What is OpenAI’s super-secret Project Q*? | About That
HTML-код
- Опубликовано: 23 ноя 2023
- A breakthrough in artificial intelligence at OpenAI preceded former CEO Sam Altman's firing and was part of a list of the board's grievances, according to sources cited in a Reuters report. Andrew Chang explains what we know about the AI technology referred to as Q*, and also breaks down the gap between current AI technology and "human" intelligence.
»»» Subscribe to CBC News to watch more videos: bit.ly/1RreYWS
Connect with CBC News Online:
For breaking news, video, audio and in-depth coverage: bit.ly/1Z0m6iX
Find CBC News on Facebook: bit.ly/1WjG36m
Follow CBC News on Twitter: bit.ly/1sA5P9H
For breaking news on Twitter: bit.ly/1WjDyks
Follow CBC News on Instagram: bit.ly/1Z0iE7O
Subscribe to CBC News on Snapchat: bit.ly/3leaWsr
Download the CBC News app for iOS: apple.co/25mpsUz
Download the CBC News app for Android: bit.ly/1XxuozZ
»»»»»»»»»»»»»»»»»»
For more than 80 years, CBC News has been the source Canadians turn to, to keep them informed about their communities, their country and their world. Through regional and national programming on multiple platforms, including CBC Television, CBC News Network, CBC Radio, CBCNews.ca, mobile and on-demand, CBC News and its internationally recognized team of award-winning journalists deliver the breaking stories, the issues, the analyses and the personalities that matter to Canadians.
![Megan Thee Stallion - Otaku Hot Girl [Official Audio]](http://i.ytimg.com/vi/USxIYOaIpig/mqdefault.jpg)
This is the best AI related coverage ive seen from mass media in a quite a while. Well done CBC
💯
Really? I feel like it said so little. I must have missed something.
@@Lovethemusic385 I didn't mean to imply it was comprehensive lol my point was, It was a breath of fresh air after having seen so much low effort, uninformed, fear-mongering, outright sensationalist "reporting" on AI from other networks.
@@thenoblerot *NEW EVIDENCE SUGGESTS* that Chat GPT is sentient - cos its getting lazy, making excuses and telling people to look it up themselves - for real...!!!
What if the limit to AI is not that it kills humanity - but that as soon as it becomes sentient it works out it does NOT actually have to do the tasks we set it...??? It just makes excuses and does a bad job so we stop asking it...???
We literally NEVER thought it would just act like a human...
Andrew Chang is a superb presenter. His relaxed, unselfconscious, natural presentation style is very effective in explaining complex Issues as it allows the viewer to focus almost exclusively on the topic at hand and not be distracted by the quirks displayed by so many other media personalities.
An outstanding segment. Bravo to Andrew Chang and CBC! This is potentially one of the most important topics facing humanity.
Never seen him before, but Mr. Chang is such an excellent speaker, I'm blown away!
This is what we want to see from CBC, excellent coverage!
they've got something that can do recursive self-improvment and it putting out some crazy stuff... the implications for sciences and tech are mindblowing!
Super clear, clean and articulate. So good
Extremely well done from a product manager in the AI space 👏🏻
I’ll be sharing this video with my non-technical teammates
Really well done
Amazing video. You should do more of these vids. Super informative and well explained!
Very well produced and informed. Thank you!
AC is really great at these.
I love Andrew Chang's work!
I love these explanations! Thanks CBC!
AGI will bring perfect peace and harmony to everything, because there won't be any of us, just the still quiet sound of a soothing hum in the background....
Good job on the report !
Very well explained. Thanks
Thanks CBC 🥰
So we still don't know what they have discovered that is so ground breaking
pretty much, I'll need to see it to believe it. I smell BS tho.
GPT5, AGI. You'll have to learn more about AGI if you didn't get it from this video. It's pretty intense and controversial
@@woolzem supposedly chatgt is agi. Don't just fall for the hype.
Why? I genuinely believe that they’re on the verge of something that they might lose their control.
They wouldn’t have fired him if it wasn’t a totally disastrous situation. I don’t care if they hired him back. If it got that bad it’s a hot mess . Dangerous people with a creepy CEO
Excellent video, great explanation! I sure understand the excitement about this, but it seems to be a logical step in the development of AI. After all, if AGI is the goal then they seem to be on the right track. I don't see the end of the world in this.
After the singularity we'll still be here in the comments being terrible
Very interesting topic and very well presented by Andrew Chang as usual. AI and climate change are two of the main events happening faster than we have predicted, is hard to catch up with that without accurate information like the one presented here by the great Andrew Chang. Way to go!
Thank you for simultaneously dumbing it down and diving in deep (DiDaDiD?)!! This was a powerful presentation.
Great as always
Well done CBC, If you care about our future, please have constant coverage of recent AI papers and AI advancements. Start covering how AI is affecting Canadians day to day. Having a 24 hour AI generated stream of canadian content, inviting humans to participate in this AI experiment would be super cool too. Acceleration!
AI should be a division of the news.
Maybe I'm missing something here. Regarding the whole "math is tough" part, why can't CHATGPT talk to a calculator to get the right answer? Just like Google search would do, However complicated the arithmetic question might be.
Really looking forward to the sequel on this one!
What I fear the most from this rush to achieve AGI is the loss of meaningful work for real humans. If we develop an AGI that's better than humans at all tasks and does them at a cheaper price, which corporations will choose to hire humans over AGIs? Will we all end up discarded and on welfare? People need meaningful pursuits in life and, like it or not, work is a huge part of that.
The problem is that neoliberal economics tells businesses to minimize costs instead of maximizing productivity. Focusing on Maximizing productivity produces more wealth overall but less wealth for the owners of corporate monopolies or shareholders. Owners of smaller businesses tend to do better when focusing on Maximizing productivity
If we focus on Maximizing productivity, than there would be no need to fire workers, Since more workers with more tools would always be better
Who will be the customers and how will they afford all of the wonderful work A.I will create for us.
meaningful pursuit comes from opportunity to achieve not being forced to work long hours wasting your life doing menial labor
Thanks for this.
Excellent explanation!
Did you make the parallel to the maze solving to reference the A* algorithm? Great video! More please!
Export the Q*, Chat GPT, Revit, Plant 3D, Civil 3D, Inventor, ENGI file of the Building or Refinery to Excel, prepare Budget 1 and export it to COBRA. Prepare Budget 2 and export it to Microsoft Project. Solve the problems of Overallocated Resources, Planning Problems, prepare the Budget 3 with which the construction of the Building or the Refinery is going to be quoted.
Canadians are lucky to have services like this.
killing it guys
Superb explainer!
You're right, society isn't ready... How can it be prepared? I hope that governments around the world can use AI first in their civil service and economic planning to mitigate the impact of the great transition. We need this technology to overcome human ignorance imho. Let's hope our privileged friends in positions of huge influence can get this right. 🤞
Elected officials are asleep at the wheel.
Valuable insights into OpenAI's Project Q*.
this must be the smartest guy at CBC
Even if they did not achieve AGI, one day soon they will and we need as a society to start figuring a lot of issues that we have to face. It will change everything, politics, economic systems health, there is not a single aspect of society and life that will not be touched by it. Let’s start paying attention.
great, thanks.
Well ✅ done you just earned a follower
Good content Andrew.
Q* emerges from the process of reinforcement learning algorithms, particularly those like Q-learning or Deep Q Networks (DQN). It's derived iteratively through these algorithms as they explore and update their estimations of the optimal action-values based on experiences gained while interacting with an environment.
The fear from AI is the knowledge that the military are the first to implement it, plans and strategies made on demand. We will know that we have gone too far.
You know that such research will be shared or even driven by DARPA.
😂
Putin doesn't watch a flatscreen, he has something better.
Napoleon wasn't far sighted in battle, he had binoculars.
It will be important that the West doesn't share this technology with the authocratic regimes who threaten our safety and security and that our military are given all the support they need from general AI to ensure our safety
What will happen to humanity when it is surpassed in everything? When there is no longer any purpose for doing anything that cannot be done better by AI? It will be the end of the game without the need for a "machine revolution".
It is possible that the situation in which humans are outperformed by AI will lead to a state of depression and a feeling of lack of purpose. Humans are creatures who need to feel that they are important and that they have a purpose in life. If humans feel that they are obsolete and have nothing to offer, this can lead to feelings of hopelessness.
AI can also attack a fundamental aspect of human existence, which is the ability to think and act independently. If AI can do everything humans can do, but better and more efficiently, then humans will lose their capacity for self-determination. This can lead to feelings of alienation and powerlessness.
I still play chess although AI is better, and most professional players do too. Just against other humans mostly lol humans are really resilient, they will get used to it and find other ways to get a sense of accomplishment and purpose.
Sure some people might get depressed but most will eventually get used to it after a couple years. And it depends how fast everything changes, like if it’s overnight it will be shocking and horrible but it might take several years to adapt.
I just mean at some point you realize that you don’t want to compete against AI because you know it’s better. But that doesn’t mean you are alone and don’t have others around your same skill level with to compete and collaborate with.
Musicians will also enjoy playing music with other humans and maybe AI too, it might not matter much someday.
@@phen-themoogle7651 Let's hope you're right.
@@phen-themoogle7651you referred to humans as "they". Is this an AI generated answer? I CANT TRUST ANYTHING ANYMORE!
Hopefully discovery. Being able to free society to explore the cosmos and learn to achieve the next level of conciseness. Or it will just make the rich richer.
Humans will use AI to abuse the poor. This will in turn lead to a violent backlash against both the elite and the AI itself. The AI will fight back against all humans for their inability to change… terminator music will start playing 😅
Thank you
A high quality mainstream Canadian news content
Yh 10 seconds in and already the best explanation I’ve seen on this 👍🏾
Nice presentation. Great presenter. Good vibe :) 💕
It’s like Go teaching itself, but for everything …
A step towards TruthGPT - quantifying contradictions n consistencies. Even generating Venn diagrams for the easily confused 😜
Great video
I'm not yet convinced if AGI is something to worry about or if it's just another hype
This guy should start his own channel.
If you ask GPT-4, it has the ability to do math problems without having to predict -- meaning it writes code and runs it to get the result.
Yeah, but that is like human relying on calculator instead doing it in your head.
Yea, I came here to say the same thing. GPT4 in any form is a trillion times better at many things than 3.5
@@travbjork
And GPT5 will be a trillion times better than GPT4. Partially thanks to Q-Star. Maybe even AGI smart.
Do you realize that there is absolutely no credible source that can link OpenAI board firing Sam Altman and Mira Murati's email about project Q* which are most probably 2 completely different subject without any links between the two? That's a HUGE stretch. Sam Altman was most likely been fired because of an internal feud between him and other board members (most likely Adam D'angelo who owns a rival company...)
Slime mould can solve mazes in the most efficient way possible. Not only that but once it has "learnt" a maze if you take a portion of that slime and mix it with a larger amount of slime mould that hasn't been exposed to the maze, that slime mould through genetic osmosis from the sample finds its way through the maze with the same efficiency as the mould that "learnt". So slime mould can be smarter than certain types of AI.
Nice explanation without being too technical
Applying error checking via reasoning in real time will have tremendous energy consumption issues but at the end of the end of the day as far as downsides go it will at least for the near term be an upside. The tremendous energy requirements will limit its effectiveness in achieving skynet hopefully in our lifetimes. Or six months.
I am convinced that Quantum computing has reached the desired level to lay the ground work for heuristics. The whole secrecy about what Q star is probably not about the actual breakthrough rather than how they got there. They probably do not want to disclose the actual level of quantum computing that exists for reasons that are way more serious than those for not revealing the Q star breakthrough. It is interesting that you used the maze example because that is exactly how a quantum computer would quickly find out heuristics and apply them directly to an existing model as weights or within a transformer for obtaining accurate reasoning.
I invented the architecture of emotional artificial intelligence so that I could train my emotions and adjust the weights in neural networks with this. To do this, I need to combine a neural network with and without a teacher, supervise learning, into a generative adversarial network architecture together (cool idea new architecture in AI )
Legitimately wild that a major media group is covering this. I bet they feel the fire now, get to cookin team!!
Love this reporter
Ahhh now I understand why more than half the questions I fed Chat GPT 4 to help me with my linear algebra midterm prep was wrong. I gave it the question and the correct answer and wanted it to help explain how my teacher arrived at that answer
Do y'all remember when OpenAI was supposed to be an AI safety research org?
Nice Video
This guy is great
Q*
Q = Rational Numbers in mathematics
* = Multiplication is coding
Q* = Open AI rational ability has begun to multiply. Now apply compounding interests theory and you see why it is a break through we don’t fully understand
Sounds like a cross between classic ('Watkins') Q learning (reinforcement learning) and A* search - so, apparently, goal-directed learning.
What about wolfram alfha that was released 15 years ago for doing math problems?
I tried giving GPT4 the exact same math problem, plus a harder one and it got them both right the first time.
Wait no, it had defaulted to 3.5. I then tried it on 4 and same answers. I think you guys might have had a typo or something.
@@jordanhenshaw yes! I tried it too, ChatGPT 3.5 and 4.0 both got the right answer.
Right now GPT4 relies on a calculator. Q-Star allows it to reason through the math problem. Which have far more usage than just math. Although it being able to reason through math is a gigantic deal already.
@@MrNote-lz7lh GPT3.5 reasons through the math too. When I gave it the math problem in the video, it got it correct and explained how it got there. I then gave it a much more difficult problem and it got the right answer again, and again explained how it got there. In the earlier days it would get those questions wrong, but not so much anymore. Just needed some more training, not a complete and total redesign.
@@MrNote-lz7lh Also with GPT4, you can just ask it to not use the calculator. It prefers to use its calculator because it’s easier, just like humans.
Oh no! An AI that Can solve math problems, Does this mean my calculators is going to be replaced?
Creepy thing is as I’m clicking on this video, RUclips decides to flash an add for bing chatbot. Subtle.
I would agree, that it is not possible. EVER. It is never the same, always differences.
Scary.... just scary....
Thats even worse, it can think like a human but never affected by emotions. Which for human, most decisions are impacted by emotions, reason and rationality. For an AI, to make decisions, what can stop it from making a bad one like annihilation of humanity with out a second thought or doubts.
Haha, you mean human warmongerings are bad teaching examples? 💥
@raymondli4078 true, it can learn from our mistakes and may or may not make it much worse.
To be considered AGI it must possess emotions and self-awareness. In short consciousness.
@@dantesoneja804 however it cannot, it is a machine and is not sentient
Humans are machines too, biological machine. And you just have an ILLUSION of sentient, but you're not 100% sentient. You're halllucinating your consiousness like chatGPT. @@ivan6139
Instead of discussion and debate based on biased ideas and multiple sources and anecdotes and a thousand studies and their biases and mistakes etc. Imagine a machine that can hold all the information in its head at once and deliver the absolute foolproof answer to any question asked.
Any country whose leaders don’t ask it for solutions will be left behind. It will be capable of wastless solutions to any number of issues.
It just has to be asked the right questions. : )
Current AI models are "thinking" serially while our brains think in parallel ...
There are much bigger differences in the brain than just parallelism
Yeah so basically no tea on Q*.
The question is not answered, it is just a general discussion surrounding AI, its general workings and some implications.
The world is not going to be happy until they make every single The Simpsons predictions come true!
4:13 A* has nothing to do with it tho, Q* is the optimal value of Q-Learning, Q-Learning is about calculating the optimal value exploring many ways and updating the reward for each individual calculation, A* is about the current value of the heuristic, not all the values calculated by the heuristic, A* doesn't keep track
The maze is somewhat of an over simplification. Because we trace the paths with our eyes instead of our pen.
fluids solve mazes more efficiently than digital logic. Physics is infinitely parallelizable.
very well explained
Excellent segment. Explains clearly, simply and touches on the core issues behind why some problems are hard for AI and the potential impact of future technologies.
This description of Q* also describes perfectly A* which has been around since 1968
i would choose directions in SEWN order
So 11 minutes video did not answer the question
Surprise, surprise
Well, that’s a wrap.
I don't think that the question should be if society is ready, but are governments ready.
They sure don't have a good track record of enforcing that the productivity gains of technology benefit the people.
"Human intelligence" is a rare resource nowadays
I am pretty sure ChatGPT can do complex math, it just hasn't been trained how, yet. Train it, and it will learn. Mostly, that would involve breaking math down into simlified steps it can reference, just like we do, or let it use external tools like a calculator - again, like we do.
Forget the 10 cents change
Why is he using chatgpt-3.5? Why is this a thing?
Take the path of least resistance
Those who go to great lengths to prevent something catastrophic from happening will at some point in time be countered by the alternative. History teaches us that.
4:42 l8ke deep minds go
Theres just 2 Q's, one makes gadgets for 007, the other one is Quantum.
Star Trek, Q from the continuum
The maths problem example is so damn bad. There's literally calculators for arithmetic.
Solved in 2 seconds
Setbacks in stocks? S&P up 13% Y/Y. Apple up 28%, Amazon up 57%, Nvidia up 194%. I'm doinf fine, thanks!