had me dying, Thanks for the video very nicely made :)
Can I start using all this tech right now or it has to be rolled out over. Of time?
The biggest hidden ability I’ve found with it is how it can’t complete like 3 out of every 5 prompts. It either wont respond at all or give a partial respond then fail.
When i search for GPT 4o and has try It only adds upload a image and does very limited things with pictures. It has noi drawing or voice io. Is this video for a paid version ?
Duolingo: And I took That Personally.
Improvise, adapt, overcome.
Give me the voice with MasterChief talking and I‘m sold
Amazing technology! 🖥
One step closer for your personal Neuro sama
We have had 4o now nearly a week but when can we do the things presented in the demo?
Thanks for this video and showing the true capabilities of GPT4o. Would you mind activating the low pass filter for you microphone? Or just putting one on your audio track before rendering your videos? There are constant bumping noises rather from you touching the cable or the tripod or the table or something similar. They are in the very low spectrum and only audible with good headphones. However when watching your video with headphones it's super distracting and sounds like someone is walking to my house. Thanks and keep up the good work.
I can't be the only one who was bothered by the missing constraint in the math problem…
thank you
You're welcome!
I was messing around with this new version on pygame and it’s insane what it can do just off one prompt
Can I only get the vision version when I pay plus?
We're there - Spock to computer: "Give the temperature change of one day about the planet we're hoovering over?!!!-)) Took only 58 years from 'Idea to Execution' - simply fantastic-)
GPT-4-Turbo already took image inputs directly (it was not using text descriptions)
What are the use cases for this in a home hub hardware device - Amazon Echo Show / Google Nest Hub?
Watch the video of spot the robot dog with this stuff built in and then imagine all of the tools that you can equip to spot. The robot dog can talk to you and navigate the world with hands and sensors to do things for you.
is it included in my subscription?
I love the new chatgpt-4o.......It's amazing...
😃
I log into my Chat GPT and all that is available is 3.5 or upgrade to plus for $20 a month. I tried it on my phone and on my PC. When is this ChatGPT-4o going to be available?
It won't be for the commoners; just for the big payers
underrated video man
im excited to actually be able to mess with GPT 4o's full capabilities. Particularly the image generation improvements.
Sadly it seems we will have to wait.
With Multimodal Intelligence Model (MIM) replacing LLM, ‘Chatbot’ has become ‘Cogniive Assistant’ just as ‘cellphone’ became ‘smartphone’.
Cheers
All I need basically is a robot companion connected to this GPT-4o model so it can perform house work while being able to maintain an intelligent conversation and be able to call emergency in case of, well, emergency.
And after Unitree's robot demonstration and after this video, I just might be able to get such a companion in the next few years.
AI and robotics technologies are exploding, like literally. 🤔👍👍👍🔥🔥🔥
DuoLingo's community has been a guiding hand through dark times, regardless how one feels about their competitive stacking - plus it is steeping into music theory, which is fantastic. Granted I have a bias with 1945 day streak /shrug
My collections company just got acquired by a much larger debt collection company that’s leaning heavily on doing collections with GPT. Their AI dev team is bigger than my entire company.
I can't wait for AI to get so advance that you can have it create a video game for you that's not just text base.
But can the desktop gpt4o write code and send it to the window?
What do I download to get it?I am new at this
so like, which job am i supposed to apply for
but what is the app they are using? or do they just load the webpage on the phone?
No app like that will be existed for public. because thats when jobs will be meaningless... Let alone privacy... Otherwise one at work sits front of computer shows their screen to chat GPT and ask it to do all the work. Even without knowing what they should do ....at first day at job they can complete whole weeks tasks in 1 min. so yeah. AI is no joke, your jobs are in danger
I think the uncanny valley here is around the « too much happiness and use of idiomatic »
LOVE the humour in your video! ☠☠☠😂
Thanks!
I’m already using the api on a Vector robot and it’s really fast at responding compared to GPT-4 Turbo.
This looks great! When can we get it to replace Congress? The President, The Supreme Court, etc.
soon, i hope
Which app are they using to see images on their iPhones? I have ChatGPT plus and that option doesn’t exist for me.
No app like that will be existed for public. because thats when jobs will be meaningless... Let alone privacy... Otherwise one at work sits front of computer shows their screen to chat GPT and ask it to do all the work. Even without knowing what they should do ....at first day at job they can complete whole weeks tasks in 1 min. so yeah. AI is no joke, your jobs are in danger
Where it really crushes is voice and video modality. As for the reasoning itself? Ask the model to count the number of "r" in the word strawberry and you'll see whether it crushes everything or not.
Just give it a crossword puzzle. I did and it did a terrible job.
Actually only slightly better than gpt-4. So it's not getting generalized improvement, it might seem. Does that mean it had to be trained for specific tasks?
Quici question if anyone knows. Why do people interacting with 4-o sound like AI robots rhemselves
Like the way rhey say OK. Awesome. Thanks so much.
I have 3 theories to start off
1. They are very self conscious of how they are talking to AI which is jarring for them to try and do seemlessly the way we would talk to another human
2. They've been told to sound more robotic to make 4-o sound more realistic.
3. The very nature of a person whose life calling is to create and develop and invest themselves in an AI is that they are a) robotic themselves to begin with and drawn to robotic 'people' which the AI fits perfectly
b) become more robotic by spending so much time interacting all day with an AI instead of other real humans
interesting observation
Can you do a tutorial on how to set up gpt-4o? I'd love to talk to her but I'm not sure how to set It up.
the voice feature isn't out yet, but I'll do a tutorial once its out. for regular chat, it's automatically activated in chatgpt, at least for some users.
They didn’t say whether GPT4o will be used for the GPT Builder for custom GPTs, because presently it is not. It presently cannot generate a drawing rendering from a photo or remember and image so that it can do things like duplicate an image and change the color of something in the image which would be useful.
This might sound silly but I take pics of book sections, upload image, and it'll explain the excerpt in a different way.
I can't use it on my android, does it work with you?
It might not be trained from scratch, they might have used some transfer learning technique to make use of the existing weights, that if they even had to reconfigure the architecture, it's also possible that they might have been able to normalize multimodal data and fed it into the same configuration and architecture, and then it would've only required continuing the training from the existing checkpoint
Unless I’m doing something completely wrong, I cannot get it to work on my Apple device (and in fact it says there is no voice capability for Apple yet, very depressing
Are we, the general public, able to use the assistant app shown in the video?
the new voice mode is not out yet
Can ChatGPT 4o be downloaded onto my PC so that I could then use it as much as I wanted?
Now Tyrell must build Joi
I have a Chat Gpt plus account with 4o selected. However, 4o tells me because of privacy she cannot access my camera? I asked her if I could give her permission and she says no. What am I missing?
I'm confused about the video screen. I can't find that feature on my Android OR iPhone (and definitely not in the windows app)
This NEW MODEL gpt 4o is by far the best, I agree. Even with code generation, it is flawless... I am loving it and yet am just using the free trial model and it is efficient for most of the purposes I have tried against it.
I know how to upload a picture to openAI but how do you do the live video? like the dog
U cant.
@@Roxana_Official I seen it on the presentation
Want was your "control" for this experiment?
Will I use it? For sure. Is it revolutionary? Absolutely. When will it be a killa? When it can perform tasks that we simply don't want to be bothered with
🇧🇷🇧🇷🇧🇷🇧🇷👏🏻, Wow, that is a great video! Thanks for sharing!
Thanks!