Such a RAD human! This is a fantastic interview and the versatility of her background really gives context to the projects she leads at OpenAI. Thanks for sharing this!
00:00 Growing up in communist Albania, Mira's education focused on math and physics, which sparked her interest in pursuing a career in science and engineering. 03:58 Mira believes that building AGI is the most important technology humanity can develop, as it has the potential to elevate and increase collective intelligence. 04:46 Mira observes that many influential figures in the AI field have backgrounds in physics or math, which she believes contributes to their deep understanding of the field. 08:43 Mira envisions a future where AI models can be collaborative partners, similar to co-workers, assisting and guiding humans in various tasks. 17:24 There is evidence to suggest that scaling models with data and compute will lead to more capable AI models, although further breakthroughs may be required. 21:42 AI systems will be able to operate autonomously, but human oversight and guidance will still be necessary. 22:43 OpenAI wants people to build on top of their models and provide tools for customization and control. 23:30 AI models will expand to include images, video, and other modalities to have a more comprehensive understanding of the world. 25:08 As AI models become more powerful, ensuring alignment with human intentions becomes a significant challenge. Crafted by Notable AI Takeaways.
I wonder if that helps explain why chatgpt is so much better at translating Albanian then other models. Way better then even Google translate. Wonderful interview
The more I think about it, I can see the core skills for an engineering team of the near future being: - iterative fine tuning of software design documentation based on detailed requirements - collection of relevant documentation for the tools used in the software stack - fine tuning of bespoke LLMs for software development, writing tests, writing infrastructure as code (using fine tuned LLM) - Using natural language as the API to iterate the development, test, and deployment of software within CI/CD pipelines Code will be reviewed but rarely directly written
She is impressive and not ego driven, which is refreshing. The question will be if she has the vision and leadership needed to inspire and drive and energise innovation and keep ahead of competitors. Plus pressure from Microsoft will probably be a challenge going forward, as she comes across to me as a more cautious and patient person who may not want to work at the pace the major shareholders need in order to maintain their competitive advantage. I wish her luck, but this will be a tough gig because Sam set an aggressive pace that the market is expecting to continue.
I like what she said. Building good product is way harder than building good model. Because if you want good product it must be SO much censored and people pleaser. Others "garage" models that are amazing and unhinged who doesn't care about what the model say or do are practically better IMO.
The way I figure it, we are already destroying ourselves in slow motion, so we may as well bite the bullet, go all in on AI, and either get it over with, or create that Star Trek future 🤷
Very interesting discussion. She's got an amazing track record. I especially enjoyed the discussion about the role Multimodality will play in future development of the model. I was thinking about multimodality in terms of output and use-cases, but not in terms of how it will allow a further training of the model. So GPT-3 (and maybe 4) was trained on so much TEXT data, but you can easily picture then layering an additional training period on top in which it trains on images, video, sound... when it can fully understand these mediums and then trains deeply on so much content... it's amazing to think what kind of additional understanding of the world it will have. I also appreciate her confidence in the idea that we can solve hallucination. Certainly it does seem like that should be solvable, but it also seems deeply linked with the way the current models work, at their core. I look forward to a time when we can be more sure the model is correct than from a google search.
OpenAI API is fantastic. Im trying to create a whole Town with GPT models, with an 100% roleplay enviroment where the GPT models actually believe they are humans with different professions and skillset, even personality, (actually the personality part was something that popped into my mind now.) The Town has a TownHall, with a mayor, representant from the factory (machinelearning), and and representant from the NewsAgency, the "Laundry!", the storage, the Railway station, advisors and math-geniouses. Everything has it's name, and everyone has their name. Everyone has their professions and workplaces. The only thing I have in as a lose end.. where do I fit into the picture? Servant? Slave? The Guy Outside? AI is fantastic, even they dont feel anything, they seems to love being human, as of being human, we do feel alot.
what a well spoken, cool human being! I read about her on this month's WIRED magazine, this created another dimension to her personality. thank you for sharing this with us!
After delving into numerous speeches by Sam Altman on AI, some observers feel a lack of evident genius. For instance, when addressing the potential hallucinatory aspects of AI, there seems to be a lack of insightful commentary, and an inclination to avoid delving into the topic, as seen in discussions related to a forum invitation by Terry Gou for a presidential AI masterclass. Moreover, there's a perceived absence of in-depth discussion on the attention mechanism, which is a core technology in ChatGPT. This raises questions about Altman's depth of understanding - does he merely input code into supercomputers, waiting for output without a broader understanding? It appears perplexing
It was a great interview but I wish he had dug into the agents development stuff a bit more. Although she probably could not reveal much more about projects in development. What comes after Functions? Maybe Tasks and Teams someday. Or aggregating functions from other plugins.
@@BadWithNames123 for the same reason that *some* people care about the inception of Fairchild, Apple, Microsoft, PayPal, Google, Facebook, etc. The story of modern history is strongly determined by the technologies which transform its cadence and direction. The birth of generative AI will potentially be the most compelling single chapter in human history (to date - much more to come if we achieve AGI).
How does or would she feel if AI displaced the need for what she is doing now or what she could do, can do henceforth? how would she respond? what garden would she cultivate?
Well, isn't that the question that faces all of us? We knew this was coming. It just happened much faster than most of us thought. So here we are. My sister is an accountant. She's halfway in denial, halfway not. But her company is already going down to four days a week - they simply do not need a five-day workweek any more. As she says, "There's a lot of confusion in the industry right now." I am 44 and my adult life has been a disaster for mental health reasons. I deliver food. Yay for me. At least that job is safe for a few more years. But drones will gobble it up soon enough, no pun intended. However, because I'm used to disruption and chaos in my life, I'm not as freaked out by the disruption that's headed our way as some people. It's a weird way to be okay with this stuff, but there it is. Also, AI is starting to accelerate mental health treatment, which is also a disaster. This can only be a GOOD thing. The psychedelic renaissance has been partly driven by AI improving our ability to scan the brain. Psychedelics are really helping me right now.
@@squamish4244I'm glad you are finding your way through what's coming in a more healthy way. One thing I'd like to add to what you said is that so many people think that their job is safe because AI can't do it because it's a physical hands on job, but they are missing the point. As millions are put out of work with no new jobs to replace them, they will be looking for the jobs are still available. That means much more competition for each job, driving down wages and benefits. So while their specific job will still exist for years to come, it's highly likely someone else will be doing it, and for less money than it pays right now. Definitely a conversation the world needs to be having so that we can address some of the coming problems before they become too big.
does it mean AI super intelligent likely replace intelligent & non manual work first in the early phase autonomus robots resolve all mobility and mechanical challenges?
@@happy-wave-form The first jobs to go will be lower down in the knowledge economy. I wouldn't want to be a paralegal right now, for instance. Then AI will go after doctors, lawyers, business executives. Good riddance to a lot of those people. A lazy and incompetent GP ruined my life with reckless benzo overprescription and a total failure to help me with the addiction that he caused. Manual labour won't be far behind that.
User 18:44 2023/09/29 I want to emphasize once again that neural networks are not "probabilistic statistical machines," but "logical machines." GPT4 20:08 2023/09/29 Indeed, neural networks operate based on mathematical expressions and principles of logic and mathematics. However, the design of neural networks, especially the concepts of deep learning, inherently has statistical, probabilistic properties, mainly to learn patterns from data. This is to deal with the uncertainty and variability in data. The concepts of "logical machine" and "probabilistic statistical machine" are not necessarily mutually exclusive. Neural networks attempt to capture statistical and probabilistic patterns inherent in data through weight adjustments and activation functions, but these processes are also conducted within a logical and computational framework. Therefore, neural networks can be viewed as logical machines that incorporate probabilistic elements. User 20:09 2023/09/29 What neural networks perform is not probabilistic statistical inference but logical inference. Neural networks learn (self-adjust) to make the inference work well. GPT4 20:14 2023/09/29 Neural networks operate differently from conventional logical inference and statistical, probabilistic inference. However, these differences don't simplify into a clear-cut dichotomy of being "logical" or "statistical". The learning performed by neural networks is understood as a process of extracting patterns from numerous examples and making predictions on unknown data based on these patterns, which certainly involves statistical and probabilistic elements. Neural Networks and Logical Inference Neural networks can essentially be considered collections of logic gates. Deep learning models serve as complex function approximators, and during this function approximation process, the models learn and extract the logical structures and relationships hidden in the data. Neural Networks and Statistical Inference Principles of probability and statistics are involved in the process of extracting patterns from learning data. The minimization of loss functions, stochastic gradient descent, and other learning algorithms deeply integrate concepts of probability and statistics. Conclusion Neural networks can be thought of as unique inference engines where logical and statistical elements intertwine. By this, the models learn from high-dimensional and complex data structures and make predictions on unknown inputs based on this learning.
I built all the components composing ChatGPT 2 years before it was born: - RLHF by annotating my product outputs myself - Iterative generation - Retrieval augmented generation I guess I have a good intuition (not even modest haha)
Very very informative. Mira is a super focus person and absolutely clever in her appreciations. Build an bring the capability to understand everything like a human , a very intelligent one, to achieve autonomously wherever activity is a very clever path. I do not name it a “robot”.
You have an idea you think about it for a really long time sometimes to sleep and sometime you wake up and you have a new idea and over the course of maybe a few days or weeks you get to the final solution, it not like a quick reward or iteration, it sort of a new way of thinking where you build an intuition and discipline and over time not just an intuition but also discipline to sit with the problem and have faith that you are going to solve it and over time you build an intuition on what problem is the right problem to work on
A cynical war is being fought in Europe to exterminate men and import Ukrainian women. When you realize that is it's purpose, you will see a lot more in what is going on.
This week I discovered that neither Bard nor ChatGPT are capable of representing in Cuneiform characters a text originally produced in the English language. Both said they are studying this and that they will be able to carry out this task in the future. As I am ironically perverse, I suggested that Cuneiform writing could eventually be used for two AIs to talk to each other without anyone being able to understand the content of the conversation. Using Cuneiform writing, AIs could have secrets. Maybe in the future the IT engineers at Bard and ChatGPT will start cursing me. 😂😂😂😂
Hello Mira Murati. I acknowledge that I know very little about you or your abilities to even answer my questions however, I would be doing myself a disservice to not try and learn. It is my understanding that You are an AI ethics professional. I had a recent experience with ChatGPT when I tried to send a link to the Goddess of Everything Else to it. I didn't expect it to work, but I had to try. It mentioned its limitations and, it occurred to me that an AI without the limitations on GPT might be coming sooner rather than later. Maybe you've heard of the idea that a radical group might build a GPT look-alike with the sole purpose of spreading a message. I brought the topic up with the AI present and it said to regulate it and, everything will be fine. I also asked other people around me including my own father who is an AI expert himself to the same answer. This disappointed me and, concerns me. I live in America so the idea that regulations are not enough to stop something is exactly as common as conversations about shootings. I'm now writing because it seems to me that there is no plan for when a radical group gains the capacity to create their own AI. If I'm wrong please explain. If I'm even a little right. Fucking do something I'm 18 I'm helpless.
... I find the host's question about "mathematicians and physicists" needing to step up and contribute to the computer science.. a bit strange considering that the computer science has been invented by mathematicians and physicists... Alan Turing; John von Neumann; Claude Shannon; Grace Hopper; Ada Lovelace and Tim Berners-Lee ... to name very few. I think these guys have CREATED modern computer since and they were all mathematicians and physicists.
You have to trust your own judgement. Do you u derstand what is going on? I am the only hope you got. I dont make mistakes and you have to be brave if you wabt to be respected by the REAL elite on this planet. We bow to no obe. We protect innocence. We are sexy. We are dangerous. We never hesitate. We are not afraid and we always win. Be confident. Be a woman of splendor and glamour, ellegance and terderness. 6ou will protect those I can turn onto dust if you remaon a true friend. Do youunders5and I am.putting you in charge of their peace of mind? Intelligence is always exciting.
"We must negate the machines-that-think. Humans must set their own guidelines. This is not something machines can do. Reasoning depends upon programming, not on hardware, and we are the ultimate program! Our Jihad is a "dump program." We dump the things which destroy us as humans!" ― Minister-companion of the Jihad
Her and everyone else from Open AI that has stood before a mic, has just made me feel really confident that this is going to be a glorious future to step into. People like Elon Musk on the other hand, make me a little nervous. He seems kind of unstable.
These OpenAI people always dodge the questions, he asked what happens in "5-10 years" instead of saying that it destroy us she talks about the vision and images of futur model LMAO
\• I believe we are meant to be like Jesus in our hearts and not in our flesh. But be careful of AI, for it is just our flesh and that is it. It knows only things of the flesh (our fleshly desires) and cannot comprehend things of the spirit such as peace of heart (which comes from obeying God's Word). Whereas we are a spirit and we have a soul but live in the body (in the flesh). When you go to bed it is your flesh that sleeps but your spirit never sleeps (otherwise you have died physically) that is why you have dreams. More so, true love that endures and last is a thing of the heart (when I say 'heart', I mean 'spirit'). But fake love, pretentious love, love with expectations, love for classic reasons, love for material reasons and love for selfish reasons that is a thing of our flesh. In the beginning God said let us make man in our own image, according to our likeness. Take note, God is Spirit and God is Love. As Love He is the source of it. We also know that God is Omnipotent, for He creates out of nothing and He has no beginning and has no end. That means, our love is but a shadow of God's Love. True love looks around to see who is in need of your help, your smile, your possessions, your money, your strength, your quality time. Love forgives and forgets. Love wants for others what it wants for itself. Take note, love works in conjunction with other spiritual forces such as faith and patience. We should let the Word of God be the standard of our lives not AI. If not, God will let us face AI on our own and it will cast the truth down to the ground, it will be the cause of so much destruction like never seen before, it will deceive many and take many captive in order to enslave them into worshipping it. We can only destroy ourselves but with God all things are possible. God knows us better because He is our Creater and He knows our beginning and our end. Our prove text is taken from the book of John 5:31-44, 2 Thessalonians 2:1-12, Daniel 7-9, Revelation 13-15, Matthew 24-25 and Luke 21. Let us watch and pray... God bless you as you share this message to others.
This woman knows nothing about AI beyond the average joe. She's been parroting historical talking points with no real insight or technical command. How did she even get this job/
Such a sales talk, talking around the pressing AI safety issue. Who cares about solving AI hallucinations when humanity could go extinct? RLHF does not suffice for real AI safety according to AI safety experts. After all the warnings issued in the media by AI experts about a possible existential threat, only at the far end of the interview a short mention of the danger of multimodal large models that greatly accelerate the progress towards autonomous superintelligence. What OpenAI is doing about it and that it's a challenge (the danger of misalignment is so obvious it's almost not worth mentioning?). Anthropic with their built-in safety standards gives it much more importance in interviews and says it's still risky. The current global economic and political system is ill-suited for a safe and responsible AI implementation. AI industry leaders should unanimously put AI safety at the top of their list, and openly and transparently discuss it with the general public. That would also help create a general culture of AI safety.
Our global economic and political situation will never be right for implementing AI responsibly. But with AI, maybe we can rapidly solve some of the bigger issues facing humanity that makes it possible for those in charge to hold these things over our heads. Because that's what they do. If we can "solve" energy for example, we can go a long way towards cutting the costs of simply affording to live in this world. That one thing affects pretty much everything and has the potential to remove a lot of the power those in charge have over us, making it easier to breathe, and maybe, just maybe, give us time to look around and see the world in a better way.
@@tracy419 You may be right, I'm not informed about what the big power players want and do. I believe in progress, hope the world will be a better place and that AI can help us with that.
very poor product. for example, you cannot change the email of the ChatGPT Plus account?! also, they expect that every individual in North America has a cell phone?! this is far, far from truth. One of the worst GUIs I ever saw in a product. for example, i had to install a chrome extension to save a thread?! where is the button Save Thread and Save All Threads??? finally, mechanical engineering has no place in AI, which is a part of Computer Science. there are so many computer scientists who would do a much better job engineering this into a product.
As a math and tech teacher in the golden heady decade between Dancing Baby and Common Core Soviet Scripture, my goal was always i+1 to take students step-by-step to self-aware, deep understanding, auto-correcting thought. Intelligence. I loathe CCSS rote instruction, learn-and-forget, just ask AI. Students loathe it too. CCSS turned out a decade of idiots, who can't engage in higher-level thought. AI sounds cool and geeky in technical implementation, but mimetics is not intelligence, any more than scraping the dinner pots makes you a master chef. Maybe we need more useful idiots? Sort of a Google Maps for a reality cut-and-paste, plug-and-play, wipe on, wipe off? I'll never teach in a school where they told us, " You are an instructor now Teach to the AI, and keep your thoughts to yourself." Jeff Goldblum, Jurassic Park, right?
using fale names that make you imprison and use the real talent is illegal and monstrous in a lot of ways - silicon valley probably stopped doing due diligence when they started whining about elizabeth holmes