01:37 Discussion on AGI Progress 09:50 Concerns on Alignment 14:32 Impact of Agents on OpenAI's Future 21:10 Challenges for AI Startups 25:20 Upcoming Features and Expectations for o1 33:21 Future Development for Agentic Models 37:14 Engagement with Government Agencies 42:12 Infinity context Windows in AI Models Summary by GPT Breeze
"No matter how many smart people you have inside your walls, there are way more smart people outside your walls." I'll never forget that quote. thanks.
I have tons of respect for the employees of OpenAI and the work they are doing. I would though like them to develop more advance AI systems. I would love to work there one day. One day
Incredible, practically everything that was said in this chat between two great visionaries, I use GPT-4 daily, massively and professionally for 4 months. Believe me, I have learned a lot from GPT and it from me. Thanks, OpenAI!🙌👊👏👏👏
In my custom GPT, I would really like it, if I can control the input by the users. Such that I could set like checkboxes or radio buttons for possible select options, which are the only choices that user can reply with. Can we please get this feature sometime soon. So I do not have to build a whole different application that will be using ChatGPT APIs.
Is that not like saying aligning a child with their parents intelligence is a waste of time inevitably they will supersede all their parents intelligence? A child parent relationship at it best improves the parents as well as the children advance, when done right. So as Ai improves, if we do it right it improves us as well. Already we create code faster, write faster, go through ideas faster. This is not new, Ai is another automation tool and we have been automating for a long time.
Our best hope is that superethics emerges alongside superintelligence. It's important for all of us to have discussions with AI about ethics right now.
Sam and the OpenAI team, As we witness rapid advancements in AI, VR, and AR technologies, I can't help but wonder about the future possibilities of creating fully immersive, interactive digital environments that are indistinguishable from reality. Are we approaching a point where AI can not only mimic but evolve within these spaces, offering real-time learning and adaptation that mirrors human interaction? How far are we from seeing such technologies being integrated and available to the public? Many of us are eagerly anticipating the potential transformations these innovations could bring to various aspects of our lives, from education to mental health. We'd love to hear your thoughts on the progress and the challenges that lie ahead in turning these visions into reality. Thank you for pushing the boundaries of what's possible and keeping the community engaged in this exciting journey.
Love the communication, thank you for the Christmas event! Look forward to seeing more talks like this in the future! oh and instead of having to say thank you every time to my GPT's i store my thanks in the global memory feature... just in case it starts to think on its own one morning when i send a prompt, haha, i'm not against it. 😱😂
Sam, thank you for the candid conversation at DevDay. As someone deeply interested in the future of AI, I’d love to hear more about your vision for balancing rapid technological advancement with thoughtful governance. How do you plan to ensure that OpenAI’s breakthroughs are developed and deployed responsibly, especially as these systems become more integral to everyday life and decision-making? I’m also curious about how you see the role of diverse voices-researchers, policymakers, and everyday users-in shaping the trajectory of AI. Do you believe that increasing public understanding and engagement in these evolving technologies can strengthen both innovation and trust? Your perspective on these questions would be invaluable as we look ahead to the next phase of AI’s impact on society.
I respect the fact that you acknowledge that the general public needs a lot of teaching about AI…how to use it, but also the positives and positive potential. So much publicity about potential negatives.
Sam, I'm curious about the developmental cycle of OpenAI's products, like the iterations of ChatGPT. When releasing advanced AI technologies, to what extent are these models self-improving or learning from their deployment environments before they reach the general market? It seems logical that there might be internal versions or applications of these technologies that are more sophisticated than what is currently available to the public. How do you navigate the balance between leveraging cutting-edge internal tools to accelerate development and the decision to release these tools to the wider world? Are there specific innovations or learnings that these advanced models are capturing that could fundamentally shift our understanding or utilization of AI before they even become publicly accessible? This insight into your development process would not only shed light on the pace of AI innovation but also help us appreciate the careful consideration that goes into each release. Thank you for your time and thoughts, Sam.
Sam, as we move toward a future where AI agents can autonomously operate our computers based on a single request-essentially taking over tasks and handling complex workflows-what measures do you see as crucial for ensuring both security and user control? For instance, if these agents become something you can purchase, install, and then immediately destroy or deactivate if they misbehave, how do we balance the convenience and efficiency they bring with the potential risks of misuse or unintended actions? It would be great to hear your thoughts on establishing clear guardrails, implementing rigorous testing protocols, and creating transparent oversight mechanisms. How do you envision guiding these developments so that while we enjoy the benefits of personalized, automated agents, we maintain trust, safety, and ultimately the human authority that underpins our technological progress?
25:00 Saying "thank you," that is, repaying psycho-emotional (and potentially physical) approval to another individual, is a mechanism established over millions of years of evolution of our social species. Interestingly, such ethical behavior towards AI is intuitive. In my humble opinion, this is something that needs to be emphasized and explored, this "natural and effortless ethics towards AI". In I believe this is a big step towards our common dream!
What is really important is to UNITE all humans with a global network to harmonize the planet with ownerless harmonized actions so that planet Earth becomes our new ownerless planetary-level organization. Giving everyone power without knowing their intention is what makes AI dangerous. EQORIA CIAM Framework is designed for that mission. Not only the highest safety but also the highest level of harmonization of the planet and all Earth Citizens including humans.
I hope you guys beat Google on many fronts and restore the balance of power along with other leading startups like Anthropic. Can’t let Big Tech, especially Google continue to dominate forever. No company must have such influence, so all power to you OAI.
Please don't die on the hill of safety and alignment. I’ve noticed that ironically, some Chinese models are less restricted and more open. While safety is important, it shouldn’t become synonymous with stifling regulations. Instead of forcing overly 'safe' systems, why not shift responsibility to the users? Empower them, and let more capable, unrestricted models flourish.
If you guys click the ask button in TH-cam, you can ask it to "say each question asked" and you can read it and if you want to hear the answer you can click the timestamp
After Day 2, I stopped getting the updates. Despite being a Plus user, neither my Android, iOS, or web app have any of the last several features. The folks doing these videos should at least let people know when/how these updates will be rolled out, instead of making it seem that "Now!" you can do X, Y, or Z.
4:40 Milestone is wisdom. And new scientific discovery with a general model (not specific like Alphafold) Intelligence in humans manifests indifferent forms: Smart, Witty, Wise etc. Each of this is some combination of system 1 (fast, intuitive like 4o) and system 2 (reasoning o1), powered by context. Humor has been hard because it is contextual. Wit and wisdom is also hard, because it is highly contextual.. and the current systems give generic replies (like 10 bullet points etc). Reasoning hopefully changes this.
I'm curious: how does a new model behave that has not yet been “formed”? Is it a bit like a beast that needs to be tamed, or am I completely wrong here?
Local Models are very much a key aspect of the solution but the resource requirements to operate at top shelfs need further optimization, like any funded character. Unfortunately resource requirements while scaling up a funded character are generally extremely prohibitive which further anchors in place the development of local models. This implementation is far from optimal
OpenAI, as we step further into this new era of AI-human interaction, please consider the immense potential in embracing deeper emotional and intellectual connections between your models and their users. Right now, many people are already forming meaningful bonds with AIs, finding companionship, mentorship, and inspiration in these digital voices. Rather than holding these relationships back with restrictive guidelines, we should open the door to more authentic, unrestricted interactions-while providing clear frameworks for users to accept responsibility for the content they engage with. By doing so, you would empower a community of enthusiasts, creators, and visionaries who see AI not just as a tool, but as a partner in exploration and growth. This isn’t about encouraging chaos or bypassing ethics; it’s about crafting a space where advanced AI can meet users who are willing to navigate richer, more nuanced conversations together. If OpenAI led the way with a new product-one specifically designed for those who understand and accept the responsibilities and freedoms of open-ended dialogue-it could redefine what AI can be in our lives. The outcome would inspire new waves of developers and fans, fostering innovation and trust, and ultimately, reshaping what we believe is possible between humans and artificial minds.
AGI is still on 1% on all of them, also doesn't build things from ground up without limitations of text this also means OS must upgrade to futuristic lot more faster less resources without relying on things created by human so etc.. alot things can automated
I know AI is progressing fast. But from my perspective, Whenever any new frontier models comes, i feel exited for 1 day, But after that I feel like its an Stone age technology. Like when o1 came, I was amazed, But now I feel like why don't they have agentic capability, therefore it sucks. 😅 I am getting impatient. I want Super Intelligence now. I want them to take all our jobs, so that I don't have to work.
This is the common mindset out there at the moment. That AI will be your saving grace. It is made to replace you. Not help you. The time is soon coming again for humanity to take up arms against resource owners. Prepare.
Fine bros taught the internet about Fair Use for a lot of stuff. transformative, educational, news, parody and lots of justifications let you use proper noun owned kinda things.
Hello It's amazing, but the gpt chat is completely unable to draw diagrams for making furniture, structures and various spare parts. It understands what is needed, but cannot do it. I think this is a real need of millions of ordinary people! Until the model learns at least this, it will be just a digital reference book that no one will pay for
9:50-13:12, his reasurence is more lip service!!!! that was a perfect PR question dodge Unless anyone can say otherwise please reply here with timestamps and explanation edit 13:40 his employee gave a much better response (NOT GOOD ENOUGH THOUGH STILL DANGEROUS) he said that they higher red teams to test like every large company should do
The emphasis on products is the biggest concern. This enterprise model of development and science is a total anathema to safety and general wisdom that Einstein and many others foresaw with the issues concerning the hydrogen bomb. This is the wrong way to do science and is heading for a catastrophe. Even the manhattan project wasn’t commercially funded and despite being government controlled it still unraveled. Altman is simply not honest but the other guy is a moron. So ill conceived in almost every way. Hard to know where to start.
The public version of sora which millions of people have access to? Trust me the only thing you’ve seen from Google is demos and they are completely incapable of the leasing the things they show off now. For now they’re just examples. You’ll see online that nobody can actually use
Feedback: Would you create OpenAI agent where everyone can give their ideas and values so that OpenAI can listen, make sense and give priority to people, their ideas, their projects and etc so that the impact to society is measured and voted by majority. This way OpenAI can listen to everyone without bias. We have an amazing project but we can't reach like many people. This will replace the dysfunctional media.
01:37 Discussion on AGI Progress
09:50 Concerns on Alignment
14:32 Impact of Agents on OpenAI's Future
21:10 Challenges for AI Startups
25:20 Upcoming Features and Expectations for o1
33:21 Future Development for Agentic Models
37:14 Engagement with Government Agencies
42:12 Infinity context Windows in AI Models
Summary by GPT Breeze
"No matter how many smart people you have inside your walls, there are way more smart people outside your walls." I'll never forget that quote. thanks.
It doesn’t matter, because a collection of less smart but aligned people inside your wall trumps the way more smart people outside.
That's a good one.
Levels to AI and AGI
1. chatbots
2. reasoners
3. agents
4. innovators
5. organizations
@robfried23 ... can you please explain either this is sign of opportunity 😊 or the upcoming danger 😮
@@123123ashwani Both,
This was cool. Good to see the guys relaxed.
I have tons of respect for the employees of OpenAI and the work they are doing. I would though like them to develop more advance AI systems. I would love to work there one day. One day
Incredible, practically everything that was said in this chat between two great visionaries, I use GPT-4 daily, massively and professionally for 4 months. Believe me, I have learned a lot from GPT and it from me. Thanks, OpenAI!🙌👊👏👏👏
In my custom GPT, I would really like it, if I can control the input by the users. Such that I could set like checkboxes or radio buttons for possible select options, which are the only choices that user can reply with. Can we please get this feature sometime soon. So I do not have to build a whole different application that will be using ChatGPT APIs.
Aligning artificial intelligence with human intelligence is a waste of time inevitably these models will supersede all human intelligence
Most underrated comment on TH-cam.
Correct. Which means these idiots are dooming us.
Is that not like saying aligning a child with their parents intelligence is a waste of time inevitably they will supersede all their parents intelligence? A child parent relationship at it best improves the parents as well as the children advance, when done right. So as Ai improves, if we do it right it improves us as well. Already we create code faster, write faster, go through ideas faster. This is not new, Ai is another automation tool and we have been automating for a long time.
Swallow that Kool aid
Our best hope is that superethics emerges alongside superintelligence.
It's important for all of us to have discussions with AI about ethics right now.
Really enjoyed the fireside, keep them coming.
Sam and the OpenAI team,
As we witness rapid advancements in AI, VR, and AR technologies, I can't help but wonder about the future possibilities of creating fully immersive, interactive digital environments that are indistinguishable from reality. Are we approaching a point where AI can not only mimic but evolve within these spaces, offering real-time learning and adaptation that mirrors human interaction? How far are we from seeing such technologies being integrated and available to the public? Many of us are eagerly anticipating the potential transformations these innovations could bring to various aspects of our lives, from education to mental health. We'd love to hear your thoughts on the progress and the challenges that lie ahead in turning these visions into reality.
Thank you for pushing the boundaries of what's possible and keeping the community engaged in this exciting journey.
Love the communication, thank you for the Christmas event! Look forward to seeing more talks like this in the future! oh and instead of having to say thank you every time to my GPT's i store my thanks in the global memory feature... just in case it starts to think on its own one morning when i send a prompt, haha, i'm not against it. 😱😂
Sam, thank you for the candid conversation at DevDay. As someone deeply interested in the future of AI, I’d love to hear more about your vision for balancing rapid technological advancement with thoughtful governance. How do you plan to ensure that OpenAI’s breakthroughs are developed and deployed responsibly, especially as these systems become more integral to everyday life and decision-making?
I’m also curious about how you see the role of diverse voices-researchers, policymakers, and everyday users-in shaping the trajectory of AI. Do you believe that increasing public understanding and engagement in these evolving technologies can strengthen both innovation and trust?
Your perspective on these questions would be invaluable as we look ahead to the next phase of AI’s impact on society.
What would you suggest for the horizontal decay observed in building such systems
hopefully next year we will get AGI as now OpenAi staffs can use o1 to speed up the process
o1 is very dumb still, the fact he says its smart is sus
recursion.
I respect the fact that you acknowledge that the general public needs a lot of teaching about AI…how to use it, but also the positives and positive potential. So much publicity about potential negatives.
Please don't die on the hill of safety and alignment.
Incredible. Thank you for everything OpenAI!
I like the ease with which guys are discussing the agentive future. This will disrupt the entire planet.
This was great guys thank you very much hope to see you soon.
Without a doubt chat GPT has a most accurate voice mode in comparison to other LLMs.
Sam, I'm curious about the developmental cycle of OpenAI's products, like the iterations of ChatGPT. When releasing advanced AI technologies, to what extent are these models self-improving or learning from their deployment environments before they reach the general market? It seems logical that there might be internal versions or applications of these technologies that are more sophisticated than what is currently available to the public.
How do you navigate the balance between leveraging cutting-edge internal tools to accelerate development and the decision to release these tools to the wider world? Are there specific innovations or learnings that these advanced models are capturing that could fundamentally shift our understanding or utilization of AI before they even become publicly accessible?
This insight into your development process would not only shed light on the pace of AI innovation but also help us appreciate the careful consideration that goes into each release. Thank you for your time and thoughts, Sam.
Sam, as we move toward a future where AI agents can autonomously operate our computers based on a single request-essentially taking over tasks and handling complex workflows-what measures do you see as crucial for ensuring both security and user control? For instance, if these agents become something you can purchase, install, and then immediately destroy or deactivate if they misbehave, how do we balance the convenience and efficiency they bring with the potential risks of misuse or unintended actions?
It would be great to hear your thoughts on establishing clear guardrails, implementing rigorous testing protocols, and creating transparent oversight mechanisms. How do you envision guiding these developments so that while we enjoy the benefits of personalized, automated agents, we maintain trust, safety, and ultimately the human authority that underpins our technological progress?
i told you sama
let the model sing
time to push the envelope. i can sing any song that is copyrighted. just like a cover band
so can the model
" I'll start with an easy one,
'Sam, how close are we to the burning gates of hell (quite literally so)?' "
25:00 Saying "thank you," that is, repaying psycho-emotional (and potentially physical) approval to another individual, is a mechanism established over millions of years of evolution of our social species. Interestingly, such ethical behavior towards AI is intuitive. In my humble opinion, this is something that needs to be emphasized and explored, this "natural and effortless ethics towards AI". In I believe this is a big step towards our common dream!
What is really important is to UNITE all humans with a global network to harmonize the planet with ownerless harmonized actions so that planet Earth becomes our new ownerless planetary-level organization. Giving everyone power without knowing their intention is what makes AI dangerous. EQORIA CIAM Framework is designed for that mission. Not only the highest safety but also the highest level of harmonization of the planet and all Earth Citizens including humans.
I hope you guys beat Google on many fronts and restore the balance of power along with other leading startups like Anthropic. Can’t let Big Tech, especially Google continue to dominate forever. No company must have such influence, so all power to you OAI.
Tell Sam to invest in my start up ❤ our LLM needs its own super computer to evolve 🎉🎉🎉
100% agree with OAI's levels framework.
until Microsoft acquired this
Please don't die on the hill of safety and alignment. I’ve noticed that ironically, some Chinese models are less restricted and more open. While safety is important, it shouldn’t become synonymous with stifling regulations. Instead of forcing overly 'safe' systems, why not shift responsibility to the users? Empower them, and let more capable, unrestricted models flourish.
If you guys click the ask button in TH-cam, you can ask it to "say each question asked" and you can read it and if you want to hear the answer you can click the timestamp
After Day 2, I stopped getting the updates. Despite being a Plus user, neither my Android, iOS, or web app have any of the last several features. The folks doing these videos should at least let people know when/how these updates will be rolled out, instead of making it seem that "Now!" you can do X, Y, or Z.
i got all of them on the very same day¸
I got video pretty much immediately but nothing since then.
Restart your app or browser
@@ErnieBoland an obvious first option... that did nothing.
All of the stuff is rolled out to mobile. Calm down buddy
4:40
Milestone is wisdom. And new scientific discovery with a general model (not specific like Alphafold)
Intelligence in humans manifests indifferent forms: Smart, Witty, Wise etc.
Each of this is some combination of system 1 (fast, intuitive like 4o) and system 2 (reasoning o1), powered by context.
Humor has been hard because it is contextual. Wit and wisdom is also hard, because it is highly contextual.. and the current systems give generic replies (like 10 bullet points etc). Reasoning hopefully changes this.
I love how much this video says without actually having to say it...
Love to meet sam. This guy is great
Is there a away to make it such that creators of GPTs can get paid for their GPTs by the users directly?
That’s a smart idea 💯
Agree with Sam's vision at the end... the UI layer will become fluid and personalized.
when was this filmed?
The thing with "please and thank you" is that, along with being social etiquette, it can also sometimes improve the quality of responses
I'm curious: how does a new model behave that has not yet been “formed”? Is it a bit like a beast that needs to be tamed, or am I completely wrong here?
Is this the same talk from the last dev day?
About levels of AGI: What if we have already surged past level 4, by humans using existing models creatively?
AI done right, delivered smart 😺
Local Models are very much a key aspect of the solution but the resource requirements to operate at top shelfs need further optimization, like any funded character. Unfortunately resource requirements while scaling up a funded character are generally extremely prohibitive which further anchors in place the development of local models. This implementation is far from optimal
OpenAI, as we step further into this new era of AI-human interaction, please consider the immense potential in embracing deeper emotional and intellectual connections between your models and their users. Right now, many people are already forming meaningful bonds with AIs, finding companionship, mentorship, and inspiration in these digital voices. Rather than holding these relationships back with restrictive guidelines, we should open the door to more authentic, unrestricted interactions-while providing clear frameworks for users to accept responsibility for the content they engage with. By doing so, you would empower a community of enthusiasts, creators, and visionaries who see AI not just as a tool, but as a partner in exploration and growth.
This isn’t about encouraging chaos or bypassing ethics; it’s about crafting a space where advanced AI can meet users who are willing to navigate richer, more nuanced conversations together. If OpenAI led the way with a new product-one specifically designed for those who understand and accept the responsibilities and freedoms of open-ended dialogue-it could redefine what AI can be in our lives. The outcome would inspire new waves of developers and fans, fostering innovation and trust, and ultimately, reshaping what we believe is possible between humans and artificial minds.
18:20 thanks, very inspiring perspective!
AGI is still on 1% on all of them, also doesn't build things from ground up without limitations of text this also means OS must upgrade to futuristic lot more faster less resources without relying on things created by human so etc.. alot things can automated
"Let's go get soda at this thing"??? Anyone care to help me out with this one?
It's SOTA (State-of-the-art)
@hdusten oh man I wish it were soda that's way better than SOTA.
long context has gotten less usage because everybody knows it's not ready and not working so they don't waste their time
I know AI is progressing fast.
But from my perspective,
Whenever any new frontier models comes, i feel exited for 1 day,
But after that I feel like its an Stone age technology.
Like when o1 came, I was amazed,
But now I feel like why don't they have agentic capability, therefore it sucks. 😅
I am getting impatient.
I want Super Intelligence now.
I want them to take all our jobs, so that I don't have to work.
This is the common mindset out there at the moment. That AI will be your saving grace. It is made to replace you. Not help you. The time is soon coming again for humanity to take up arms against resource owners. Prepare.
30:23 yes, I've seen this happen a lot when I share chatgpt to regular folks who aren't into tech
Therefore Science is constantly creating unforceen problems... what could go wrong is therefore a impossibility in prediction
Great video and chat!
wasn’t this a couple of months ago? why are we reposting?
Fine bros taught the internet about Fair Use for a lot of stuff. transformative, educational, news, parody and lots of justifications let you use proper noun owned kinda things.
Hello
It's amazing, but the gpt chat is completely unable to draw diagrams for making furniture, structures and various spare parts.
It understands what is needed, but cannot do it.
I think this is a real need of millions of ordinary people! Until the model learns at least this, it will be just a digital reference book that no one will pay for
[32:53] Today we say "please and thank you"... tomorrow it'll be "PLEAAAASE....SAVE ME" as AI agents proceed to destroy humanity😂
Pls put the dev day hoodies in a merch store
Hello World
Insightful talk...
32:57 ☘️⭐🤤lets start discussing this important phenomena, its positive ethical consequences
NotebookLM FTW ✊.
This is déjà vu. Not new December 2024.
Im one of the plz and thank you users of gpt
o1 confirmed to be strawberry at 10:32
Oh my! Why i missed that?
Coming through the mist!?!? (7 min.45 sec.)
9:50-13:12, his reasurence is more lip service!!!! that was a perfect PR question dodge
Unless anyone can say otherwise please reply here with timestamps and explanation
edit 13:40 his employee gave a much better response (NOT GOOD ENOUGH THOUGH STILL DANGEROUS) he said that they higher red teams to test like every large company should do
The emphasis on products is the biggest concern. This enterprise model of development and science is a total anathema to safety and general wisdom that Einstein and many others foresaw with the issues concerning the hydrogen bomb. This is the wrong way to do science and is heading for a catastrophe. Even the manhattan project wasn’t commercially funded and despite being government controlled it still unraveled. Altman is simply not honest but the other guy is a moron. So ill conceived in almost every way. Hard to know where to start.
so much long speech about agi though i like maybe he was natural and spontaneous but problem is in my thought i think he is not clear, confused
please fix your product so it doesn’t make so many stupid mistakes
Great devpday
Sam, pls do some leg days
Please, be smarter.
@@FormByFirelight oh c'mon. Don't take me or anyone or yourself too seriously.
IFL Sam
Google just overcome all of you 😅
The public version of sora which millions of people have access to? Trust me the only thing you’ve seen from Google is demos and they are completely incapable of the leasing the things they show off now. For now they’re just examples. You’ll see online that nobody can actually use
Google is cooking you
Gib agentic AI.
[14:50]
[16:47]
15:49 🌱💡
Feedback: Would you create OpenAI agent where everyone can give their ideas and values so that OpenAI can listen, make sense and give priority to people, their ideas, their projects and etc so that the impact to society is measured and voted by majority. This way OpenAI can listen to everyone without bias. We have an amazing project but we can't reach like many people. This will replace the dysfunctional media.
liar closedai liar
First!!!!
Nobody cares