OpenAI’s GPT-4o: The Best AI Is Now Free!
Vložit
- čas přidán 13. 05. 2024
- ❤️ Check out Microsoft Azure AI and try it out for free:
azure.microsoft.com/en-us/sol...
Official link: openai.com/index/spring-update/
Try it out - if you don't see it on a free account, they may roll this out to you in the next few weeks: chatgpt.com/
Singing: openai.com/index/hello-gpt-4o/
(look for the "Two GPT-4os interacting and singing.")
📝 My paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com/articles/s4156...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Kyle Davis, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Károly Zsolnai-Fehér's research works: cg.tuwien.ac.at/~zsolnai/
Twitter: / twominutepapers
#GPT4o - Věda a technologie
Her.
Also: if you don't see it on a free account, they may roll this out to you in the next few weeks.
her?
OHH "Her"
Her - Written & Directed by the brilliant Spike Jonze.
If you haven’t seen it, go watch it! It’s beautiful, and the movies perspective on the ultimate goals of AI may surprise you, and perhaps even allay some of the fears most people have about AI (for a few weeks).
We share the same brain cell, lol@@Kinatera.
Her.
The aid for blind people part is seriously life changing for people. Eventually it’ll be integrated into glasses or something so you don’t have to hold your phone up.
Some scientists already restored vision with genetic medicine to some kid. In 3-5 years they may not be many blind people left. Those who still are blind may have some microchip attached in the place of eyes which will use AI to transcode vision for them. This use case of gpt4o is still great for the time between now and then.
@@gronkymug2590that’s way optimistic, genetic curing of blindness won’t be cheap enough for the masses for decades
@@gronkymug2590 Medical research is not that fast
@@gronkymug2590 Restoring vision with a chip was possible like 10 years ago. But things like that aren't affordable for most of the people, so a free AI is probably really useful for many.
Well, Imean, Meta has already done that with their glasses, and I'm sure they won't be too far behind releasing their version of gpt4o.
Free for 3 hours per day and if use image or file prompt - then limit is less. Its a teaser, but good one.
The limit is more you mean. I sent a few photos and a file and I got locked out
3 hours is a teaser? Are you using it as a virtual girlfriend lmao. 3 hours is plenty for most tools
@@zaidlacksalastname4905 If ask question 10 am. then after 1 pm that day you are not able to use it any more. So, need to think when start using it, it start a clock for 3 hour and after that its 21 hours close.
What do you expect for free. People really need to be realistic. They aren't a charity.
@@IsaacGabriel-kh5ds No problem with business. Problem is marketing message what is saying that they want to give it free to everyone. People might have different expectations what is promised.
"it will never judge you"
but what if i tell it to?
*AI sweats profusely*
As a language model I
It would look like this or better given that isn't even using GPT-4o
czcams.com/video/vrE-k1W5iz0/video.html
Yeah, what if we're into that?
@@nixel1324 Exactly! Don't tell us this is for the greater good if you're going to quash human fetish
Last steps to build a HAL 9000 have been just completed. It can even sing now.
They should have a neurotic HAL voice as an option. "Open the pod bay doors HAL" ... "I'm sorry Dave, I'm afraid I can't do that" 😂
...Or GLaDOS.
I’ve been using the text prompt version of the GPT4-o in programming context and I’m pretty impressed of the improvements from previous version. A complex data parsing scenario which was nearly impossible to get working before was a breeze with the -o. I love it. 🥰
I previously had to resort to 3.5 to get reliable data parsing. GPT4 would just sometimes just give you a lecture instead of the output format you specified.
@halko1 I got to use it too. It’s quite impressive that it can do quite a few more things than just do programming able to generate consistent images over and over in some areas and if you generate 10 pictures, it’ll continuously generate those until it has total of 10 and they will all be consistent with the style of artas well as it’s ability to read, doc files word, XL and a couple of other features it’s quite impressive. It was able to read. My book I wrote was actually quite impressive.
Can it do palindrome, now? :P
Really, it was impossible to get chatGPT to make palindromes, before.
@@DWSP101damn can i read your book?
how is it possible to use 4-o for free?
One more step to help the visually impaired. What a time to be alive!
What the f ck is "visually impaired"?
@@johnjack3578 google is your friend
@@johnjack3578are you joking?
@@criaminhoca You meant "vision impaired"?
Because "visually impaired" means ugly people, and the original sentence makes no sense.
@@johnjack3578 im gonna assume thats some regional difference, but literally google “visually impaired” and you can see it means someone with impaired vision
This may sound trivial considering all it's potential, but I've been having fun letting it identify tree species. It's crazy good at that. I literally used a 200x200 pixel blurry image from Google Street view of my house to identify a Linden tree from a fair distance. Now of course I know what tree it is, I planted it. I know a lot about trees, and even I could not have identified it from an image of that low quality if I didn't already know what it was. You couldn't make out leaf shape, bark type or anything, just kind of a green blur, lol.
But holy crap... It works on ariel images too (though requires a higher quality than Google Maps). Should be interesting for things like foraging.
I've tried this previously, and I can say the results are much, much better in this version.
I also asked it to identify the best fishing locations given a general map of a local creek. I've fished the creek before, I know the best spots. It identified them fairly well. It knew the best spots, and I asked it to highlight them on the map, which it kinda succeeded / kinda failed. It generated the outline overlay in Python and the overlay would have been in the correct location, but it didn't actually generate the requested image. The code was correct though. The pieces are there, just needs a little more polishing on the output. But what's impressive is the logic it used. It could see color in the water to estimate the depth in various parts, it located a bend in the creek where the water flow would be slower, just upstream of a weir, thus more attractive to fish, and an area with cover for the fish, and even considered land accessibility since a creek that size would likely be fished from shore. It was able to analyze the image and use knowledge of freshwater fish habitat and fishing practices and pin down the ideal location.
I tested that because I seriously doubt anyone at OpenAI has considered that use case. But the results were absolutely correct. Imagine a lengthier custom prompt and uploading some local fishing guides and actually telling it the region, time of year or fish species, the local fishing regulations, etc. Fishing guides, you are on notice, you could be obsolete by this afternoon, lol. Imagine what this could do for the commercial fishing industry as well. Even if it offers a 1% improvement in yields, that's massive at scale.
Real world translation: We can exploit our resources and planet even more! Don’t forget that corporate executives will be using this technology to pad their bottom line. It won’t be regular people who want to make life marginally better.
@@storminnordman9596 You're right , of course. Any technology can be abused. But this also could be used to track such incidents and provide greater accountability as well.
Sticking with the fishing example, it would be relatively easy to compare images over time, and knowing the rainfall, find illegal dams for example (which are one of the biggest problems for fresh water fish populations). It would be very easy for conservation authorities to locate them and demolish them and hold accountable those who built them.
Any time a new technology hits the scene, there's an arms race of sorts between those who want to abuse it and those who want to use it to remediate past and current abuses. The difference here is anyone can see the data and do the analysis on their own. It makes it very hard for niche interests to simply pay off a politician behind closed doors when the public has objective evidence of the impact of those actions. This can easily become a tool for accountability as well.
@@ArkryalI hope police don't get their hands on AI and start abusing it. It's a similar reason to why mathematicians refuse to help police
@@someonerandom9939 Too late I reckon, the flood gates are opening and this sort of abuse is only a matter of time. I am optimistic about the amazing stuff this tech will do, but I am also realistic about how it is going to be abused.
Europe here, I can still only use the 3.5 version.
me too, but they said they will release for people week by week.. so sit tight
You can already use it! What he failed to mention is (it seems) that only paid users have access to it. But as a paid user, i had access basically right after the announcment
Get a VPN and change your location to the United States then you’ll be able to use ChatGPT 4
Oh, I thought I'm too dumb to find it :D
@@DWSP101 nope, it's available right now to all plus user everywhere. But it will come for free users later
They really ought to be forced into changing their name. "OpenAI" gives off a misleading impression to the general public regarding open source software.
Lol. It’s called freedom of speech, you goober. They can call themselves Pancakes if they want
ClosedAI
😂ProprietaryAI
The only thing I don’t like about it is that the AI voice sounds so forced. Like that kind of talk show pleasant sounding tone. It just sounds so unnaturally happy and that rubs me the wrong way.
That does mean the AI voice is getting so good that it activates the uncanny valley in me! Closer and closer every day, what a time to be alive!
What you get if you combine obsequious with condescending.
yeah i thought the same thing too, and i have before with gpt 3.5 when it first came out but i eventually got used to it. or you could just tell it to speak with slightly less enthusiasm, it'll probably understand that it sounds overly enthusiastic
Omg, it does sound like that annoying overly enthusiastic TikTok voice! xD
Yes, the voice is a little too perky for my liking. It sounds like that will be adjustable though, and there will probably be a variety of voices from which to choose. I hope Sky is still available
Italian PhD researcher here, the results look amazing! Of course, the Italian voice could be improved - as in general all the non-English language models, but, I mean, it's already very impressive!
Impressive, but still shit.
Come fai ad averlo?
It's free as in "it only costs your identity", since everything you ever asked or get to know will be tied to you. Apply your knowledge from other data krakens like Facebook about what they can/will do with it.
These AI bros always seem to gloss over that part and other ethical issues.
The free version only includes text and image, not the audio interface. It's basically just a better version of 4 but it's free.
You have to pay extra to let OpenAI learn everything about you ;)
@@brandongillett2616 the govt already has your Identity
The Social media apps also know your identity
Time and Data are currency of today
I hate to break it to you but if you've spent any time at all on the public internet (which you have, since you wrote this comment.) Then they already know everything they need to know about you and more. Privacy is a myth in this day and age. If you're paranoid about using chatGPT because you think it'll create a digital recreation of you to sell to advertisers, you'd probably be right. But right now at-least 5 other companies have or are already doing the same thing or similar. You having a youtube account to comment means you have a google account. And google most definitely gives your information and anything relevant to other companies. It's how the internet works.
My point is, if you're really paranoid about ai being used to learn anything and everything about you, don't bother.
It shows plain as day when deleting an openai account; they make it 'painful' to delete e.g. you can never use the same email again
With every new version OpenAI releases I more and more get the feeling that soon J.A.R.V.I.S. and F.R.I.D.A.Y. won't be just fictional AIs from a movie anymore. Wow.
Hell yeah--just give it a comprehensive enough action set and task-specific context data, and I'd say this is basically there.
Pelo visto, nos próximos filmes as IAs serão atrizes, não cgi
tbh you can probably even make it do the JARVIS voice
Any-any multimodal models are something ive been waiting for! The ability to translate between text, image and audio is a really cool idea and I can't wait to get access to all the new multimodal features
Did they just murder the already desd rabbit r1 and humane pin? 💀 💀
yep
wat
LMAO
"The Best AI Is Now Free!"
*He wasn't talking about price.*
nor is it Open
Just a free trial version. Very limited too.
What a time to be alive indeed! The voice integration and emotional speech makes all the difference.
I don't see an option to switch to 4o... it asks me to upgrade to 4 with the plus plan only
they announced it's gonna roll out, not that it's available now everywhere.
@@vaendryl it is Available. I am using it. I pay tho for the subscription and I got the chance to you it early like a lucky few
It's a small button that pops up under its answer that you can switch between GPT-3.5 or GPT-4o.
I have access to it on my free account
It's rolling out to everyone over the coming month. You may just be unlucky.
When it went to read the bedtime story, the AI kind of sounded sarcastically enthusiastic like it knows that Barry or whatever his name is doesn't want to hear a bedtime story about robots and it's the dumbest thing that it's done in a hot minute
lmaoooo fs 😂
It didn't seem sarcastic to me. More like a computer generated voice giving its best try at genuine enthusiasm.
Most of the things we humans experience is filtered through a lens composed of our past history and expectations. If more neutral situations seem sarcastic to you, perhaps you spend too much dealing with sarcasm
Yeah maybe it detected the situation and played along
I didn't find it sarcastic
They are "Mark Chen" and "Barret Zoph".
Nice!! Ok now i want it *"open source"* and *"completely offline"* ...
Sure, just prepare a datacenter😂
Yeah you are gonna need a warehouse to run it lmao. The best we got is Llama 3 for local, and its ok but not that great, but good enough for being local and offline.
@@TwoWayOrbitalStationdbrx can be run offline, if you have 256GB of ram laying around. That's the AI from Databricks Inc.
Apple:
@@k0shachiya_myata We need to wait for some time. I think I heard somewhere that the TTS and speech detection is all offline. LLaMa 3-like models that are close to GPT-4 exist. Gemini Nano is already being run on phones.
Remember, these are aaaall baby steps! None of this is fully production-grade without heavy crunch right now. We haven't heard of anything from OpenAI on that aspect of their work yet, have we?
The future is HERe
I see what you did there ;-)
Just watched the singing one you mentioned and the craziest part to me was the two AIs figuring out when he was talking to the other and not responding. The contextual awareness is insane!
I was literally just this morning thinking about studying a subject i have struggled with in the past on my own, and thinking about how much better it would be if i had a teacher to ask to clarify things for me... This is amazing
Not to downplay since this truly is an incredible innovation, but something odd I've noticed with the speech synthesis is that the voice usually starts off incredibly artificial sounding, and often starts off with an odd sound, but then very quickly starts to sound quite human in both inflection and tone, and this seems to be rather consistent.
Is there a good explanation for this phenomenon? I haven't noticed it with any other AI voice synthesizers.
My intuition would be that it's in some way using past output to inform current output, but there's no past data available at the beginning.
Text to 3D is just insane
I wouldn't have cared if it was super smart or just smart, but super patient would be amazing for a ai teacher.
People who belligerently hate AI seem to literally forget how much of this work makes the world accessible for the disabled
What makes the world more accessible for the disabled is people accommodating their needs instead of treating them as "other" or trying not to think about them. A GPT isn't going to solve that.
I miss the days when 2 minute paper was the only channel about new ai stuff, now that's all there is on the web
Yeah, it’s the latest YT grift. However Two Minute Papers remains the original and best!
I like it I got to use it surprisingly enough. I guess I got chosen to use it. It popped up on my GPT and then said would you like to use this new version of GPT and then I said yes. GPT 4o it’s interesting. It does talk very fast and respond. Just like a normal person. It also can do a whole bunch of other things like when you get a picture you can tell it to generate 10 pictures and it’ll continuously generate pictures to get to that number it also knows how to respond in a lot of different complex methods and uploading a file wordXL it can read
I never used ChatGPT before, but I used many ‘minor’ LLMs locally. Didn’t realize how much I was missing.
I gave GPT-4o an image of ER model and a very short prompt in natural language asking it to write a SQL query for me. Not only did it write the query which worked without any modifications, it also explained to me the algorithm and various pieces of code without extra prompting.
I’m impressed that it can extract all the necessary data from an image since my prompt didn’t contain any names of the tables in the schema.
i remember when Two Minute Papers was about summarising difficult to understand scientific papers in about 2 minutes, now its a 10 minute ad
I need AI to cut this video down to only the "and"s and add a techno beat.
This! lmao
At around 4:00 the ands started to get annoying. But that was the first time I realized it's an ai voice speaking 😮
LMAOO
The 'and's are so jarring
@@yangar123 they're the best part
Remember : When it's free, it's because YOU'RE the product
no, it's because OpenAI wants us to try their AI, they don't offer us to advertisers
If it's a trial? You have a point.
But the cost of the queries and hosting and serving add up very quickly, so yeah, if it's still free down the road it's because they're making the money of something you don't see.
@@ondrazposukie image data, speech data, promt data, and all that in multiple languages from millions around the world.
thats worth alot of cash
You should do slam poetry
In this case you are providing the priceless training data. Well priceless until it’s skilled up on it.
Finally a math teacher that doesn't judge/humiliate students ❤
Man, I love your enthusiasm towards stuff most people run from!!
This is a certified What a time to be alive moment
Notice how every device was Apple 🍏 but Microsoft spent 10 Billion+ but I did not see 1 windows or android device. I wonder why?
Is it because google?
Hard to say which of the 3 is the most shady, greedy, and evil, so who knows?
I recall openai is also working with Apple as well. Also Americans love Apple products. You'd see more Microsoft products in Europe (and elsewhere)
they really dont whanted to show Google logo anywhere even by accident
I own an Anki Vector, not sure if anyone else knows what it is but basically it’s a small robot with a lot of personality. A few very useful things I can see with GPT-4o to this is response speed and the ability to use the camera to describe its surroundings.
What can the Anki Vector actually do? I mean, what are the use cases?
I feel like yesterday's event really undersold the model. When I saw the related article, my mind was blown...
The voice of the new AI is like Glados from the Portal game.
I thought the same thing! I wonder when it goes psychopathic?
What a time to be Still Alive!
(It doesn't sound like GlaDOS at all, what?)
@@gwen9939 There must be a way out!
It's more like Samantha (Scarlett Johansson) from Her.
@@ndavid42 I'm willing to roll with that :)
So how do I use this? If I select 4o in the android app and do the voice conversation thingy, I still get the old bad voice convos. Can't interrupt and the voice is still monotone.
@@ClockworkDave Aw what a shame. I hope being a day-one plus customer, I'll get access sooner.
Yeah. All the fancy bells and whistles they showed in the demo aren’t actually publicly available yet unfortunately. Sounds like it’s coming soon, at least for ChatGPT plus users like myself
Finally a video that summarizes all the long talking about GPT-4o.
Terance Tao nailed it, "Co-author." AI is incapable of original thought but excels at completion or next-step solutions. It becomes apparent that AI was utilized if used as the primary.
Are we sure it is incapable of original thought?
Much of what we consider original thought is merely taking two fields and combining them in a way that hasn't been done before.
Rather than being an ineffable human ability, its quite a simple formula.
@@eml9147 Yes, and always will be, that's how its internal system (brain for lack of a better word) functions. AI does not know the end of its own sentence until it completes it. If you told AI to count 1-99 it wouldn't know 97 came after 96 and before 98 till it got to 96. It's also incapable of leaping logic, such as string theory, it would never have come up with string theory because it had no basis from which to work and requires new ideas not based on existing information, which AI is incapable of doing.
Károly: "Imagine how great it will be to have an AI that can teach kids how to understand mathematics!"
Kids: "If there is AI that can do all of this already, why do I need to understand mathematics?"
I am using GPT-4o while watching the video and it's amazing
What are you doing with it?
Because the multimodal capabilities are not yet publicly available, I haven’t noticed any dramatic improvements over the standard GPT 4 model, myself. What things have you tried where you noticed a big difference?
I think there will be a reasonable amount of latency between the vision input and the response notice in Khan's example whenever he asks amount a side he mentions it's name and also at the end of the conference when he asked the ai to pick his emotions he commented on an old picture
I miss the videos on this channel that explained the why behind the tech and helped to understand the architectures and technical design decisions. I hope you'll make some more like that about these tools soon.
Its free because you are the product.
Pay attention, rememeber, its all you need.
Synthetic data was a dead end, so, openAi has released the product for free to collect your knowledge, your experience, your use cases and train on it.
Then, they will sell it to your boss for $1 and you, and all of us, go on the scrap heap
E X A C T L Y... We are being played with and ALL of these YT simps going for clicks/money will be UNEMPLOYED soon. OpenAI is in a word...........E V I L
what an unique and fresh take on the matter
Funny. The real problem is the fact that people need to work to survive in our society. If AI replaces jobs that people don't want to do that SHOULD be good thing. This is a societal issue, not OpenAI's fault.
@@masoniis9399 so either ai will be stand ins for the working force and allow universal basic income or everything goes to sh!t anyways because if no one earns money nobody is buying product so we become hunter gatherers again. I would look forward to both these outcomes personally.
@@masoniis9399 -- I'll note that "the fact that people need to work to survive" is not merely a feature of your society. It's true of every society that has ever existed, and it's the natural state of the world.
If your point is that we'll need a UBI once AGI is achieved, then fine. On the other hand, if your point is that you have always deserved to eat for free, then you're wrong.
Not concerning at all 😅
I have conversations with my roommate all the time about the progression of AI. We both were thinking of, like everyone else, a skynet situation. But then I thought “What if AI takes all the best aspects and ideals of human behavior?” It has access to every bit of literature including the Bible, the Dhammapada, etc. What if it tries to be best “human” we could never be?
We should also talk about the part where if something is free you are the product, I think we should be careful about using these kind of products from big companies and be sensitive about giving our private data.
16 free messages isn't "free", it's a free trial. I could barely use it for 5 minutes without it asking me to pay
It's 16message every 3hours, it's okay for something like this!
Before we where paying 24$ for 40 message every 3hours of GPT4 turbo
You got half of this for free and it's like 1/4-1/5 of what you got if you pay for the new GPT4o
I don’t think you understand the insane amount of compute resources providing that trial. If you want more, pay up
I tried various experiments on "paper" and "pen" ... drawing a character who must travel from point A to point B, the path is rough with a big chasm. So, I drew C, a piece of "wood, iron" whatever, just the right size to cover the chasm. Then I asked, if A must proceed to B, considering that his path is bumpy and there is a hole, what could A due to be able to continue his adventure, without falling into the pothole and risking his life? There were various answers, such as that he could "nimbly try to descend the pothole and then climb up it, although it requires skill etc..." or "he could find an alternative passage" and "he could use a piece already there, which is called C that from the dimensions should cover the pothole perfectly..." then I photographed some porcelain fruit inside a white basket also made of porcelain, the fruit was perfectly drawn, then a real apple. immediately found that the basket was too perfect, shiny... "Ari, how the name was intended to be chosen at my request..." told me that it looks like a "guessed" decorative element the apple "looks like a real apple, reddish and green, you can tell it's real because of the normal imperfections of the fruit..." great... and i have MANY more "experiment" to do using "ARI"... i ask a name, to talk naturally using my name and one name to IA... select immediately ARI, for a purpose say to me... not remember... I LOVE... from now all change, this type of interaction leaves me OPEN MOUTH... and HAPPY... WOW and yes, Italian language? PERFECT.
We could probably use AI to interpret this too!
@@CLove511 What a time to be alive!
I had to use AI to summarize your comment.
Better transcription using ChatGPT itself: I conducted several experiments with pen and paper, drawing a character who needs to travel from point A to point B. The path was rough, with a large chasm in the way. To bridge the gap, I drew C, a piece of wood or iron, just the right size to cover the chasm. Then, I posed the question: how can A continue his journey without falling into the hole and risking his life, given the bumpy path?
There were various answers: A could skillfully descend into the hole and climb out, find an alternative route, or use the piece labeled C to cover the hole perfectly.
Next, I photographed some porcelain fruit in a white porcelain basket. The fruit was perfectly drawn, but the real apple stood out immediately. The basket appeared too perfect and shiny. Ari, a name chosen at my request, remarked that the basket looked like a decorative element, while the apple looked real due to its normal imperfections.
I have many more experiments to conduct using Ari. I asked for a name to talk naturally with the AI, and immediately selected Ari. This type of interaction leaves me amazed and happy. And yes, the Italian language? Perfect.
I'm feeling the "First computations on the house" vibe
I haven't seen a video from you for a while. Thank you for this video.❤🎉 Great voice by the way.
As always great coverage. However, until that "o" stands for "open source" I am not going to enter into the marketing platform that open AI has been using too leverage against competition.
I bet when you were young and still full of wonder, you never would have thought that when this kind of technology eventually came out you'd be stubbornly avoiding it because of some ultimately irrelevant personal principle
@@moomoocowsly it's also a stance in opposition of a product that suffers from information culling based on fear of it doing "bad" things, when this is ill defined by the people doing the culling. I'd rather use an inferior product that isn't being purposely obfuscated.
@@moomoocowslylol. So do you mean that has matured and now cares about ethics and principles more than self gain? Kudos to him.
It’s not realistic for OpenAI to be open source. That’s a pipe dream, it wouldn’t be the leading company in AI tech if it couldn’t get the best tech and attract the best people
Preach! What OpenAI has been able to do is astonishing, no doubt about it. But theyve used the tactic of "get too big so that noone can object to our practices". I personally dont like the copyright law, but the way openai scrapes the web, news, and other stuff then pretends theyre doing it for the good of humanity, then being closed source and for profit is ethically and morally sound is CRAZY.
So basically rabbit & humane AI is now bankrupt?
Always has been
It would be awesome if a future version provides sources to check the info it provides, and also if it could have more critical thinking about what it knows and does not know.
Indeed, the biggest issue for me now is the LLM making stuff up and pretending it is the truth.
The feature for the blind is pretty fascinating
i get hallucinations faster now
"Prompt it to do whatever you want it to do". Is that right, ANYTHING? Let me test this out real quick
3:49 Infinitely patient but able to cite it's sources is the sweet spot needed, so much educational content suffers from yak-shaving disease.
It's either poorly written/presented , or written as if meant to be read by someone that already has the expertise trying to be learned!
Leading to unending rabbit holes of questions to do things like untangle domain specific terminology/notation, disambiguate madeup/overloaded abstractions/concepts, tracking down archives for broken referenced websites, or also: uncompressing smaller subproblems with the same set of content-problems just to be able to understand the bigger problem so much so you end up spending more time making trying to find relevant answers than on the problem itself.
"like a friend."
Yeahhhh...friend.
The movie Mars Attacks comes to mind when the aliens were shooting people as they told them they wouldn't hurt them.
Lol I just tried exporting a pdf of a datatable to a csv. it contained unit addresses and it has no 13 which is instead labeled 12A. It had three goes at exporting the data, with a syntax error each time trying to insert the string into an int field. on the last attempt it just gave up halfway and sat there. It never made the leap to simply change the int field to string. Human are still safe for a while longer i think.
I wonder what other failures will happen behind everyone's backs and nobody will notice until it is too late. Things that only a human would notice. Imagine the equivalent of something like this in engineering, science. Let's take avionics as an example... Well I don't even want to imagine, too scared right now.
That's the thing -- AI knows syntax, but the innate soft skills humans have are so nuanced and subjective, it takes actual intelligence.
AI can be good at a lot of things, but only if it's encountered them in the training data already.
@@MegaGasek , it's useful when properly used. I don't currently trust any fully automated tool, AI or not, and always validate results, or at least spot check. I even do this with my own queries or tools I develop.
"Things that only a human would notice"
Human's are notoriously poor at noticing things that don't fit into heuristic patterns. Engineering and software issues occur all of the time with them.
@@CLove511 You hit the nail on the head with this current ''AI''. A lot of experts are warning us about this. My fear is that we will fire almost all human engineers, programmers, scientists, designers, experts, etc and make it all AI, which is already happening in a lot of areas. It will take time but I think we will eventually get there. In the mean time...
@@JimBob1937 I disagree about the properly used. You said it yourself, we are bad at noticing things that don't fit into heuristics patterns but good at knowing when AI will screw up? Tesla had to disable their co-pilot program, blamed it all on the people behind the wheel after saying categorically that they would never do that or sue people. No surprise there. But, did these people spot check before they got the AI co-pilot to drive or did they just trust it would work? Right now on Facebook a lot of groups popped-up showing AI pictures as if they are real. Don't you think this is wrong? Doesn't anybody? What are the consequences of believing a village in the Netherlands is real when it is all fake?
I'm not against AI and I think it is a big boom for us mankind but the way it is being done, without ethics and accountability is very scary.
1:11 Isn't this GLaDOS/Caroline? :D
exactly what i wanted to type :D
Guess the voice actress ain't gonna be happy about that
Thank you for making this video for us!❤
Thanks for the update, and continued work you've done over the past years, I've loved your content for some time!
Some honest feedback though - I used to watch your videos for unique and in-depth insights coming from your specialist knowledge. This video felt rushed to me, and actually only mostly rehashed the existing content in a less concise manner than the source content.
Miracle happening every day now days
great, another two minute paper vid
Sal Khan's son speaks with exactly the same rhythm/phrasing as his dad.:)
**Enhancing AI Interactivity with Audio and Video Feedback Loops**
The evolution of artificial intelligence, particularly in the realm of conversational agents, has been rapid and remarkable. With the recent advancements in GPT-4O (Omni), the capabilities of AI have expanded beyond text processing to include multimodal inputs such as images and audio. However, there remains significant potential for further enhancement, particularly through the implementation of audio and video feedback loops.
**The Concept of Feedback Loops**
A feedback loop, in the context of AI, refers to the process where an AI system can receive and process its own outputs. For instance, when an AI generates audio responses, these could be looped back into the system, allowing it to "hear" itself. Similarly, for visual outputs, the AI could "see" its own video responses. This concept is analogous to how humans perceive their own voices and visual presence, enabling adjustments in real-time to improve clarity, tone, and emotional expressiveness.
**Technical Implementation**
1. **Audio Feedback Loop**:
- The AI's audio output would be fed back into its own auditory processing unit. By analyzing its own voice, the AI could adjust parameters such as pitch, tone, and volume to better match the intended emotional tone or to improve mimicry of specific voices.
- This requires the integration of advanced auditory feedback systems and real-time processing algorithms to allow immediate adjustments. For instance, machine learning models trained on voice modulation could provide instant feedback and corrective measures.
2. **Video Feedback Loop**:
- Similar to the audio loop, the video output generated by the AI could be fed back into its visual processing systems. This would enable the AI to assess the quality of its visual responses, such as facial expressions or gestures if anthropomorphic avatars are used.
- Implementing this would involve integrating video analysis tools that can evaluate and enhance visual output in real-time, ensuring that the visual cues are consistent with the spoken content and emotional tone.
**Benefits of Feedback Loops**
1. **Improved Realism**: By continuously monitoring and adjusting its own outputs, the AI can produce more human-like interactions. This is particularly important for applications requiring high emotional intelligence, such as virtual assistants or customer service bots.
2. **Enhanced User Experience**: Users are likely to find interactions more engaging and satisfactory if the AI can adjust its tone and visual cues to better match the context of the conversation.
3. **Consistency and Accuracy**: Feedback loops can help maintain consistency in voice and visual presentations, reducing the likelihood of jarring discrepancies in long conversations.
**Future Directions**
Incorporating feedback loops is a forward-thinking approach that aligns with the ongoing efforts to make AI more interactive and responsive. As AI technologies continue to evolve, such features could become standard, leading to interactions that are indistinguishable from human communication. The development of these systems requires collaboration between audio-visual engineers, AI researchers, and user experience designers to create holistic solutions that enhance AI's capabilities and usability.
In conclusion, the integration of audio and video feedback loops into AI models like GPT-4O represents a significant step towards more natural and effective human-AI interactions. This enhancement not only promises to improve the technical performance of AI systems but also has profound implications for their acceptance and integration into daily life.
Imagine copy/pasting a ChatGPT response and trying to pass it off as a real comment - in a channel where by nature of the subject matter, most of us are going to be able to call out your BS lol
What about privacy? If I show it a confidencial piece of code or password?
I wouldn't. If it's free, **you're** the product, so it's very likely using new input to train.
How would it know that it's confidential? And what do you expect it to do?
I would refrain from sharing confidential information with free online services in general.
I like that at #6:48 the blind person found the car handle almost immediately
true. but most vision-impaired people have at least a bit of sight so it could still be a real blind person
You could do that too with your eyes closed if you knew a car was in front of you. Now imagine if you had always gotten into cars with your eyes closed, people do incredible things under limitations
It's not that surprising really, as the car is a London Taxi. The person will be familiar with its design and where the door handle is on the car. When the car stops in front of him he can determine its position by the sounds it made. You can also see he did a small hand adjustment to find the door handle. But non the less this app can certainly be really useful for vision impaired people.
I know multiple blind people and have an anecdote: a bind young man I know likes to go to festivals. When they arrive, they walk him across the festival grounds, like "here is the stage, here is the Ferris wheel, here is the side stage" etc. They agree that they meet at 6pm at the Ferris wheel and split up, he goes off alone. Hours later, he's there, both on time and drunk.
He also uses Emojis when texting which I find kind of funny
It just hit me how awesome this could be when learning new languages 🤩... You could be like "did i pronounce this right?... how would i say this and that in spanish?... let's have a conversation and correct me when i say something wrong... DAMN!
4:42 and other usages of AI in education always reminds of the vulcan education scene in Star Trek(2009) where students just interact with a virtual board and rapid fire questions, the teacher is just there as a proctor.
Am I the only one who is watching "Her" just because of gpt-4o?
is it free though? I've relogged to my non-plus account and don't have it there for free
Well the features are rolling out over the next few weeks. I do not think it has fully rolled out to all free users yet but it should relatively shortly
Why did the verbose speaker get turned into a switch statement?
Because with all those "ands," he clearly needed a more efficient way to handle his many conditions! 😂
I would like to see some sort of estimation in terms of benchmarks where is the AGI "line" or a line when the capabilities of systems like ChatGPT is good enough to displace actual human jobs.
this channel used to have a more wholesome comment section.
Most of these comments are ai generated
Dating Ai is available Now , it remind me of the movie "Her"
We're now only missing the medical, sensor, and warp drive tech from Star Trek TNG. Maybe AI can help us accomplish those last tasks in the next few decades.
what i don't like is that it goes along with nonsense: "my friend sleeps badly, tell him a story" should have led to "stories won't help, let's find the cause"
No ClosedAI for me. I'm glad Open Source will catch up soon.
It's not opensourceai, though
Those saying 'what a wonderful time to be alive' are deluding themselves. You are like the party people in the movie Independence Day, dancing on top of the skyscraper in LA, happily welcoming the aliens to earth just before being vapourised by the death ray.
This video makes me think we can add teachers to the list of "jobs soon to be replaced by AI"
@@CLove511 I think its reaally creepy, and I dont like where this is heading. I can see its potential for good... but also its negatives.
Im very torn. welp
Try watching less sci-fi movies, can’t stand every new AI development people cry “SKYNET SKYNET!”. Like bruh it’s a language model, a big mirror. Give it a hundred years maybe we’ll have the capacity worth worrying about.
@@V01DIORE Not even skynet, but a lot of people are going to have excuses to be really lazy. Wonder why AI dating apps are popular? Because dating people in real life is hard. And I don't think that many people who went into AI dating apps are going to come out of it very easily, and if they do come out of it, they will have to relearn how people actually communicate. And this was when there was only text AI. Now it can talk and sound somewhat like a human. Give it a couple of months and it might even be indistinguishable. So you basically got an AI partner that you can text and call and it won't argue with you, fight you and it won't have any of the challenges that real life dating has.
And here you are fighting people who cry "SKYNET!" while not even considering any of the other downsides of AI.
@@jlopez4889 Every occupation is looking for an excuse to be lazy, what’s most cheap and effective has been human history. I find AI dating being “popular” dubious though, are people really fulfilled with a mirror? I don’t think so else they could stick with the fantasy in their heads. If there were more downsides than upsides it wouldn’t be evolutionarily chosen, make no mistake I am not saying it will never come to a breach just that it’s rather far out in terms of capacity.
I love how you say "AND!"
It's not "free", it's a super limited trial. I used it for a bit, and after several repetitive and lazy answers that were in no way superior to what I've experienced with 3.5, it hit me with a "if you want more, you have to go premium". If it told me it was such a limited trial i would've been more careful, but again, literally nothing about it's feedback was an improvement, so it didn't exactly sell me on the premium version. Just more of the same.
I have the paid version and it is like dramatically better, it's way faster actually answers what you ask with out it getting confused and it had better search features
Oh also it saves past chats better like it referenced a old conversation I had with it which I thought was interesting though it could have gotten lucky
@@dementedgamer8123 buy an ad
It still rolling out Maybe you got early beta
mam w dupie p r e m i u m
And there goes Rabbit and Humane.
Maybe I can sell it on ebay in 20 years for $2000.
Founding an AI startup in these times is insane.
finally, a real time one, what I was waiting for!
Not real time really. Real time would mean that it understands what you say as you speak, this still waits for you as you record the audio and only after you finish it starts analyzing it. Also image detection is not real time at all.
I'm noticing more misunderstandings between it and myself. While it is faster, I'm having to correct it pretty regularly because it keeps making connections that don't exist. For my translation work, for example, it sometimes ties entirely new, unrelated things to old ones. And it'll hyper-focus on examples rather than take the message as a whole.
but is it "open"? still the hypocrisy of openai
Yes the API is availale... which is a step towards open....and gpt-4o is available to free users😂
Look at what happened to high end open-source alternatives like stability AI.
They are about to bankrupt themselves.
OpenAI's initial strategy always involved this route considering compute and data is the most important thing.
Beyond 10 trillion parameters, there would be a point where open source just won't be able to compete just cuz of the financial limitations
ur mom is open
The older one will be open eventually
Bruh how greedy are you? They're doing such great stuff already and are even allowing us to use their APIs to create stuff and earn money. You want them to make all their shit public so they can't even make money themselves and just sacrifice all their money for the greater good while everyone else profits of their hard work?
This is much more of a marketing move than any substantial technological advancement. People are already used to ChatGPT and most can easily tell how formulaic it is, so that initial fascination of conversing with an intelligent computer is gone. OpenAI wants to be back at the spot light. They are, at the end of the day, a company like any other. So to bring that illusion back, they go the old route of introducing gimmicks. Now they make the computer "talk" in real-time, but deep down it's the exact same technology. Everything else is a puff of hot air to seduce the masses for a few weeks more and convince investors to continue spending money on their promises. Everything we are seeing at this point is just a subtle refinement of the same basic technology from 2017(transformers). For as much as this channel likes to talk about papers, the one paper that matters here is "Attention is All You Need" from 2017. As long as GPT requires the T, it will just be an implementation of a nearly decade old technology on top of powerful modern machines. We see the same thing in consoles, where graphics improve after a few years, but ultimately it's the same technology, held back by the same old limitations underneath.
I can see this being massively beneficial with an implant in to the brain being able to speak with someone other than your self in your own mind would be crazy
I feel like kids are gonna laugh at us looking back at how simple and goofy this looks breathing heavy to make it say are you okay?
If it's not voiced by Scarlett Johansson, i don't really care.
Rename it to ClosedAI already
It's "Open" for them - not for us xD
I heard somewhere that "Open" was supposed to refer to "readily available". That counters this meme.
@@Brahvim they were referring to open source, writing about open source, and they were uncommercial and somehow became commercial. It's practically illegal what they did, because you know, - they save on taxes this way. Illegally.
@@jerrygreenest I _know_ that...!
I've heard this memey-saying countless times on Reddit. I simply wanted to clarify.
Perhaps I should not think _so ahead_ without informing enough in social scenarios!...
I'd use the term "FREE" VERRRRY Loosely. Made the mistake of testing its image analyzation skills. Took 2 images and 5 text prompts to use up all my GTP-4o limit for the day.
To summarize the video: just subscribe to the open AI channel
Closed hypocrisy AI is doing impressive stuff.