STUNNING GPT-5 Details Are Finally Released....
Vložit
- čas přidán 10. 07. 2024
- Learn A.I With me - www.skool.com/postagiprepardness
🐤 Follow Me on Twitter / theaigrid
🌐 Checkout My website - theaigrid.com/
Links From Todays Video:
www.bloomberg.com/news/articl...
arxiv.org/pdf/2311.02462
www.perplexity.ai/search/what...
x.com/tsarnick/status/1803901...
x.com/Coulmock_Yov/status/179...
x.com/search?q=TSARNICK%20AGE...
x.com/tsarnick/status/1808252...
www.bloomberg.com/news/articl...
video.twimg.com/ext_tw_video/...
video.twimg.com/ext_tw_video/...
video.twimg.com/ext_tw_video/...
/ hard_to_believe_these_...
Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.
Was there anything i missed?
(For Business Enquiries) contact@theaigrid.com
#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience - Věda a technologie
![The moment we stopped understanding AI [AlexNet]](/img/n.gif)
When producing videos, please be considerate of your *viewer's time.* If there was a leak about 5.0 having PhD-level intelligence 18 months from now, please get to that point without redundancies to fill 20 minutes.
I'm not complaining, I appreciate you. However, I believe that if you appreciate your audience's time, it will lead to *more growth for your channel.*
Ty
🙌🏽
Wes Roth made the same content fit into 8 minutes. Clear and concise. This is just rambling and reiterating previous points over and over. I would love to see him release a video 3 hours later if it means he can sit down and write himself a short script to stick to rather than trying to get his video out first.
subtitle on and 1.5 speed. That should save you some time...
And timelines as well with the points of interest marked so we can get straight to it.
Aren’t we tired of the over-hype?
Its better for all of us if this technology will slow down a little bit, giving time to optimize it high power consumption and utilize usage.
Its clear what the technology can do and its limitations, unless there is a new paper out that we don’t know about many true experts have pointed out that the foundation of current LLM have reach its peak. They can keep throwing data and computing power but that won’t make them smarter. Phd level refers that can recite books as good as a Phd but they still can’t reason like a human, and that means that in a hyper-complex and nuanced world this machines are still tools and not “replacements” as this videos lead us to believe.
Other way you will get no funds 🤷🏽
@@milanpospisil8024nah, I vote for full throttle 🤘🏽
@@milanpospisil8024 I vote for full throttle 😊🤘🏽
Tbh, it's like a running joke at this point... 4.o isn't even fully released and we're already rustling the vines with 5.0???
wdym? Isn’t free for everyone rn?
@@user-gs7pr7zi8kI think he's referencing some of the things 4o was supposed to come with as shown in the demo.
It's just a version update .. just that every update is miles (or so) ahead from the other versions.
Supposedly that stuff is to be released by the end of this month, so we are getting close🤞@@georgemontgomery1892
5? This is so old tech... 6 is on the way! Wait... I mean 7. Wait....
The new voice engine isn't even out yet.. LOL
They are likely worried about getting sued by sound alike or liable for illegal activity and abuse. I'm sure scammers will take advantage of it.
I wish I didn't know about it. It's a tease. At least it's going to probably be better than the demo.
This will be released after the elections, so serveral months from now
Keep the hype train moving!!!
@@trappedcat3615 Well, that's like saying remove all phones and computers since scammers take advantage of those things as well.
Given that the whole point of having a PhD is to generate new knowledge, which Gpt 5 is unlikely to do, it seems misguided to says it’s at the level of a PhD as opposed to a Masters degree
Agreed! We don’t have a general intelligence nor is the current paradigm likely to give us that. That being said, we don’t know the details of how OpenAI has built the thing.
Good point, and this is likely. However, enough knowledge may give a system better pattern recognizing skills. Especially with statistical analysis. Data gathering is one thing, knowing how to apply it to create new insights it something ai might be able to do.
With enough reasoning ability and the tools to complete tasks it will replace you. It doesn't need to create new knowledge. How many times a day does the average employee who is encumbered by company SOPs and policies come up with new knowledge.
Exactly! To me models are like infants/children, not high schoolers (where they claim they are at). They hallucinate or make stuff up like a 4 or 5 year old would. Only in some narrow ways do they reach higher levels, but they don't exhibit near enough types of intelligence to compare to humans...
I think it will be able to generate new knowledge or at least new insights. Sounds like they are talking about clinician-type PhDs or those not in research.
The idea that they would only give access to the elites and not the people is concerning.
Dude AGI would never ever be released to the average person or the public as a whole.. If it was an AGI then by definition it could autonomously navigate the internet to perform real tasks. Thus if it was an AGI the first thing every user would do is prompt it to go and make them a million dollars or some similar amount of money. 1.This can never and would never be allowed. 2. if they "red teamed" or "patched" the AGI such that the presumably millions of ways a PHD level AGI could operate in the world to make money, then the result would be a lobotomized AI system that in no way resembles an AGI... The public will never get an AGI. Don't wish for it, don't hope for it, just accept what we have, it's never going to get any better.
GPT-4:
Level 1 (Chatbots): Excels at this level, demonstrating advanced conversational abilities.
Level 2 (Reasoners): Shows strong capabilities in problem-solving, often at or near human level for many tasks.
Level 3 (Agents): Limited. While it can suggest actions, it cannot directly take them.
Level 4 (Innovators): Shows potential in aiding invention through idea generation and problem analysis.
Level 5 (Organizations): Limited. Can assist with organizational tasks but cannot autonomously perform them.
Estimated Level: Strong 2, with elements of 4.
Hypothetical GPT-5:
Level 1 (Chatbots): Would likely exceed human-level conversational abilities.
Level 2 (Reasoners): Would likely surpass human-level problem solving in most domains.
Level 3 (Agents): Might show improved capabilities in action planning, but still limited in direct action.
Level 4 (Innovators): Could show enhanced ability to aid in invention and creative problem-solving.
Level 5 (Organizations): Might show improved capabilities in coordinating complex tasks, but still limited by lack of direct agency.
Estimated Level: Strong 2, significant 4, with potential elements of 3 and 5.
Your sora theory has huge implications - the elites having an AI system light-years ahead of what the public has ... exactly what Open AI was meant to prevent
It’s starting to worry me 😑 Unbelievable that Musk and Zuck may be the hero’s now 😐🤦🏻♂️
@@tituscrow4951 And Google... I mean, they are trying lol hope they stop to suck at AI
I know people who have access to Sora and they say Its no better than RunwayML's Gen3 Alpha, only much more resource intensive.
People got to stop with this child/high schooler/PhD comparison, it's such a bad analogy. GPT-4 is not a high smart schooler level. It's much smarter and much dumber at the same time. It knows more than all the high schoolers on the planet together and can do lots of things even the smartest high schooler won't be able to do. At the same time, it would fail at some easy tasks an elementary kid would do, just because it relies too much on the training and often fails miserably when tasks are a bit different from the training data. So that brings us to the PhD comparison? Is it gonna be just being able to solve some of the things PhD could do? Or is it about doing original research PhD could do? PhD is not just about solving textbook problems, it's about providing refined original ideas.
I don't understand why authors of channels like this one can't be just a little bit more sceptical, without eating up and hyping up anything AI companies bring up. That would make such a better content
Nice grammar
Given most any topic you will find both skepticism and belief. In this case he's a believer in the technology. Massive amounts of comments on channels such as these blindly list things that our current technology can't do, while ignoring the fact that many of those things *are* actively being done and that the issue is their lack of serious testing (hands on) and creativity. It works both ways and I for one am incredibly happy that there is a broad range of opinions on the topic, because it brings forward a bunch of worthwhile questions and ideas.
I'd say it's more so about the level of reasoning and problem solving capabilities rather than intelligence as a whole, when it comes to these comparisons. Which makes sense to confine the context to as it's an important part of intelligence that it's lacking that is also practically quantifiable through exams/tasks at those levels (high school/masters/PHD/etc.). Knowledge isn't much of an issue, like you said it's already way beyond most people when it comes to that.
Btw as for questioning the legitimacy of those claims when it comes to PHD level I do agree, am very curious to see the benchmark on that.
The terminology is used take make it easier for the public to understand.
As you pointed out, this suggests that GPT5 may not be deployed as a chatbot, meaning ChatGPT and GPTx are going to diverge as different product lines.
Also, it would be VERY helpful if you included sources (as links) in your video description - at least the sources shown on screen. For example, if I wanted to watch the rest of the talk by Dario Amodei, I would have no idea which one it was...
inlcuded now
Talk about long-winded.
OpenAI is going to give Sora to the gatekeepers. That's rich. Innovation in filmmaking just took a leap backwards.
and GPT-5 for corporate use only - they go fully commercial, making the rich richer and causing unemployment among the poor. I think this is why a lot of people left OpenAI.
@@micbab-vg2muI had this same thought. We’re taking the 1st slide into dystopia. I’m now rooting for everyone else in the Ai space. Closed Ai ain’t it… Did you see Sam Altman supercar?
And the executives and investors will be shocked when nobody consumes their content because the cut out all of the creatives and mass produced whatever their algorithm told them would be popular. It will figuratively be comparable to Chinese made dollar store quality products but for movies. There's already such little quality in Hollywood today and the slop that will be coming will be the final nail in the coffin.
I don't think it will matter. There's already several competing AI solutions for Sora.
@@tituscrow4951 no
I don't trust them.
They have not even released Sora and GPT-4o
They said the end of this month.
But I bet they are losing lots of sleep worrying about what you think.
@@tracy419 I, however, am gaining tons of valuable Zzz's after listing to what OpenAI thinks about my thoughts about things regarding thinking and more specifically, what the hell was I talking about?
They might just not release to the public due to how exploitative its potential is.
@@ConfusedPlushiee Well that is a really dumb reason.
@BionicAnimations Idk, Sora at least is worrisome to me. A lot of people barely use their reasoning skills when ingesting information, especially videos. So in my mind it is silly to think that people will not use this stuff to exploit others and bend their sense of reality.
In my mind I think that we won't really be able to tell what is real at all in digital space (it is already hard enough now at times). Meat space is gonna be very valuable for people's psychological well being I think.
Either way just on break at work and thought I would just type into the ether. Have a good day my friend.
A whole bunch of nothing
Agreed. Im still here trying to get GPTo to create a Eventhandler in JS for a smal Game demo and this guy is talking about Level 2+ AI with resoning etc. AI for now only got more expensive and accurate in domains where there is data to train on. But i have jet to see anyone actually doing anything impressive with this that goes beyond a junior developer. To me it looks like AI is just good for summarizing and finding knoledge semantically quicker than a google search or for writing boilerplate code... Thats already good but you guys realy need to stop hyping and try using this tool and you will see the glaring gap that new models never were able to bridge. "Its the worst it will ever be. Future Versions will be able to solve X Y Z..." im waiting. If GPT5 6 7 cost 10 times and again and again more than current models then they will not be viable reguardless of what they are capable of.
So in short: STOP HYPING > TRY WHATS POSSIBLE > IMPROVE AND INOVATE TO PUSH THE CUTTING EDGE. There are 40+ Videos on AI Agents and Frameworks and Futurepredictions but nothing actually showing anything impressive (If i see one more Snake or Twitter engagement bot im going to break)
I think OpenAI is so mighty that they expend more energy containing their power than actually putting upgrades out.
I actually think it's anthropic that's going to get there first.
Important clarification: the next frontier model that they recently started training will not be released until 2026 or so. GPT 5 or whatever they call it finished training earlier this year, and is in red team mode. It will be coming out sometime later this year (December or anytime post election?). Check Alan D Thompson’s The Memo report.
Yeah, the question is whether they’re talking about 4.5 or 5.0 or what they think 5.0 will be able to do. The article says it’s a research project based on 4o though. I could see them getting a big improvement in reasoning with just self-reflection and perhaps some kind of an iterative ensemble system. (Set up external feedback that would guide multiple instances “confer” or debate with each other to yield superior reasoning than the base model is capable of, zero-shot.)
@@DaveEtchells exactly. Good observation
True true true. My brother does cyber security for NASA. They have already seen demos of GPT-5 fine tuned to operate their software.
@@user-ty9ho4ct4k Interesting. Does it look like to him and colleagues that LLMs are plateauing from demos of GPT5? That seems to be the impression I get from hearing 2nd hand about other businesses that got early sight of GPT5.
Conservative public do not like big jumps or any jump at all, criticized SORA's demos, public criticized SUNO and UDIO music, criticized Figure 1 demo.
If they choose to give more power to well established big monopolies like Hollywood and armies over public i wouldn't be surprised at all. Welcome to the Crystal Era.
LoL you ever tried using it to program python language. 6 hours later and a lot of mistakes the code won't be right. Same thing with HTML. It can't do ph D level stuff lol 😆 it's computer code trying to mimic human response.
@TheAiGrid There will be LOTS of "Reasoner" LLMs soon. Haven't you read the "Grokking" papers ?!?
By over training the model to the extreme an new emergent behaviour occurs...REASONING. A paper was published on May 27th the SOLVES the Grokking technique and a github project called "GrokFast" cuts the training time for this by up to 50X !!!
We should be flooded with "Grokked" models sometime in September or October.
This is NOT an OpenAI breakthrough.
Really.....once reasoning is cracked the rest becomes easier and easier. Generalised intelligence is reasoning chains all the way down, just more elaborate and over longer timescales. If this is true I'd therefore expect AGI (any definition of it that puts it on par with humans, I don't care about having a super precise definition of it) within the year.
Until it's actually released I don't believe a word they say
Exactly. 😂
Agreed, OpenAI used to be a company you could believe the demos, but today they show a demo and just dont release the products
I regret the fact that I was born in 1961, this stuff is just starting to get good and I would’ve loved a couple of years to play with it
Don’t worry you’ll have exactly same amount time to play with it s everyone else.
Can someone explain to me why we're considering current models to be "smart high schoolers" and next level models to be "PhD's" when they aren't reliable?
It doesn't matter how on-average smart it is if you have to double check the output.
The magic i get from all this is that we dont have a hard barrier in the approach, technology, theory and application.
This means we re just on our accelerating path to Ai-economy.
We need to make sure that this is for the benefit of everyone. I think that is our task.
Yes. Abundance can be for all. But what will people do for income?
Excellent content. Can you do us a favor and get an Ad Blocker on your browser, so we don't get continually visually distracted from your presentation.
If you want any CZcams content without ads you need to pay Google a subscription fee.
Kinda sad that Ai is slowing down so much right now. We used to think couple months ago that AGI is around the corner, and now I dont know if the next good update is around the corner xd
bu at the same time, I feel like we just need to push it far enough to make it truly understand things, and reason to innovate, create and improve to see next Ai explosion.
Nvidea is getting major AI core improvements that havent yet gone to OpenAi. If OpenAi then also improves on current models at the speed theyve been going, then shit will go fast enough soon enough. I just hope the political landscape will allow us to create utopia instead of regressing us to a post-human dystopia.
@@ConfusedPlushiee I hope so too. I hope....
Sam Altman specifically said that they recognize that the public do not like big jumps which would be startling and so they will try to make a iterative improvements a.k.a. GPT 4.0,4.5, 4.5 turbo, 4.o etc.
and he wouldn't be more wrong, we love big drops and big updates
It has nothing to do with what the public wants.
It is entirely based on two things:
1. They need the military industrial complex's blessing, and
2. Too much advancement/attention == regulation.
@@itsallgoodaversa well, they need more power to introduce models at a large scale. So it's more like they can't release more intelligent models too fast at a large scale. They don't have a choice until they get a few reactors running.
@@edgaralwe love it, but not the general public, especially in the political climate in America. People were freaking out when Sora was first demonstrated. Man, I wish they went fast, but they know the public backlash from the Luddites.
Why do you think these 💉 were and are for???
Pretty pretty
truly
Also "really really" and "crazy"
The AGI levels all seem pretty nebulous and not very useful. For example, ai chatbots can already 'aid in invention', whatever that means. They can bring up issues you didn't consider and relevant topics that might apply. Even if only one of the ten suggestions it gives are actually actionable, it's still 'aiding in invention'.
The amount of simpletons who:
- overestimate how good this crap is now
- cheer for the superior beings to be created and take control and cause our extinction if this is ever finished in next 10-‘15 years
Is absolutely insane..
And where are minuses? 🤔
Maybe a student with PHD in English or Literature with only limited "book smarts" with no other smarts and a photographic memory that has memorized every piece of text on the internet. Not even great book smarts, as most people don't have photographic memories so they still need to be able to reason and figure out the answers based on their studies for questions they haven't seen before. Anything that would take experience or being able to adapt to something outside of the normal or most predictable outcomes, these LLM models fail at. They are more akin to an idiot savant.
When the transformer based LLMs are able to integrate other deep neural networks functions like DeepMind's AlphaGo and are able to simulate and learn through trial and error to gain that experience, then I would think they are closer to being able to match at least some aspects of human intelligence. Not to mention they still need to do a lot better at basic reasoning, planning ahead, and answering questions that require gaining additional insights, knowledge, and correlations before answering.
Govts will have agi and asi before all...
So more announcements, but not product?
Could you please use an adblocker if you show browser windows? It's very distracting to have random video ads play around while trying to read text.
Mira Murati said "PhD in specific tasks" - sounds like it could be some kind of overfitting of known problems. They seem to use the term PhD quite lightly here. I will only accept that it is PhD level if it can do arrive at conclusions that are truly new to humanity.
I can tell you have made a real effort into cleaning up your use of filler words. Bravo 🎉
This is going to revolutionize programming. As a senior engineer with 10 years of experience in web apps, I've doubled my productivity over the past few years
Rather than this list of steps, I am more interested in why Mira Murati said it would take a year and a half before the next breakthrough comes out, but she can already see what'll be in it. I wish the interviewer had asked her what could be some of the reasons for the long lead time. I can't visualize how they can be testing and seeing something that works in the lab (which the MS-guy Kevin Scott said he was already seeing now), but still say "ah, I guess it'll probably be 18 months before we can put this capability into GPT." In some way, it must be a complete departure from the current architecture, or... what? I wish I had more quotes to allow me to make more sense of it.
Power constraints and economic conditions. it has to work at scale for 3/4th of the Ai world using it if it’s what they say. And then. Can we make a profit on it. Then ( maybe ) is the government supervised red-team tests coming back clean. Because now we’re talking about systems that can actually do very dangerous things untrammelled.
HOLLYWOOD????? That is the last place I would let have a super powerful AI ... JC
So OpenAI's idea of "safety" is to put the tools into the hands of existing powerful agents, figuring they'll use the tech to maintain their status, where if released freely to the world, there would be a big competitive scramble for new dominance, which might mean giving real AGI/ASI more freedom. It's not the worst theory, I guess.
It doesn't eliminate existing power struggles (e.g. China vs USA vs Russia - where OpenAI is pretty clearly aligning itself with 'the West'), and there's always the potential for open source to be sufficiently empowering of small actors to create problems. But it is 'something' at least.
More Hype, Less Substance. Every day low expectations.
That’s scary for coding, based on the mountain of PhD-level code I had to completely rewrite.
Sad
Phd level code you had to rewrite isn't Phd level code.
PhD level means they are giving their own technology and code away.They would never even if they have achieved it
Nice video, I noticed that you’ve slowed down your speaking speed and cadence as well as reducing the repetition of words.
7:55 Leaving us hanging!
I'm not sure education level is a measure of intelligence. Seems like it's more of measure of knowledge than intelligence. Sure, AI can have tremendous access to knowledge, but does that make it a Newton, Rembrandt, or Beethoven?
PHD level? I will be happy if chatGPT will analyze simple Excel tables without making mistakes and generate simple graph images without experiencing problems each 5 minutes.
When chat gpt can count the number of Rs in strawberry, ill believe its high school level
Ask it to count the number of p's in рerhaрs. And don't type it, but copy exactly what I wrote. I expect it to fail.
Know why? Because those aren't ordinary p's in "рerhaрs"; those are the Russian letter that is equivalent to r.
This is a continuation of Ray Kurzweils "Law of Accelerating Returns" where we use today's tech to create tomorrows tech. This is what give its exponential growth. Now when an A.I. agent begins to directly assist with innovation this will be knee of the curve on the exponential path.
Public release is not a very good indicator of advancement.
Most of the serious research done in the world is done as PhD research, so I don't buy it. That would certainly be AGI. I think it won't be *anywhere near* that level, and OpenAI are going to get slapped down really fast if they try marketing their next system that way. If they really start doing that, I'd say they've run into major problems scaling and are planning a pump and dump!
To be completely honest about it, I think there's just a misunderstanding here. I could believe they can solve small problems as well as a talented high school student, that's it.
PHD without tools. What I got from that was a system with a lot of general knowledge & good self awareness & autonomy. Like a PHD student just accepted & discussing their dissertation with the institution. So before drilling into a lot of datasets & standing on decades of other research. Another words it’s just the possibility of intelligence but very unfocused.
There are chances here that GPT-5 can solve complex math problems like "how much is 2+2, beware this is a logical riddle", or logic ones like "I have 2 sisters and 3 brothers, how many kids are in my family", exciting times!
[TAS] Speedrun to unemployment 100%
GPT-5 for corporate use only, just like SORA is exclusively for Hollywood. Why pay humans when you can build an army of PhD AI agents?
[TAS] Help TAS bot is taking over the game by itself oh no he called the cops on me oh no aaaahhh 100%
Its nice so hear that stuff from openai but i think we should give them more time, training costs time and energy, i think honestly the next BIG step/update/next thing comes when Stargate is finished and running for a while, until then i think we get some cool updates like memory or some sort of low level agents
1.5 years? So basically everyone else is going to catch up with/surpass OpenAI.
Jesus, they are going full own grift mode now. Gpt 4.0 cant even generate an image without hallucinations. Ask it to draw a transformer model and it can't even do that. Much less attending conferences, giving 30 minute plus overviews of its research, navigating through windows 11 to create power points or run experiments, giving updates to advisor, collaborating with lab mates, etc.
No shit.. its a chatbot
A good one, but still a chatbot
Making declarations is not the same as having it. You need the actual algorithms and they aren't talking about that
Sounds like they just wanna stay in the new cycle at this point.
Truly pretty crazy.
Truly pretty really crazy
Basically
ok but where is 4o voice features ?
All I want is for it to roleplay as Goku being my gym trainer, hope that PhD helps
Honest question: why do you have subtitles that do not match what you say? I suggest using Whisper to create your subtitles for more accurate (and less distracting) text on your videos.
I want to know how much more electricity GPT 5 will use. If they're just throwing more processors at the problem, it'll quickly become a problem because building new power stations is really expensive, so only large companies will be able to afford it. If they're writing code itself more intelligently then that's super exciting.
✦
GPT-5, also known as SLG, is the next level of AI with reasoning capabilities.
00:01
✦
Advancement from lower AI levels to organizational level AI.
01:40
✦
Next GPT model expected to be at PhD level intelligence
05:16
✦
GPT-5 details reveal advancements, potential release in 2025
07:16
✦
Le modèle GPT-5 intègre des flux de travail agents pour pousser ses capacités encore plus loin.
10:28
✦
GPT-5 is working towards achieving Artificial General Intelligence (AGI)
12:12
✦
GPT-5 models can potentially evolve into human-level problem solvers
15:29
✦
GPT-5 is expected to have a significant impact in problem-solving, surpassing current AI systems.
17:08
✦
GPT-5 models are expected to be rapidly deployed
19:59
✦
GPT-5 reliability and internal working
21:22
They need to raise more cash to handle their coming legal battles
Am I the only one who got distracted by keanu reeves
Was anyone else distracted by that ad featuring Keanu Reeves on the right? 0:56
It's pretty remarkable how we got AI that surpassed the cognitive ability of the leader of the most powerful nation and arguably the most powerful human on earth and NOW it's almost ready to get it's PHD.
Sounds like Open AI are not open at all. I don’t like this change of direction so I’ll look beyond open Ai
I never understood, how can a GPT research. It predicts the next Token.
How can it research??
Someone explain me??
Its about time you know how to use google search before trying ai
It can't. They are engaging in a grift at this point.
It does predict the next token based on finetuned data and stuff
I mean, if you want to be philosophical about it: research is mostly just extrapolating from existing data and reasoning from that. If AI can do that, then shit will go FAST.
@@ConfusedPlushieewell, predicting the next token is not actually reasoning.
That means that it just iterates through the output of the LLM?
And she said: „for certain tasks“. Machines based on LLM can‘t think.
They haven’t figured out how to make Sora profitable yet. Same with, probably, a lot of upcoming technology. Keep up the great work🤛
Model reasoning must include the ability to understand the "What", "Why", "When" and "How" concepts of any action. It must be able to fluidly refer to any of the concepts in order to maintain stability, validity, accuracy, and security, we humans perform these checks intuitively. Someone smarter than me could say it better, but that's my two cents!
PhD before the internet or after the internet?
It is not PhD level it is a marketing Gimminck
Before, a PhD without tools was like a mathematician without a calculator or a programmer without a computer.
GPT5 is just sam altman typing really fast.
They do not deliver half of what they promised - I believe it when I see it. Now I have fully switched to Anthropic's models.
The fact they all use the exact same example feels a little bit like marketing
hopefully they dont think that the public cant have access to these kinds of systems, where the economic benefits are only available to entrenched companies.
It should be released for everyone!
What are the ethical guidelines for the accelerated development of this technology?
they went fully commercial and still act as a pro bono organization, but rich people can do what they want. I am not surprised. Ethic is just for common folks.
@micbab-vg2mu I get what you are saying. Though rich, wealthy, or poor, people can not do whatever they want when it can certainly bring horrible repercussions to society. Some wealthy people are powerful in insidious ways, and that's how they can make their unscrupulous moves. But corrupt power doesn't last too long, and amongst themselves, they annihilate each other. Ethics are for everyone. They are inescapable because of what we experience through what we learn from consequences of cause and effect. The course of life itself teaches everyone so.
Shit on Sora
There are other video generators who are releasing to the public
So in other words, OpenAI + MS says something and this is now a fact? From my humble perspective of a MSc. aren't we still in toddler level with GPT-4o (reasoning, reliability, hallucination etc.) ????
Thanks for the video.
I don't think the strategy to keep Sora behind an enterprise level paywall will actually work because it is quite clear that several other AI companies including open source are working on the exact same thing.
In my experience, 4o. can generate 90% of the work of many roles. But the key is that it does thisvery fast and for pennies.
To be honest, if we can get this level of AI the next few years would be fine with me. In the short term, I just want what Apple Intelligence promised and what Samsung is doing with translation.
Seems like it could go both ways. Who is to be determined anyone is qualified enough to use these next level AI? What if the power still end up in the wrong hands
PHd level for tasks ... yes, but not for interaction. Its still going to feel like a machine no matter how much you teach it
Maybe They are not need to teaching any of things. Because that is a skill of PhD level.
Ai for all humanity. "I think they will only release to companys" I mean I need my money back then and all investors.
i do wonder if these things are probably going to be extremely different. 👍 enlightening.
Having taken PhD qualifying exams, I would say GPT-4o could pass them now, as long as it isn't a math PhD.
Sounds like OpenAI's "level 3" is nearly achieved as soon as they achieve 'level 2' - we know how to make a chatbot into an agent - they just perform poorly, mostly due to poor reasoning/problem solving abilities. Even 'level 4' doesn't sound out of the question - put a team of level 2 agents to work, throw in some randomness along with goal-driven reasoning, and invention seems likely to fall out of it. Level 5/organizations might still be beyond a level 2 general intelligence, if only in terms of amount of processing power required.
Love your work fella. Excellemnt content.
there’s gotta be some tomatoes I can throw around here 🍅
I’m betting they’ll avoid using numbers in their names. GPT-5 will be GPT-R (-R is for reasoning)
Sounds like an excellent time to purchase a secluded hut.. On the moon.
Why doesnt the text gpt 4o feel as good as the/its voice mod that they showed in the demo. It doesnt feel right.
All sounds like pie the sky marketing hype if you ask me.
At this point I am losing interest in what OpenAi have to say about their future models. I've got GPT4, which is good enough. When they actually release Omni, I will believe it. When they actually release GPTS, I will believe it.
Until they do something new with what I use I'm switching off from the whole subject of systems better than this current released generation.
Mentioning all those academic capabilities their model will have means nothing if they don't make a difference in real-world scenarios.
Im very interested, when AI is not only relying on pre-trained data - only when it can interact and learn by its own and gains the ability to memorize - then call me impressed. As for now, AI is currently very knowledgeable but cant pick up past interactions....Bummer.
Ilya Sutskever was the brain behind this company, and he has left it.
Lvl2 -> 25
Lvl3 -> 26
Lvl4 -> 27
Lvl5 -> 28
LvlBoooom 🎉