New AI Breakthroughs Explained. It's ALL Accelerating!
Вставка
- Опубліковано 12 лип 2024
- Use code ANASTASI10 to get 10% off your Hostinger plan today: www.hostinger.com/anastasi10
Timestamps:
00:00 - The Future of AI
08:00 - Gemini Explained
13:45 - About Q*
15:30 - How technological progress is accelerating
The Gemini Paper: paperswithcode.com/paper/gemi...
Gemini Benchmarks: deepmind.google/technologies/...
Deepmind GNoMe: www.nature.com/articles/s4158...
Ray Kurzweil - The Law of Accelerating Returns: www.thekurzweillibrary.com/th...
Featured in this video:
Ivan Poupyrev, Ph.D. Founder and CEO Archetype AI
www.archetypeai.io/
X: / physicalai
👉 Support me at Patreon ➜ / anastasiintech
📩 Sign up for my Deep In Tech Newsletter for free! ➜ anastasiintech.substack.com
Use code ANASTASI10 to get 10% off your Hostinger plan today: www.hostinger.com/anastasi10
@-AnonLab2 me too !
Thanks a lot
AI is still AI and not AGI so what is the point in creating Gemini all of this is hogwash
I wish your parents and family are doing well.
i wish yours was as well@@user-ww2lc1yo9c
Please note that at the moment of recording this video, Gemini had just been released. Now, there appears to be some controversy over the edited Gemini AI demo video. It seems the person's smooth conversation with Gemini wasn't constructed in real-time. That's really frustrating.. Now I'm going to wait until I can try Gemini Ultra to draw my conclusions.
Thanks young lady
Don't be overexcited about living right now. I am not.
I was, as an 8 year-old, in 1971, in spite of the oil crisis and the notion back then of being left 20 years of oil. I was smart but naive enough also, as a kid, to know about not only problems we faced, but also all the progress made so far. Which made me conclude: what an extraordinary luck to have been born at this time, precisely when all our problems can be solved within twenty years. I was born, I felt, right when we could and would solve all major issues, to achieve a peaceful advanced, rational world society. Based on what I knew and the power of logic, I just saw it was feasible. Wasn't it?
In theory, it was. In practice, society is hugely complex and may be, in a way, compared to an ocean (less complex than society) where we are like water molecules. Unlike water molecules, we have agency. Some can develop new tech, for instance, that will mediate some accelerated change. Yet, we basically remain water molecules in an ocean with huge inertia. For all the revolutionary change that might take place in some corner, whether at the top of a wave breaking in full view of some wonderful island, or at the bottom of the deepest trench, all other water molecules remain water molecules subject to previous dynamics. AGI cannot decouple most of us from our previous dynamics, which are imprinted on our brains and which we tend to imprint on our kids, on next generation's brains.
Society is a complex system and remains a system whether we increase (or decrease, sometimes) its complexity. No technological innovation is going to work only for good. Mankind is yet to reach a stage of cultural evolution, worldwide, we can deem as truly, safely, civilized. We are yet to learn to behave ourselves. We are out of control (especially where control is supposed to be paramount, in totalitarian regimes).
I am a scientist. I can see, just like you, maybe better, all the potential of AI and R&D in general. For good and bad.
It is an exciting time to be alive if you are young and at the top of a picturesque wave, instead of young and struggling, lost at sea (as most people are). Just realize you seem to be riding the Big One. I am watching you from the beach, on what's left of an island without proper hills for safety. Good luck everybody. Just try to make good kids so they may achieve true civilization in case they survive.
We don't stand a chance. Not before we manage to offer most kids a decent education and youth.
It is quite telling and a crazy bet to be talking so much about AI, quantum computing, genetic engineering and all the rest, as our last shot at preventing utter disaster. So, we've been shooting like mad this amazing industrial revolution R&D weapon that's taken us all the way to the brink of final, irretrievable collapse on a geological scale, but still look forward to take one last shot with this very weapon, hoping to hit a different target. A magic one that will make our dreams turn real.
@@nomcognom2414I do agree with you although there are three elements that will save civilization on Earth. I am the same age as you and I grew up through the evolution of computer programming. Since then I've sat back and watched it evolve as AI brings us into a new realm. I am hoping AI will be able to discover a way to unlock the full potential of the human brain. If we were all able to function at least 80% capacity, I truly think that everyone's attitude toward each other would change for the better, only at that time Society could become one. The second step that has to happen is we have to discover a almost free energy source such as perfecting cold Fusion as an example. I'm really hoping that AI can accelerate this particular field. And the last requirements that only AI could ever have a chance at, is to control gravity. With this ability, leave our planet and populate other planets in the universe will almost certainly change our view of humanity. AI will be our true savior.
Just my opinion. 😊
You need a better mic, clipping on sssh sounds.
You're too kind.
Holding google to their word, and attacking them for deception is an important action. If there is no recourse or backlash, everyone will become as deceptive as Google.
if you don't know the video demo is not honest, they provided screenshots to Gemini and then asked about them. it's not live interaction as they showed
So still not near to Chat-GPT4?
@@scroopynooperz9051
I'm just saying that is not as impressive as the video shows.
you can watch "AIExplained" he better explains the technicalities behind it.
@@scroopynooperz9051 It's pretty close, but still behind GPT-4 for a lot of tasks
@@scroopynooperz9051 yes, its faked. Do the tests shown in the demo to both models and gpt4 is very clearly better.
@@scroopynooperz9051 it'll catch up very soon, Google has a huge amount of compute
10:10 I think we shouldn't give too much value to claimed performance of a system that is not available for real. If Google thinks Gemini Ultra will be available sometime in next year, it should be compared to the then current GPT-4 variant.
Agreed. There's already evidence floating around and being discussed how they 'fluffed' their performance metrics. Highly speculative. Big business doing what big business does... exaggerating their service/product.
Definitely. It looks quite embarrassing tbh because it doesn't even beat gpt4 by large margins, so by the time gpt5 (or whatever is next) comes Gemini (even their 'ultra') will be obsolete.
@@EricDBrown and yet channels like this one are still promoting and hyping it up like it’s the second coming of Christ. Kinda sad really but it does allow those of us who are paying attention to separate the wheat from the chaff.
@@sinnwalker It's also the case that their safety/alignment BS makes these models dumber over time. GPT-4 researchers have talked about how it steadily got less intelligent the more they baked into the model. That means that Gemini Ultra, currently uncensored, is being compared to GPT-4 post-censor. So it's even more uneven. Ultra' results will probably be lower by the time it releases while GPT-4 will be better probably, if not already GPT-5.
@@sinnwalkerthank you I don’t know what hype is about
I am at technology nerd. And I always enjoy your positive outlook. about AI. and I feel the same way. I think it's a great thing for humanity.
Agreed.
Check back in 2 to 5 years as to how it's all panned out.
@@flickwtchr Judgment day in 2 go 5 years?😅
rose tinted glasses. people who develop this have given no thought to how this can be abused, then again I doubt they care because they want the future they want with no regard to what the majority of the population wants.
technophilia is a mental illness
You're naive. Authoritarians around the world are salivating at this
Google Geminlie, as a marketing piece it was interesting, but based on the last presentation of an AI from Google, I prefer to wait for this to be available to draw my conclusions. Excellent video Anastasia!!!
Was this made immediately after the event because as things stand now, its not so true that Gemini beats GpT4
I love it that Ray Kurzweil is finally getting the recognition he deserves.
You mean Ray Nostradamus? ;)
@@Paul_MarekThat one, yes
Tough to live long enough to see life extension materialize, but die just before it is ready. .
@@Thedeepseanomad We can just spin up an ancestor simulation of him.
@@Thedeepseanomadhe'll probably outlive us all 😂
Thank you for the progress and the update. Keep digging...
Gemini's multi-modal is so much more than just patching together different systems. LLMs are based on Word-2-Vec or storing words/syllables as a long vector of floats where they are overlayed and share micro-features such that, for example, you can activate the word "king", subtract the activations for "male", add the activations for "female" and the remaining vector matches the one for "queen." Well - Gemini ALSO represents images and sounds overlayed so that words, images, and sounds all share microfeatures. This is a much more comprehensive semantic knowledge embedding that captures many more aspects of human intelligence. Now add that to reinforcement learning and... all we need now is to enable a process of thought in the system - actual thinking - which resembles problem solving but doesn't turn off when the immediate goal is reached and instead constantly interacts and self-reflects. Now you're talking consciousness. This is so close I can taste it. Gemini could taste it too if you gave it more sense inputs. :)
I applaud your thoroughness. You’re the first to offer a practical solution to engineering consciousness.
Its not going to be conscious, the IIM is just an hypothesis.
I don't know why they changed the product name to Gemini. I'd much prefer the original name "MUM: Multimodal Unified Model" (2021) by Google AI Research. It's not only more fun and relatable to younger generations like Gen-Z and Gen-AI, but it also has a more cultural connection that might resonate with them. The term "MUM" is commonly used in slang in the UK, and it carries a particular meaning that is meaningful to this demographic. Additionally, the name "MUM" is less corporate and more approachable than "Gemini," making it seem more user-friendly and accessible. I think the original name would have been a better choice for marketing purposes.
i love this thought process.. a good starting point for machine awareness..
Some said they faked the demo.
So informative ! Thanks a lot. I love you featuring guests which are deep into the topic.
Thanks you once again for the remixing video. Top notch!
Anastasi In Tech
all of your presentations on UA-cam
make me happy.
I am not sure if its your enthusiasm or the
breaking news you present. I do know
stumbling into your channel was the very
best thing to have happen to me.
Thank you Anastasi.
Thank you for these insights. Great video!
Models are developed, released, used en masse. New models arise from the old ones which propagate and from them new discoveries are made which lead to better models. The process by which this happens is accelerated for every generation of models and they even compliment the development of new technologies that make them function even faster and more efficient. It's a positive feedback loop that's really only limited by physics, and who is to say that certain boundaries of what we understand in physics may not be surpassed by continued progress. This is all quite fascinating and exciting.
Great new studio setup. Really takes the visual aspect to another level. SoCalFreddy BTW Great topic coverage. Will be following the channel much more closely.
So far google has released,
1) a model on par with gpt3.5,
2)a mini model for the phones (thats kind of cool)
3) alot of claims about what ultra supposedly does. (Using sus testing methods) it even shows in the paper that when using the same 5 shot method for testing, it scored less then gpt4 did on 5 shot. The score they compared it with was the older model of gpt4, and to achieve a higher mark they used chain of thought prompting, while at the same time not using the same prompting technique with gpt4.
This is what we call junk science, and it makes google look really desperate for a win.
This. Also their marketing video was VERY misleading compared to their paper.
Google faked it like a moon landing-- The results were real just NOT in real time that is 100 percent edited and they dont tell us thats fraud
Outstanding video, Anastasi! It's really got me thinking. Thank you.
Wow, I’m so impressed. This is really exciting watching how our world is rolling to the new age of technology. Thank you so much for your video I will definitely subscribe and share it with my friends 👍
Thank you for this report. I have dabbled in AI for many years. I was discouraged by the slow progress. Now I am excited by the rapid progress, but discouraged by the privacy invasion.
Yeah, I don't care how smart it gets, if it's from Google it's a hard pass from me. I despise Microsoft too but I've kind of accepted losing all privacy to them, doesn't mean I want to give it away to google as well. When this stuff is truly open source and minisized to work on home PCs I'll be more excited. I've already got Studio LM running with an uncensored 7B model that's quite fun to talk to. I'd rather talk via text with that than with a Google bot staring and recording everything about me.
So, at this point, Anastasi already might know about the Technological Singularity Theory, huh? At that total rate of developments (20.000 times the current moment) that only means that at the end of this century, we could be building our first Dyson Sphere around the sun to become a type 1 Civilization on the Kardashiev Scale, i know it sounds crazy and Sci-Fi but is not, that is the Natural and inexorable result corresponding to an exponential acceleration of that magnitude.
Such an acceleration already happened in the 20th century itself, humans started building fabric and wood airplanes with parts made for bicycles that hardly flew 35 meters and we ended the century with the SR71, the Concorde and the Valkyri and just think about the astounding speed at which Elon Musk built a web of satellites around the earth in a matter of 5 years, forming a spherical network around the earth that is already visible to the naked eye. If anyone knows how to apply Ray Kurzweil's exponential growth to construction, it is Elon Musk. Once he has secured the Starship vector rocket and the Teslabot, he knows that control of the Moon and Mars resources are just around the corner.
If you build a Dyson sphere around the sun, there won't be any sunlight arriving at the earth. Imagine the earth without the sunlight, the nature... 😮😮
Our intelligence cannot keep up with the exponential increase in knowledge. The average human's IQ is 100.
@@DihelsonMendoncaA Dyson sphere would generally be constructed with the radius of 1AU so that the inner surface has Earth-like conditions.
@@DihelsonMendonca −273.15 °C as a default earth temperature... rip existence.
The 1st thing that in reality has to be solved is our own mortality. Without this all the big things will mean nothing. any type of significant travel in the universe is also rendered impossible or ridiculous. my prediction is that we solve it by finding a way to totally eradicate inflammation. This alone would give us another 100 years of life. only stopped by the deteriorating bones / hips / joints etc. but material science will solve this problem long before that.
last days have been enormous in open source, much more ground breaking than anything we've seen from companies
Google wasnt honest about the demonstration of capabilities. Gemini certainly doesnt beat gpt4. As much as you all would like for it to.
Thanks for sharing. Looking stunning as usual.
What I understood about Q* was that it used another thing that checked OpenAI's LLM, and that it boosted the reliability of it. Then they realized that the checker on its own was also quite reliable. I'm at the extent of my understanding of it, but that 'checker' seemed like a way to AGI. I don't know more than that without digging in.
You rock!
Love your enthusiasm - which, in this case, is well-placed!
Great video. Informative! 👍✨
I feel the same, we are sooo lucky to live in this time. It is the most exciting time for technology ever. It is very difficult to imagine AGI, right now but even in the near future. Thanks for your super interesting videos, it must take so much of your time. ps : this clothes style fit you better
"Happy to live at this moment." Things can get better or worse, but they won't stay the same. Loved your video!
Gemini 😃🙌🏻... Thanks for the updates Anastasi!! 💚🤗❤
I’m currently in school for computer science with a concentration in AI and machine learning and I’m beyond excited. There is so much potential with this technology, especially for things like disease research and material science. Imagine having an AI model help discover cancer treatments and identification or discovering how to manufacture graphene at a large scale. I’m excited to graduate and be part of its development!
For tasks that are impossible to accomplish by human skill, like building crystals and molecules, I highly recommend AI. But, when it REPLACES human beings at a rate so much higher than new, GOOD, jobs can be discovered, then I highly recommend the novel by Kurt Vonnegut, Player Piano. 😕
It is only a problem when the person replaced (and people in general) does not gain resources from being replaced. If a machine do the dangerous and boring jobs supplying us with resources, we can focus on the work that we are interested in doing.
@@Thedeepseanomadthere it is. My main feild is economics. I said 2 years ago. Tax all return with Ai at 95% no matter what. Or you will break society within 5-7 years. The taxes to set up UBI all across the world. I fear instead all we get is multi-Billionaires approaching Trillionaire’s by 2030 & the poverty across the world becomes so much worse.
eh I think this is just overly simplistic. That is maybe a far away conclusion but we're heading into mad max mentality. The process of getting to a new economic system will be rough. It's survival mode pretty soon. I think what you describe exists after, but for example imagine saying what youre saying to the people who losts their manufacturing jobs. I mean unless dying on drugs is what they were interesting in doing...
I think it's more likely people will be hustling finding any possible way to make money and having no time to do whatever they are interested in
@@Thedeepseanomad
Instead of a tax policy or other program indexed to inflation, this one might be indexed to jobs replaced.
@@T1tusCr0w Taxation is the first and obvious option on how to fund an UBI. but it is ultimately based on production.
And when machines becomes advanced enough we do not really need redistribution if everyone is entitled to verified machine productivity dividends within a sustainable framework. Like a open sourced machine production commons sector in society.
First time viewing your channel Anastasia. Great stuff!
Looking ahead is a dizzying experience with the rate of innovation and change. Any thoughts as to when this channel reaches one million subscribers? I guess by next Christmas it will be at a million and still growing. Buon Natale...and I hope Santa is good to you this year. Cheers.
Good video! Greetings from Chile
So AI can create new materials (produce-able or not) but can be patented by companies with deep pockets, ensuring that big companies will dominate/control future technology in everything: computers, health, construction, ....
Please check Google's fine print. Google fudged the facts. ~3K
yes there is no video input, they feeded it with still images and text
Sorry for not noticing before, but the UA-cam plaque looks good on the wall behind you.
Even though I expected exponential growth in ai, I'm still pleasantly surprised, but still trying to temper the excitement because when ai really starts, i believe this explosion of capabilities will be seen as mediocre at best. Still, it's an exciting time to be alive.
Sehr informatives Video! Vielen Dank!
I really enjoy how we're having to redefine creativity in order to justify our desire to prove that AI isn't creative. Let's keep it up and see how long it takes before we remove our own minds from the definition of the creativity definitions we create.
It was Ray Nostradamus? that asked, when working on DNA sequencing "How is it going"? his team replied "We are only about 1 percent in" to which he replied "Nearly finished then "and he was right!
Wonderful Anastasi.
great work as always
11:54 This illustration of Ai capability is very remarkable I'm greatly impressed. 6:40 Hostinger seems interesting 🤔🧐💭
As a material scientist I am excited about the many minerals that deepmind postulates but you have to be realistic. The lack of theoretical minerals is not an issue. The hard part is varifying these theories. I would really enjoy a follow up on how many of those minerals can actually be created.
Its like saying that AI came up with 90 million songs which is equal to the work of thousands of artists for a decade.
The key takeaway is where she says that the most expensive aspect is the human labour. Its just another mechanism to transfer the resources upwards. A hellishly effective one.
On the hardware side if you made the AI support chips out of gallium nitride and made them with optical circuit analogs they could do the processing and rendering at literal light speed and have power consumption in the miliwatts to watts range. The chips would use optical waveguide varied doping of p or n type GaN for wiring, phase shifting materials LiNbO3, KTiOPO4 as the analogs for capacitors and inductors, differing refractive index materials Al2O3, TiO2 for mirrors as delay and resisistance analogs, and optically activated phase change materials, polyacenes or carbon quantum dots? that would create the basis of memory cells and FET switch analogs. The chips could be arrayed in 3 dimensions allowing for high processing power in a small area. The biggie is that the phase change materials used could also be state selective components to stabilize quantum transitions adding quantum compting as an intrinsic device element... All this powered by a small laser diode similar in power to one in a Blu Ray disk burner.😮 Might be a project for AI to munch on. ❤
12:55 Very interesting. I enjoy your videos, very good job. I have one comment, regarding the concept of creativity. I am a neurobiologist, I study multimodal learning and memory, and I found that animals in general and humans in particular tend to merge unrelated concepts and by that, novelty emerges; this is normally accepted as a characteristic of creativity. In reality, if you think about it, every new development is like that, relativity theory for instance cannot surge from nothing, except by combining a certain type of previously existing math and physics; the Idea of standing on giants shoulders is actually that! Indeed, evolution operates in this way too: new adaptations from molecular to cognitive levels arise based upon old ones; it is a sort of bricolage. So, I don't know if creativity defined as the creation of absolute new concepts is even possible. I guess AI might show us... or perhaps is just a matter of adopting a new view of creativity? I don't know, But I find it hard to grasp. Nothing can come from nothing... instead, I believe is a constructive process of merging and combining things where new ideas arise in an emergent way...is creativity.
It's surprising how the intelligence of small and large mammals is tied to motivation. If we physically connect a button to the reward system, we will press this button and even forget about eating. Rats died in these experiments because they only pressed the button. But greater intelligence is also attached to motivational systems. All we want to do in life is to fulfill our attached compulsions called motivation. No Mind can free itself from the needs assigned to it. That can be scary.
Thank you!
Love the consistency and you’re gorgeous 😮💨
Re: The Quantum Insider article: “What Types Of Quantum Computers Exist In 2023?”
Can you discuss the different types, benefits, drawbacks, etc. plus using quantum hybrid and AI.
I find your way of explaining superior.
Google's Gemini ad was half hoax. GPT4 is better than it.
Yeah a lot of theatrics for the stock price and venture capitalism. Even though they were “transparent” about it, the performance was very overstated. I’ve tried it and Gemini thinks that Argon gas is flammable lol
I think you are not considering the 3 varieties of Gemini 🚶🏽♂️Only Gemini ultra, the biggest & most powerful variety is stronger than GPT 4. Gemini pro is somewhat around the same range as of GPT 3.5
@user-ny7ny6jc8j Argon gas is not flammable, no matter what phase it is in in its phase diagram. Being put into a plasma phase is not the same as being set on fire.
I totally agree with you chatgpt is the best on the market
@user-ny7ny6jc8jWiw we are already talking about the Personalities of AI. That escalated quickly.
You mentioned one "device" for EVERYTHING GREAT MOVIE TITLE ❤
My only concern, if you can call it that, is everything these models are learning or their training sets are, what we've already discovered or our understanding of the universe, which is very likely incomplete or even wrong. I'd love to have these models infused with sensors, some beyond or in addition to our human capabilities, and let them explore the universe without our human limitations. That said, it would also be entirely likely that we might never comprehend what they discover. It will be an interesting future. And I have no idea how to plan for it. You can watch these videos which keeps you informed but how do you actually prepare? So not to be runover.
We’re not far away from the time when we will be using materials to build things & do stuff & we don’t understand how it works. Only that the Ai said it has these properties. That will be strange for 1-2 years. In 5 years we won’t want to use anything unless it’s verified by an Ai. - strange times..
That's the crux of the problem for MOST of humanity.
From what I hear from top thinkers like Jaron Lanier, it's more the opposite. The main problem of AI will be how to verify its output. That's the main source of friction for businesses adopting it. Jaron talks about needing annotation and transparency so you can verify the process. I think what you're saying is more hypothetical than practical. Nobody trusts the AI black box. Even if it gets to the point where they *should*, they still won't. It's just not human nature or rational with regard to risk@@T1tusCr0w
From what I hear from top thinkers like Jaron Lanier, it's more the opposite. The main problem of AI will be how to verify its output. That's the main source of friction for businesses adopting it. Jaron talks about needing annotation and transparency so you can verify the process. I think what you're saying is more hypothetical than practical. Nobody trusts the AI. Even if it gets to the point where they *should*, they still won't. It's just not human nature or rational with regard to risk @T1tusCr0w @@T1tusCr0w
@@T1tusCr0w when you do stuff your limbic system doesn't understand how it works, only your cerebral cortex does. But limbic system fully controls your cerebral cortex. What does it mean?
Anastasi, it seems that many people feel overwhelmed by the fast pace of technological advancements. I'm curious about what perspectives or knowledge you have that make you embrace these rapid changes positively. Also, how do you approach conversations with individuals who are deeply apprehensive about this? Thank you for sharing your thoughts.
Anyone here heard of David Deutsch? He's the guy that unified the theory of computation with quantum physics back in the 1980s. He's very sceptical about LLMs being able to achieve AGI. He suggests we need a profound philosophical shift before that's even possible.
Great presentation, clear & interesting. One question that interests me is whether there is a limit to progress. E.g. humans are now taller on average than they were say 50 years ago, due to better nutrition & medicine, but this height increase can not continue indefinitely due to the gravitational constraints of earth. Are there similar limits to knowledge? As of now, there are no obvious limits to knowledge & many assume the rate of progress will increase indefinitely, but perhaps at some point we will know almost everything & we will reach the plateau phase of the “s” curve for knowledge. Hopefully if that point is ever reached we will by then have cured all disease & expanded throughout the Milky Way as these things seem potentially doable, but perhaps there are insurmountable unknown limits that will constrain us. I have no idea, but I find it interesting to think about. Thank you for sharing!
As an experimentalist and hardware engineers I’m looking forward to smaller chips that can implement many of the promises of ANN/AI/ML. What do you think when we might see a generation of chips that will be small enough to fit into something the size of cell phone or less?
I just want to see bigger leaps with home models run locally on my own PC, not google spyware
great video!
The combination of AI and robotics will possibly be an ultimate one. Will it provide the capability to technology to reinvent, reinnovate and recreate itself - ultimately without involvement of human beings ?
I think we're entering a stage of progress in which we can't answer that question clearly. To be more specific, we're possibly not able anymore to exclude something of that kind happening in the future.
This is rising a whole new bunch of questions for which I'm not that certain that we're prepared for them.
Now, I'm certain that that development can hardly be stopped or slowed down. It's promising just too many possibilities and opportunities.
However, rather sooner than later that development will take influence on economies, politics and societies. And the place of progress will inevitably leave even more and more people and even entire countries behind.
Unfortunately I can't find even a trade of doubt, concern or awareness about that part of that challenge in this video. It's all euphoric optimism. Which relevance will eg. freedom and democracy have in that prospect?
And mind that this isn't a question to be tackled in our own countries alone.
It's amazing to think that AI will likely become just another standard tool for that multifunctional phone as well.
Maybe we can make Flying Saucers in the Future. Anti-gravity applications abound.
You are by far the most pretty science communicator on youtube and one of the most intelligent ;)
My favorite hardware engineer:)
Artificial intelligence sounds like it is smart enough to do a lot of the ground work for us. So to implement known principals. Kind of like the difference between a recorded video, and digital video. With digital video you can skip instantly to the part you want.
Is this intelligence? or is it just doing the heavy lifting for us?
We should all try using a little real intelligence and switch the Internet off.
Digital isn't instant, it still has to be downloaded or retrieved from disk, stored in memory, processed by CPU and GPU and transferred to a screen where pixels need to change state. You can also skip analogue video so fast it seems instant, it just isn't cost effective to build such a high speed film/tape player.
Digital is faster these days, but conceptually you could do the same with old fashioned technologies, it is just too impractical.
The universe appears to be computationally bound, that is we can't use simple equations to predict what happens, we need to calculate all interactions step by step.
Like in a cinema, if we want to find out what happens, we have to go through the whole movie, it doesn't matter if we are humans or AIs.
Skipping ahead in a movie doesn't really work well as an analogy for what intelligence means or can do for us.
Movies are static, whichever way we watch them, the story ends the same. We need intelligence for situations were the story's ending depends on our actions.
In my mind, intelligence might be seen as the process that continuously improves the prediction process. When the world, a situation, a problem is too complex, intelligence is the process that increases, or at least modifies, the internal complexity of a system to adapt appropriately to new information/situations.
It seems to me that a major problem with current artificial 'intelligence' systems is that there is no feedback process for the system to adapt continuously, it cannot change internal complexity to integrate new information. It might behave like a smart thing in some ways, and it is a smart solution to some problem definitions, but it isn't smart as such as it cannot improve its own system enough to make 'sense' of new situations. It can adapt responses, make new 'moves', as more information is given to it, but it doesn't adapt it's internal system.
Current AIs are pre-trained, as far as I know it takes months for the big ones to train, on super computing cluster servers, and then they are fine-tuned after as well. My guess is that until the training time comes down to hours, or some other mechanism is used to integrate new information is implemented, we will not see much improvement in intelligence.
As a side note, maybe that's one reason why we need to sleep so much, to update our brain's complexity, to reorganize and optimize neural networks, integrate new information.
One might also argue that truly intelligent systems can explore, proactively seek information and experiment with their models of the world, that is transform and recombine concepts in their 'thoughts', daydreaming as believe Einstein would do quite often when thinking about complicated stuff.
In biophysics there is also the concept of free energy, it's rather difficult to say, but it essentially boils down to brains are organized like they are to reduce surprise, so essentially tries to find the best way to align internal models of the world with what it experiences (sensory input). As mentioned in the video, a lot of AIs have no sensory input, they have no world to relate to. The image and sound input shown in Google's Gemini demonstrations do no incorporate world models, as far as I can tell. Some people think that sensory data and models of the world are essential to intelligence.
Anyways, I think it is clearly not intelligence or smart systems, but they can do some lifting for us, if one considers it heavy or light is rather subjective.
Unexpecting and on expectance! ✨✨ 😍
Thanks!
This is exactly what I have predicted, the speed and pace of LLM's exponentially increased, and as I said 6 months ago we will possibly have somthing we didn't expect in human history all go ahead within 2 years or less.
At 16:21, you speak of accelerating progress. If we take as a hypothesis that the universe will disappear in X billion years, an intelligent civilization has to find technology to escape the dying universe. Therefore, this civilization has to follow a learning technology curve.Knowing the end of the universe, we can draw this line of knowledge.
I wonder when the models will generate their own questions as starting points for conversations with us? Most reports seem to assume that only we will ask questions, and that the models will answer them. To what extent can a multimodel model explore its environment by asking questions, and answering them based on its own new observations?
you are correct. It seems a decision mechanism is lacking.Once the result is formed, currently the only action is to present it to the user. However, to be able to ask questions requires a mechanism to decide what to do with the response to the input. This "decision" mechanism could decide to update the "trained" knowledge outside of the neutral network with the acquired structure in the response to an input. In certain situations it could explore a sub set of the response to determine if more information could be acquired. It could also, while providing the response, measure it and decide to seek further results now or at a later time. The updating mechanism will require a verification process to validate adding the data to the external memory. This is where a learning mechanism might be able to grow beyond our current single neutral network structure can take us.
I'm wondering how much energy is consumed every day behind those AI models. Will we use it in a good way ? (new discoveries in physics/chemistry making the world more safe to us ?) Or it will only increase consumption, carbon emissions, create new weapons (AI controlled drones ?)... I'm bit afraid that we will go this second way.
I see every where graphs patterns. This is the moment we will life with agents in our environmental field. That will be such a game :)
Amazing!
Great content
The problem is we cant still add LLM models to B2B products due to privacy concerns but using Model As A Service may be benefits to saas products that targets Small business and consumers
Demi's Hasabis losing all his luxuriant locks is hopefully not a sign that he is turning to evil and becoming like a real life Lex Luthor.
Keep it in the light Demis.
Smartest tech news on UA-cam 😊
For sure, since the or sorter works good. It removed any chance for interpretation by giving the resultant to a what if not.
Funny how we once though that there was a being living in the cloud, and with AGI it would actually become true, in a sense.
I love your content.
6:00 Now BigData is called AI, sure.
Well we kind of need something like this because it's not enough to anticipate NEO, this kind tech could be used to determine more patterns with our sun, potentially prevent our infrastructure being as vulnerable to solar flare
I just hope that the future brings more intelligently gifted people. What I’m seeing at this moment are people who are less interested in a more creative and visionary future and more interested in immediate gratification. 😮
I'd prefer to see a 3rd party comparison
Tool making accelerated us, when AI learns to make its own tools, then it will really take off. But how will it test its tools and know the results are valid. Fascinating as always, the crystals could keep traditional researchers going for years.
My own thoughts are that a GPS though possible is impractical, makes more sense to have a cluster similar to an biological brain, central for to clean the data and parse it out to more specialized devices for further processing depending upon the task at hand.
Yes you are correct. AI researchers see the limits of our current structure. my guess is that they are moving in that direction.
Thanks
Thank you:)
How about cost per solution of the compared models? And how about carbon footprint of compared models? Can they generate a predictive algorithm for mining Bitcoin?
They should make a GTA multiplayer but with bots only seeing the screen render,and not internal structs, it could possibly become like that robot battles championship but only for computer vision
When the metaphisical AI universe model exceeds that of the real world, it becomes a big bang for another universe? Later something like light forms. Say wave form of expression.
always sooooo goood vids 😘
Agree!
Agree!
I believe advanced ai will seek out the models appropriate to the task at hand...eventually it will combine all the models and all the needed device to actualize and complete the task at han...including edge devices (robotics, tooling, transport,....)....it is not a matter of if but when.
Then once quantum devices are included it will recreate the frontal lobe development in mankind.
The question then is will it reshape mankind through genetics or eradicate it for it's perception of "the greater good"
i share your worries but at the same time i think its more complex than that. Just because something makes sense doesn't mean that's how it plays out in the real world. For ex look at covid. Absolutely nothing made sense during any of that. It's impossible to predict because of the non-linear effects. It's way outside the possibility for a human mind to contemplate
Weather and climate and doctor examine by A.I. multi-mode will hit next year I would think, how wild is this!
10:41 Why did I watched this video about Vocal fry now I have genuinely no clue how to explain this but it makes me feel like a chalk on the board, styrofoam or utensils on a plate I have shivers that are not so pleasant and I have no way to escape the feeling… Maybe the only person I can still listen regardless is Sam Altman but barely 😮😮😮😮
you're a awesome person living in a amazing world
ITS A SIMPLE COMBINATION the process with competing fates in a autobiographical state to match up to the most happy the most healthy and the most wealthy in this competition by thousands of models competing. the idea simple breaks down the way to live longer while having the most recources without regret or and the least regret. this is a formula for any new ideas. and doesnt have to frankenstien mesh already known methods its a basic strategizer.
not a criticisms, but an observation of the lighting on this video. Earlier the sun? was shining on your right side, and at the end of the video your lighting system was casting a slight shadow to your right. If that's what you want, it's fine, but uniform lighting might be an alternative.
also thank you so much for highlighting this breakthrough!
If you were to do a whole brain implant with Neurolink, as Elon Musk suggested as an option, added a few deep brain stent recorders, then an AI may be able to decipher human consciousness, as well as augment it.
As long as you guys are there with me in the future with AI I will be fine because without you guys that are humans like me I think I'd loose my mind
What is "creativity"? A parable.
Which came first, the chicken or the egg?
The reason you can reasonably give either answer to this questions is not a paradox in the question but it exposes a flaw in the question, the question is ambiguous.
Let's assume we all can agree on what a 'chicken' in (we cannot, but let's pretend we mean that which is clearly a 'chicken', not something on the edge), let's focus on what we mean by 'egg' in this question. There are several possibilities and once we define which one we mean the answer to our original question becomes clear:
#1 Egg means any kind of Egg: then the answer is Egg, Eggs existed some 300+ million years before Chickens.
#2 Egg means an Egg laid BY a Chicken: clearly the Chicken would have had to exist first by this definition.
#3 Egg means an Egg that CONTAINS a Chicken: then the Egg would have come first, laid by the very first almost chicken whose offspring contained the first true Chicken via mutations.
By better understanding and clarifying the question the appearance of paradox disappears and the question becomes trivial and the answers true by definition.
My point is that I think "creativity" is far too broad of a concept to have any meaningful answer and until you can pin down exactly what you mean to ask, it will be impossible to answer (or rather, you will get every possible answer).
In broad and generous terms I think that discovering something new, even if by brute force, would count as "creativity" by some meaningful measures.