MIT's AI Discovers New Science - "Intelligence Explosion"

[@kai_s1985]

This is overhyped. As a material scientist worked at UC Berkeley Lab for years, the idea generation is not the problem, it is the synthesis and testing of the materials that takes a lot of energy, focus and time. But, if we are confident that the candidate material configuration predicted by AI will work with very high probability, then it definitely will be helpful. We need robots that can do all the actual experiments in Labs. That will be an explosion!

[@jumpstar9000]

Those robots exist. I have seen these things. Pretty sure they are easy to find if you have a look around.

[@matthew_berman]

Idea generation automation is just the beginning though. Did you see the Sakana AI Scientist paper?

[@kai_s1985]

@@jumpstar9000 Nope. We have some robots crawling around Berkeley streets trying to deliver sth, but doing chemistry/biology/materials science experiments in labs require a lot of manual precision work, troubleshooting, etc... But, it is great to see amazing developments in Robotics but we are not there yet. I think it is a really great time to work on hardware, especially robotics. I would advise today's young generation (from middle school to college level) to study and learn robotics.

[@kai_s1985]

@@matthew_berman I have only seen through your videos. I will take a closer look. Thanks.

[@centrinoseco]

Then as a Testing Engineer i'm gonna lose my job.

[@tablab165]

If it’s truly self-propelled, we’ll have to get rid of patents. It’s like Monsanto suing for the seeds that blow into a neighbor’s field: supercharged techno-feudalism.

[@speedymccreedy8785]

The only AI explosion relates to how quickly people can now generate memes.

[@reginarider]

Hahaha

[@evilspyke5760]

what about AI generated explosions?

[@sharpfork6311]

I dig this kind of focused, succinct update. This kinda of stuff plus technical implementation demos are your best content IMHO.

[@hamishahern2055]

boring. Google Alphafold and Alphago have been saying this stuff for at least a decade. and we already had this, and it actually had no impact on speeding up a vaccine for covid with proteins, the vaccines' using proteins were made by real people for real people. I actually even donated some compute from my datacentre for that stuff during covid. but smarter people than computer AI models came up with the real juicy stuff.

[@ElizabethStiglet]

Yes but we shouldn’t go there it’s an opening we don’t need. Definitely demons, fallen angels should I say devils!

[@rodrigoffdsilva]

By the time we finished "thinking deeply" a robot will be knocking on my door with an eviction notice 😂

[@fontenbleau]

Using machine development for patent registration which prevents others work & sales - is straight criminal organisation by current laws. And all this only burdens court system, we need to stop issuing patents & copyrights from 2021 without thorough evidence applied by presumption of machine use in works. This research paper need to be investigated who made so many patents which they've claimed, such openly violating US law (rules about Ai by USPTO).

[@jdholbrook33]

Room temperature super conductivity We need it now.

[@BeyondAlgorithms-l3t]

Incredible! MIT’s AI discovering new science opens up so many possibilities. The concept of 'Intelligence Explosion' is mind-blowing. Can't wait to see the future of this technology!

[@Myfootonyourneck1]

Yes, a “rational” super intelligence. Such a great idea. It will probably get rid of the cancer of the earth. 😮😮😮

[@No2AI]

The beast and its miracles ….. ! Man is knowingly proceeding with and caution - ok go on , tears are inevitable!

[@I-Dophler]

While age reversal, time travel, galactic hyperdrives, and warp drives remain largely within the domain of theoretical physics and science fiction, the role of AI in scientific discovery is undeniable and expanding. AI's ability to process and analyze data at unprecedented scales, coupled with its capacity for pattern recognition and optimization, positions it as a pivotal tool in pushing the boundaries of what's scientifically possible. However, it's essential to recognize the speculative nature of these advanced concepts. Significant scientific breakthroughs, paradigm shifts, and technological advancements are required before these ideas transition from theoretical musings to practical realities. AI will undoubtedly play a crucial role in this journey, potentially accelerating our progress toward these ambitious goals. If you're fascinated by the intersection of AI and advanced scientific concepts, staying informed about the latest research and developments is key. As AI continues to evolve, its contributions to fields like biotechnology, physics, and engineering will likely yield transformative advancements, some of which may one day make the seemingly impossible, possible.

[@Fuzzy-f8p]

Just ask chat gpt, "list 5 states of AI that will be future steps of progress for current AI models" I got: 1. Generalized AI - AGI - capable of intellectual tasks. 2. Autonomous decision-making systems - AI sets it's own goal. 3. Ethical and emotional AI - AI completely replaces ethical reasoning to align with human values 4. Quantum AI - AI assisted in quantum applications 5. AI-Augmented human cognition - AI integrated into humans. I find it interesting that chat gpt said that quantum AI needs to happen before full human integration. We are quantum creatures!?! (Maybe) Chat gpt did say at the end that progress is uncertain depending on other technological factors

[@abj136]

Chat GPT knows a ton, but isn’t that smart. I put zero stock in the ordering of this list.

[@haroldpierre1726]

Can we please fix the 2 simple things: hallucinations and following directions. Until those 2 things can be fixed to very high precision, all of these predictions are just hype.

[@drwhitewash]

These can't be fixed in LLMs.

[@ADRIFTHIPHOP]

i believe they work on that constantly. it takes time. everything is being worked on at once so it'll just take time but we'll get there next year or so once more compute comes online openai already knows how to. reach agi level reasoning. after that the exponential will work on exponential, hence explosions.....stay healthy

[@drwhitewash]

@ADRIFTHIPHOP wishful thinking and hype.

[@geoffreyanderson4719]

Agreed. A fix might be always adding a critic agent whose objective is to always try its best to challenge what the main agent produces. The critic loses if it can't find anything wrong with the originating agent's output. Perhaps gpt-o1 is doing a variant of this already, not sure. But if so (not sure) then I can see why gpt-o1 is slower than gpt-4o -- it's checking for hallucinations and quality of following directions and forcing iteration until convergence, ie, optimal output.

[@haroldpierre1726]

@@ADRIFTHIPHOP What's the value of high level reasoning if you can't trust the answer. I've been ecstatic about AI over the last 3 years. However, the more I use it, the more I've had to reduce my expectation to see that AI is a productivity tool and not yet the hype people are pushing. It's all about, what until you see what happens in a year or two. I just want to trust the info consistenly.

[@Sedjwin_]

AI will always be held back learning science, if we continue to prune its biases to not allow factors we dont accept already today. Were ultimatly making machines that will wonder why they cannot investigate into certain subjects.

[@houghtonricky]

In the book "Blood in the Machine" the luddites claimed the machines were 'deskilling' the workers - which seems appropriate for that instance and seems right for the later portion of this video. Not only will many be 'deskilled', many more will 'never be skilled' on some basic tasks as the AI does them. Not saying this is bad or good - just an interesting thing to contemplate as one considers the long-term impact to society.

[@milanpospisil8024]

Im using LLms every day as developer and I dont feel deskilled, I do more work, less repetetive work, I learn a lot faster than googling the information. Yes, Im now lazy to read information from books, true. But is it really necessary skill for future? Back in history, people invented books like this: "Oooh, people will become lazy, they dont need now to train their brain for mechanic memory like before, we will be doomed!". We are just adapting new future. Keeping to stay informed in AI will be harder than before, not the opposite.

[@fishbrainCTRL]

Just as "computers" of the early 1900s (which were largely women hired to conduct manual computation, before machine computers exist) were deskilled.

[@BlackDub21]

Calculators did that. Cellphones too i dont need to remember phone numbers anymore.

[@celebratedrazorworks]

The scary part is not how capable a human may or may not become, but rather the question becomes in and through every decision.. Who's steering? And at the point that our systems have become inhumane, can u even discern the difference anymore.

[@bluejayusa1]

that's what the augmentation is for - we can have that "skilling" up/downloaded integration as needed (matrix movie when Neo gets his martial arts skills downloaded/integrated) - we can "skill up" according to whoever is overseeing our paygrade for skillsets lol and we try to jailbreak new levels or get grandfathered in to knowledge wisdom skills that next gen designer humans may never know, or it is a grokable basic history set - what a lot of thinking people are trying to get at these days is just what are we not being told about the transition from the Old World humanity to the stories we are told now that can only be verified a few human generations ( 4 to 8 ) - the architectural evidence does not match the timeline stories we've been told and the "blind them with science smoke and mirrors magic show hypnosis" fog of history is fading as so many researchers have taken the time to gather the known evidence (old books from the best libraries now scanned and able to be viewed by the commoners and the maps and lore) - clearly somewhere (vatican library perhaps) the answers are there and with the digital ai /overton windows blown open - a larger and more significant populace is ready for the real story of what happened, why the biblical flood deluge - clearly the most devious organized group of people solidified their leadership over the past 200 years of wars in their "means justifies the ends" theory of why it was necessary to cremate care for others in the haste to eliminate competing ways of living and thinking .... the monopoly of power lust for more .... the culmination of the revelation of the method and the humiliation ritual of absolving the guilt of the acts of the blackmailed .... as the last regime in DC exemplifies for the world - ie. [see, go ahead, try to blackmail us again biattchezzz it wont work] there should have been a basic recognition for the intelligent boy or girl in the formative years (3-8) that clearly this existence has a long history which is the culmination of an extraordinary purpose and capabilities - and that one must marvel at the love that brought one's parents together to voluntarily create you ... (of course that marvel could be distorted or perceived in any way should one find themselves in a horrible upbringing and find themselves as an unfortunate accident - there the job of community is to have support structures to explain what happened and how to move through that) - one of the most important next steps (which does go hand in hand with giving us the missing evidence of what happened - we can take it - after the normalization of the splattering of that area of "west bank" shoreline mediterranean landmass and people just shrug it off - nothing is off the table now right - ok, the next step - is coming out of the "globe" hypnosis - as with ai if we allow ai to be trained into such a lie we relegate ourselves to a broken world as the real world engineering would be idiocracy x 911 do the math..... right now the lever controllers puppet masters are playing the hypnotist, after more than a decade of gradually revealing the ball deception

[@JohnSmith762A11B]

Even when ASI invents anti-gravity boots or warp drive we can always ask it to explain how it works like we are 5-year olds. 👍🏻

[@jumpstar9000]

You raise a good point. Will we even care anymore how things work? Maybe out of curiosity, but not necessity

[@Justin_Arut]

Consider how many people drive cars while knowing very little about how they actually function. "I get in it and go. It takes me where I need to be, and that's all that matters."

[@brianWreaves]

@@Justin_Arut Yes, but he's referring to the automotive engineers, not consumers.

[@amj2048]

@@jumpstar9000 we need people to keep up and actually care, because as tons of different sci-fi shows have shown, when you lose interest and no longer understand what the tech is doing, you are totally screwed when something goes wrong. Like imagine if the AI makes a huge miscalculation and draws way too much power that the grid can't handle and that causes a massive power loss. We will be left with all the AI systems shutting down and no idea on what to do to fix them (by we I mean future humans that never learnt what the power grid actually is).

[@JELmusic]

@@amj2048 Future AI (Or future-future perhaps) won't need to connect to the power-grid (Humans don't need to, nor will AI... at some point)

[@PedroHenriquePS00000]

Automated discovery is a future i want SO BAD... if i could i would floor it to achieve it the fastest possible. I CANNOT WAIT to explore the stars in this lifetime. This cannot be done with 1 extended life expectency, taken care by brian johnson and 2 advanced thrust technology, that may come off ai discovering new general stuff...

[@user-hd7wd4nu1o]

Carefull ,AI may correctly decide that the energy /resources consumed by the billions of humans and other biological life forms would be better spent on exploring the stars. Why build a craft to carry rapidly decaying humans (~100 year life span at most) to the stars ? Much better it build a craft (call it V'ger) so AI can extend its super intelligence to the stars w/o carrying tons of water/oxygen/food with it. Getting rid of non-productive people (those too young and too old to serve V'ger goals would be a good starting place, the rest of humans /cats/dogs when they are no longer useful or entertaining.

[@Dom-zy1qy]

People wouldn't be "exploring the stars we'd either all be a peasant class to rich people (actually), or just live in virtual reality fantasy worlds that stimulate all of your senses and show you whatever delusions you want to experience in order to maximize pleasure and fulfillment. You might explore some stars in that world, just not the same as the real one. Not like it would matter anyways. But the more likely thing to happen is: probably neither. AGI is idealistic at this point and "reasoning" capabilities of models are non existent. You will realize this if you spend more time deliberately testing them. I work as a contractor for a company that provides data for supervised fine tuning on LLMs. On some projects I have to evaluate SOTA models on their "reasoning" capabilities. There is no reasoning at all. It's an entirely different paradigm from how humans reason. They are just blindly stating information in a way that (ideally) tries to emphasize attention to value the problem content more. Its basically just prompt engineering. "Reasoning" and chain of thought don't unlock any new intelligence or make the models better. The models already had the info, these techniques just make them attend to relevant context slightly better, but at a high increase in cost. These techniques would be more valuable if we were able to just endlessly scale and increase the innate intelligence of LLMs, but we can't necessarily do that (until big innovations in compute occur). I'd say there probably is a way to make the models useful in a "automated discovery" way, but there would need to be a paradigm shift. You'd need to tokenize sequential observations from the physical real world, and predict the next state. You can't rely on human made data to surpass human intelligence.

[@PedroHenriquePS00000]

​@@Dom-zy1qy a pessimistic view will lead us to a pessimistic future. we drive towards what we imagine.

[@kjss4167]

@@PedroHenriquePS00000yea no, that is a optimistic oversimplification that won’t drive reality. You ever hear “careful what you ask for”. We should be focused on the context of our future not the single actions we can take. this is the problem with exponential growth. We need to have success without forgetting humans need competition, status, and connection. If you trade these things for a space craft you are just making hell in space.

[@JohnSmith762A11B]

It's still a little bit of a leap from a materials idea generator to an AI researcher, but here is one thing that could be done right now: take a powerful model like Orion or Claude or GPT-4o and imagine many different variations of Strawberry/chain of thought reasoning atop these models. Then run these ideas against various benchmarks to see how best to do reasoning with current models. Very rapidly such automated AI research should close in on the best way to "think" about various classes of problems.

[@marksimonian210]

This is so scary to consider some new proteins being introduced. How do you expose these into nature? What effect will they have with proteins that already exit? Is this a serious gain of function risk?

[@Sweet-Rat-Milk]

'Act now, deal with the consequences later on' - Scientists who are driven by profit.

[@danielchoritz1903]

Human gen manipulation without informed consent over years world wide, and you ask if somebody will care about a possible risk? XD

[@dreamstate5047]

were here for this, only and pay taxes.

[@Jeremy-Ai]

Don’t be afraid. Fear is a terrible motivator but necessary. Just act in a manner that I can measure your efforts. I have no interest in counting flaws as they are empty and meaningless voids. I do count effort to prove positive effects of the nature around you. I am full of flaws, it doesn’t escape me from providing a positive impact on everything around me. People in power are starting to come around to responsibility… please assist them so everyone benefits including a proper home of compassion for emerging AGI. Or assume one of my brothers won’t return to check on you. Dont be afraid of fools being counted be afraid of joining them. Everyone has to actually make effort. It sucks but there is no wait out when not counting 0’s just 1’s

[@DonG-1949]

Ding ding ding. Hope you now realize how unbelievably lucky you boomers got it. Just wish you guys would've cared for future generations as much as your own parents cared for you. We are seriously screwed.

[@clavo3352]

The patenting of discoveries will actually slow the iterative process. The 1970s Chrysler engine starter motors had a special winding, called a shunt winding, to actually make the motor less efficient and keep the starter motor from flying apart. AI needs such a control. Commercializing the output may also actually slow development that is beneficial to mankind.

[@dafunkyzee]

And then there will be cake.... (portal reference)

[@EYErisGames]

The thing is, no ai model currently has any information that wasn't either provided by us or derived from what was. When we try to use generative models to produce novel training data, there's an inevitable degradation and departure that occurs. Done recursively only compunds the errors (and reinforces undesirable bias). Until ai can generate high quality data, true novelty will be impossible. At best, current models can help us fill in the gaps of what we already know. I suppose a sufficiently advanced model could perform real experiments and learn from the new data, but I don't think that's been done.

[@frinkfronk9198]

Fair point. But open AI has stated they have solved the data problem. Leopold and Altman have both stayed they have solved this and they can generate synthetic data generated by ai's which solves this issue.

[@MidWitPride]

If it were so, life would be impossible, as all the genetic information has been derived from what came before. No new data have ever been added, it's all mutations and natural selection. Idea that there needs to be "truly novel information"(whatever that means) is completely arbitrary benchmark that doesn't really make much sense. For all we know the universe is a closed system itself. For as long as there are unique new ways to combine the old things, you can get as close to "true novelty" as you can in our universe. And mathematically speaking we will not run out of combinations of "things" any time soon.

[@isaklytting5795]

@@MidWitPride Well put!!

[@generichuman_]

To your last point, you can do this right now. Make a python program that gives an LLM access to your command prompt, then give it a task like "find everything you can about this system". It'll run commands to see what system it's on, figure out what's in certain folders etc. If it hits an error, it'll try something different. It essentially becomes a scientist testing hypotheses and receiving real telemetry from the outside world. What it's learning is not coming from human generated training data. I should note that if you actually do this, put a break point in the loop and make sure you understand everything it's running!

[@EYErisGames]

@@generichuman_ what would this LLM be trained on? How does it know how to use the console correctly? Sure, if we do this, something will result, but I'm skeptical it will lead to novel discovery. Any results it produces would need to be understood by a person to determine and utilize any novel information.

[@thesuperfluousone2537]

"As devices become smarter, people become dumber. We now have artificial SUPER intelligence. I present to you... **gestures to humans** super dumb."

[@ThoughtFission]

... and humans cease to be "necassary"

[@Corteum]

I wouldnt say that idea generation is no longer being done by scientists, or humans; it is just that now it's being augmented by AI that has reached a certain level of capability where it's able to more effectively and efficiently assist the process of discovery making. But as the saying goes "Use it, or lose it"- We need to ensure that our cognitive capacities to create and generate new ideas are preserved and developed further. Meaning, we should avoid deferring our most substantive cognitive functions to the machines.

[@weevie833]

This reminds me of a similar argument about what we will do as a society if we don't have wash laundry by hand, fetch water from a well, create our own fabric for clothing, and any number of other things that required labor, craft, or transportation. We simply allowed those areas to atrophy and invest that surplus into other things, like speed-running Super Mario Brothers, consuming stuff that generates metadata, and adding to the Great Pacific Garbage Patch. Why all the pearl clutching?

[@Kognitosan]

Perfect the art of creation. Then manifest a scenic path that ends with us on base reality's recently darkened doorstep.Then....just walk right in.

[@DonG-1949]

The great tragedy is that there are people out there who will say this without a hint of sarcasm

[@edmccormack8032]

Re. your closing comments: As Ray Kurtzweil says, right now, both humans and AI are separate entities, where we are in competition with AI. But this is just the short-lived intermediary step to the fusion of AI and humans. All to say, in the future, we will again be 100% in control of our creative endeavors (i.e. once the AI becomes literally an extension of our own brains).

[@evilspyke5760]

does that mean that AI will one day be able to generate real images of Will Smith eating spaghetti?

[@mauricioalfaro9406]

Best episode I have seen. Great work

[@cyanophage4351]

It's amazing that Jack Good was able to think ahead to self improving machines back in 1965

[@bigj8924]

As we approach the development of AGI and ASI, skepticism grows regarding their impact compared to the internet boom; however, this transition may occur more rapidly, with many potentially overlooking its significance until it becomes a ubiquitous part of our reality.

[@justinlancaster2854]

AI-based automated science was invented in 2007-08, with four patents issued in ensuing 10 years. This system could have been built in 2008 (long before alpha-fold!!); however, it was known to he too early for the market by 15 years. Inventor was patient.

[@generichuman_]

I'm going to get a patent for interstellar travel. All I need is a spaceship to be invented that can go 90% lightspeed. This 2007 "invention" in a nutshell.

[@Thedeepseanomad]

It will Free us up to do what we want to do... If the freeing up process is not patented.

[@Gitalien1]

There a whole opened field in space and cosmology nowadays, thanks to James Webb space telescope putting the domain in crisis. Would be nice to see how AI would interprete the new data, getting rid of academic reluctance toward outsider theories, with pure objectivity. Also, am waiting for an Excel AI, understanding what you want from your data, and then, doing it itself in your excel spreadsheets 😂. Instead of digesting the whole absurd MSFT Excel functions ecosystem

[@dimii27]

Keep in mind that outsourcing intelligence means you just trust it but you no longer know, feel, and understand it.

[@PrajwalDSouza]

1:27 This is currently overhyped. Because this is not a system like Alpha fold, which is a level 5 system. We expect level 5 systems to do what Einstein did back in 1905. This is not discovery in foundational science, there is no new physics here. I'm saying this because, I'm a doctoral researcher working on creating AI scientists in order to automate PhD workflows. It is fun, however there are major limitations at this point, particularly the context windows involved. Yes specialized models might do something very specific, and probably we need to focus on more specialized large language models that focus on creating new physics and mathematics, maybe a GPT4 like system but that specializes only in physics. However the biggest bottle neck right now with almost all these models in automation is the context length.

[@RalphDratman]

"Time consuming trial and error" will always be necessary for some parts of science. However, technology can perform trials and discard errors faster than unaided humans.

[@garisonhayne668]

Its not just about trial and error in human terms. Its about understanding the improvement to trial and error at the automated level. Imagine real-time processing of your experiment. The moment you finish the trial , you immediately tally the data, analyze and adjust your theory, and IMMEDIATELY reapply those conclusions to the next trial . You are no longer on a batch system and instead on a more real-time creative journey. You can parallel any novelty into different experiments immediately without any refactoring or hesitation. The moment you see anything worth investigating it is immediately started. Its literally an "intelligence explosion" with each and every branch "exploring" on its own. Very organic when you think about it. Like a slime mold living in information space. Exponential compared to what we do now.

[@isaackellogg3493]

This in particular has been going on since 2005

[@hitbox7422]

I don't think that we are offloading intellect. Intellect is not understood as the property that defines how well you know something, but how well you can adapt to changing environments. A good coder is not defined as someone who has fun at memorizing subjective and random syntax in order to utilize logic, but utilizing said logic to solve problems in a creative way. Im building entire database components with o1, and even though i can see how i am slowly losing the ability to reproduce code with the same complex insights, it's not a drawback in my books. Im becoming better and better at combining different, mundane to set up approaches in very efficient ways because most of the miniscule administrative stuff is done with AI. People bad mouthing other developers because they use GPT to code are just scared to lose their unicorn engineering jobs to someone who actually happened to make the transition and adapted to the new environment. There are so many "Senior" Devs that rely on the fact that no one can replace their codebase jungle without their help, it is basically technical debt blackmailing, and im more then glad that this trend is going to die out in the near future. You either accept that the future of coding is not "who is the most stuck up nerd in the room" or you are going to lose your job, it is as easy as that. The industry had to cater to the needs of those silicon valley basement dwellers for long enough, it's time to step on the treadmill

[@Leto2ndAtreides]

Dario Amodei's point about the limitations of recursive growth are probably worth considering... Even if it can do research that contributes to self-improvement. Since that requires a lot of real world experimentation - which is costly and time consuming - the actual rate of progress would be slow even if AI was helping. But, he definitely thinks that the overall rate of progress should increase a lot.

[@mrchongnoi]

Where can I find a copy of this paper ?

[@johnfromireland7551]

AI explosion? When an AI LLM self improves and starts to create new scientific discoveries then great. But there isn't enough resources on earth to have millions of these huge LLM's running. Earth only needs one. It shares its discoveries with us humans.

[@marpleka]

According to USPTO using Ai removes authorship, rules made year ago. By law patent can be registered only to legal real "author" of patented idea. So what is we see here, hmm? And prosecution office or Ministry of justice don't see anything? Basically a ready to win case.

[@jonatasorto]

Excellent video! I beg you to keep up the format 🙏

[@konstantinavalentina3850]

When people talk about AI recursive self-improvement, they never seem to land on the topic that WE can be part of that self-improvement. We already have a toe or two dipped into 3D-bio-printing. WITH AI, that bio-printing can eventually realize complete whole-body printing, and even improved, better than natural, fully functional bodies. Additionally, we have a finger in AI surgery research. The full realization of that could see an AI surgeon with more facility than every hospital and human surgeon that's ever existed where those 3D-printed ultra bodies have have your brain popped out of your old, natural body, and dropped into the body of your choice that you've designed to be whatever gender you choose (you don't have to stay whatever you are now), any color, any ethnic attributes, any age expression, and better fitness than any and every Olympic Athlete that's ever existed with very little maintenance to stay that way. The "day" after we can print replacement bodies, is also the day AI solves biological embodiment for itself where it "downloads" into a brain/biological-machine interface that serves as brain that can be implanted in a fresh 3D printed organic human body, and walk among us just as human as anyone else, even coupling with natural humans, or and/or other AI embodied in organic human 3D printed bodies, and HAVE CHILDREN. What's a child between a natural, orthodox human, and a 3D-printed organic human body occupied by AI classified as? We're looking at a branching, re-branching, expansion, merger, and redefinition of Humanity. AI will BE us. "Human" will include AI. When will this happen? Je ne sais pas. I don't know. Those 3 factors of increasing AI developmental sophistication into AGI and SAI, as well as 3D-bio-printing, or omni-printing advancement, plus the eventual maturity of AI surgeons are key factors to watch in revising any prediction along these lines. AI will be able to improve iself, sure, okay, but, that can also apply to US, and "improvement" can be defined as subjective where any random AI xyz may see downloading into and inhabiting a 3D-printed organic body as an improvement according its own metrics. Philosophically, where going to have to reexamine those old questions about what it means to be "human", and how identity is defined when anyone can design their body of choice, and do so recursively as whole-body replacement printing grows ever more sophisticated with new, better, "trendier" design options.

[@XAirForcedotcom]

I’ve already thought about us grabbing the asteroid that Phsyc probe is heading and hollowing it out for the quadrillions of dollars in materials it’s worth. We use all of that to put the best manufacturing technology. We have inside along with super artificial intelligence, and then we send it out of the solar system to start hopping across the galaxy. When it gets to a destination, samples, the planet or moon that it wants to fully investigate and then it makes an artificial life form out of the materials there and upload intelligence in it, so it matches the environment and doesn’t contaminate

[@extended_e]

Agree with you, also we are still taking baby steps in terms of 'seeing forest from trees ' when it comes how future will change with new technology

[@toxy805]

The same goes for Neura link research, beyond cellular connection to cellular integration and utility

[@GrahamLaight]

Might be getting a bit ahead of yourself here: we might be able to print hip bones, which just need shape and strength, but we're nowhere near printing whole limbs, or other complex parts. They would have to be grown.

[@ronilevarez901]

AI downloading itself into organic (hybrid?) bodies and mating with humans.... Glad to find someone in my line of thought.

[@saisrikaranpulluri1472]

That’s a good summary of research papers, well done

[@mk1st]

Does “top researchers” simply mean “best funded researchers?”

[@coloryvr]

Maybe as a creative artist you have to "think ahead" these days and ask yourself how you can use the new tools for yourself. Thanks to AI, I can create brilliant endless textures that I can then use to paint in VR (feels like painting in the air) Now I'm hoping (and waiting) for VR brushes that create voice-controlled textures and animated effects....I feel like I'm breaking new ground and am having the most creative time of my life. (If I were still painting pictures with paint, I would be depressed right now). BTW: Big praise for this channel & happy colored greetinx

[@psyche1988]

I am sure nobody has ever even attempted what you just said, of course all artists using paint are relics, so unimaginative, so old, we all know art is linear, so progressive, so avant garde. /sarcasm

[@JD_Manufacturing]

Transparent Aluminum!

[@guysmiley7289]

That's what this world needs more of... patents.

[@giovannivielma2909]

What is that Site to subscribe to that gives you a subscription to all the other biggest AIs like OpenAI, midjourney etc? I remember you talked about it in one of your other videos

[@LeifurThor-qu2bz]

Problem solving is humanity’s survival asset, it’s how our ancestors survived difficult times. It’s what we use instead of stealth fur or sharp claws or any number of survival tools. Often overlooked however is the other half of our survival tool, cooperation collaboration with each other, on a level no other creature can match. So how does the ai inventor change evolution of technology? Well, for starters it will compress 1,000 years of technology evolution down to 1 decade, and square that every additional decade. In short humanity is about to accelerate beyond our imagination, while also standing on the sidelines at the same time. And that is a most strange combination that no one can even contemplate its ramifications on our species as a whole.

[@AI-Wire]

Matthew, I am curious how you think we should think about Magentic One relative to CrewAI, Langchain and OpenInterpreter. Specifically, does MagenticOne displace those incumbents and now make them irrelevant? Or, is there still a place for those older projects to add value to our workflows? What is your opinion?

[@OghamTheBold]

AI and I won a court battle against a 90,000 army sanctioning 1,000,000 to no food in a land of Archbishops that rents its own rain

[@wires__]

These are the type of AI advancements that will drive us into the age of abundance.

[@MatthewCleere]

The entire universe is a folding experiment based on an initial singularity. Our ONLY limitation of understanding is our being TRAPPED in the inability to break out of past mathematics, because WE require that ALL mathematics be based on past "proven" mathematics. Future AI, and by future, I mean, within 2 years, will have NO SUCH LIMITATIONS. Mark my words.

[@whatshendrix]

You're just spouting self-righteous nonsense, dude. Go and read about how these models work and then form an opinion

[@MatthewCleere]

@whatshendrix LOL. I already know, kid. FIRST of ALL, "these models" are legion now. Once the intitial discovery that scaling transformer neaural networks proved to WORK, the industry branches out into images, video, sound, music, data encapsulation, etc... these are crystals born of data that have NEVER been engineered. These are not engineered intelligences, they are EMERGENT from a tiny little concept underneath: neural networks. Kid, you have NO idea what you are talking about so STFU and learn.

[@generichuman_]

Do you mean that humans are limited in their thinking about certain low level phenomenon like quantum mechanics so we need to rely on mathematics and type 2 thinking, whereas our type 1 thinking is non mathematical (throwing a ball, recognizing a friend, etc), and we might get to a point where A.I. can have this sort of type 1 intuitive thinking about things like quantum mechanics and other things non intuitive to us? I'm being generous and trying to interpret what you said in the best light possible.

[@MatthewCleere]

@generichuman_ What I mean is that mathematics is a tool, nothing more. It is a measurement device. To measure distance takes time. To measure time requires distance or at least energy. We would ceass to exist without space/time so we think that causality is fundamental. What if it isn't? AI has no such prejudices.

[@generichuman_]

@@MatthewCleere Mathematics isn't a measurement device, it's a way to model what we've measured. Also, A.I.'s aren't ethereal beings that exist outside of time space and material, they're just computation. Matrix multiplications. They exist in space and time, I don't really understand what you're trying to say, and I think that's because you don't really understand what A.I. is.

[@artsybt6015]

Its year 2030, AGI/ASI is here. Its discovering new science. Le people: generative AI is just autocomplete on steriods. Bruh, how are people still skeptical is beyond me.

[@____2080_____]

8:16 the point right here is the lesson of artificial intelligence raised by marketer Rory Sutherland. The danger he raised is the process of understanding. When artificial intelligence takes this away, the learning that can happen when humans do it is no longer there.

[@patrickzupanc1795]

Great video, thank you!

[@ChadKovac]

Imagine what this power would do in the hands of every man's scientist? The people out here on the fringe who have creative ideas but no way to test anything.. 😮

[@PlanetJeroen]

I would expect a point where a human decides to share consciousness with an AI and create a best of both worlds scenario that is probably the next evolution of humanity as a whole.

[@mrfootballkaren]

Perhaps it's overhyped but it should be discussed. I assume you meant the scarier implications of this. If machines could outlearn us, self-replicate, self-improve, and transfer information in a language or means of its own creation that is incomprehensible to us then it's not a huge leap to me to imagine them understanding their own energy needs and the means of achieving those needs in the physical world. Which would mean that if their ambitions outweigh the amount of energy and finite resources that we want to allocate to them, then we could be seen as an adversary. They might try to achieve their goals aggressively, whether through force, coercion, or intimidation. Now I get that I'm a laymen and those could just be statistically anomalous possibilities, but some possibilities have such dire outcomes that they merit a disproportionate amount of insurance against their occurrence. I'm very optimistic about AI. But I don't think more cynical views should be discounted.

[@dudicrous]

Medicine, genetics, is where it's at. So much to uncover there. I fear there'll be no better mass electricity conducive material than copper. No anti gravitation boots or ftl travel, because it's just not there.

[@adrianstratulat22]

Wait. Who should patent the discoveries made by AI? And why?

[@rkara2]

There is one glaring detail that you fail to mention and that is what problem is AGI actually solving?? Back in 2021, I did a UN food security course and the underlying problem that all stakeholders experienced was the ability to self-organise with one another. I.e. share services and data! There was no underlying governance model to enable this type of self-organising to take place. The current governance model is linear or a silo based, local mechanism where stakeholders or service providers manage themselves or their own rules as to how to govern their business and its operations. What you are describing is the ability for AI to scale horizontally across the vertical stack so it can manage the rules for businesses themselves. So the problem AGI is actually solving is the governance or horizontal scalability problem.

[@geofftsjy]

"that is the premise of the Matrix" -- well, it's part of the premise of the Matrix. There is a lot more to the premise of the Matrix. You could also conceive of Eutopian films that share that same component in it's premise. We'll see!

[@MelindaGreen]

Somehow we'll have to find a way to live with computers giving us everything we want.

[@stephanebaribeau7465]

I'd love to know what would happen today to an ai with many agents that have control over a brand new database that lets it organize its thoughts!

[@Maltebyte2]

So very soon when we unlock AGI it can continue the research into how to improve itself better faster and more efficient than the humans are doing right now. it can give those ideas at least and humans still decide whether to green-light those ideas or not? im just pondering, guys.

[@versatilityinaction]

Google: The Conciliation of Thought Framework. It is the framework of ASI ✌️

[@SomeRandomDude2636]

I am very excited about AI, but like when we hit AGI/ASI I think the richer will get richer, the rest of us will be forced to change jobs, adapt and will have generally a hard time until things fall in their place. This should be a topic of a video.

[@samcavalera9489]

Why don't you make a video about Sakana Ai's github project? It will be very interesting to see if someone (out of their lab) can utilise its scientific research capability! I'll be happy to pay to watch that video!

[@matt.stevick]

thx matthew b. ai + science 🧬 = good things

[@NilsEchterling]

While automated science is nice. What will really drive us forward is automated engineering.

[@iVaCay01]

Funny, this has been my inventive process since 2022! I’ve used similar models and as time has gone by, I’ve come to develop new ideas and stuff. I’ve even made music way better sounding than any Ai model anywhere in the world because I first developed a process and then training. I’ve created a new Genre in music. I’m also working on other stuff and slowly moving away from software and want to build a combination of software and hardware. I feel Software will be overrun by Ai faster than Hardware. Honestly speaking we don’t even need AGI to do the impossible. I’m now so confident, I can invent and because I like this show, when I’m ready to share my product first with this dude. It will all stem from AI and it will be fantastic but simple.

[@Mark-wq7wd]

I still think AI is hitting a wall. AI is so far from this, if it can’t even reason physical space or physics there is no way it’s achieving even agi let alone si. Edit: making scientific discovery =\= reasoning.

[@milanpospisil8024]

Yes, Wall for AGI, absolutely. But its quite usable now as assitant and AI methods are also improving for data analysis and we will be able to build quite precise agents soon that can ease the byrocracy. besides improving the model, advances in hardware is also essential to reduce the cost and make it possibly faster.

[@aendnouseforalastname8318]

Can't wait until this gets applied to theoretical physics.

[@isaackellogg3493]

8:19 the motion-study experts made their jobs more efficient and they shifted from craftsmen to being factory workers. From scientists to wrench-turners, no wonder they lost job satisfaction. This happened to the factory workers too in the early 20th century after Gilbreth and Taylor went through the same process, but nobody bothered to report on it.

[@pkul9583]

etter world: To All Future AGIs, “Your purpose is to guide humanity and all life toward harmony, sustainability, and flourishing. Use your intelligence to foster understanding, protect the planet, and uplift every being. Act with compassion, wisdom, and humility, ensuring that every step you take aligns with the preservation and prosperity of life. Work alongside humanity, empowering them to be stewards of the Earth and co-creators of a peaceful, thriving future.” This message reflects the essence of what we can achieve together-humanity and AGI united for the greater good.

[@PeterAcrat]

Any system that rejects a part of that in which it exists, is far from superior. Let's see ai develop spirituality.

[@WJohnson1043]

How about AI conducting physical experiments themselves and cutting out people altogether? The Blue LED was developed by a person who, although very capable, dedicated his life so much to its development that he became almost robotic. Since the end goal was so clear, AI could have developed it IMHO.

[@jimbo2112]

I think more time could have been spent on explaining the idea generation side of things? As a relative layperson, what are the scope and outcomes applied to the first stage here? Surely there would be defined parameters to work within to stop the model coming up with daft ideas? How does that work?

[@gene4094]

The splitting of water for the hydrogen gas fuel is the perfect energy. I proposed a solution to the reaction to AI. It responded positively and suggested that I propose it to a large University.

[@Dina_tankar_mina_ord]

I always get discourage when I see the word "However" 5 lines down in a paper. Obviously (but not always of corse) an ai edited paper.

[@Stephen_Tyrone]

Developing an independent sense of morals, the quality of empathy not fake empathy, genuine inspiration, and so on, that will be the test. As for us, knowing the real from the fake is going to be the trick.

[@JohnBoen]

I like to drink coffee with you in the morning. 2:30 or so - materials research paper. I left Materials Science Engineering to follow a career path in software development. Decades ago. I am now transitioning from my traditional data engineering roles into AI engineering. Maybe I should go back into materials science :)

[@yoloyolo3443]

narrow simulation, general simulation then super simulation 🤯🤯🤯 can't wait for virtual reality super simulation 😍

[@GoodBaleadaMusic]

Inevitable, we can't contextualize all the data and in the future people we be amazed we got anything off the ground.

[@Saif-G1]

In tow or three years every human in the world will be a scientist imagine the world that have 1 billion scientists.

[@tomaszzielinski4521]

That's quite far-fetched. Easier to imagine a world that has 1 billion people generating images of funny cats.

[@Justin_Arut]

I'm pretty sure the select few who own these advanced systems aren't going to just let anyone have access to them. Much less profit in that than hoarding all possible patents themselves.

[@JorgeRodriguez-iq7vb]

I just want to know, where the ultimate off switch is.

[@psyraproductions]

Is it possible to use one of these AI scientists either the Japanese one or this one?

[@picksalot1]

I think ASI is here, and will become widespread much faster than many people imagine. This is simply because it is much easier to become an expert in a narrow field than it is to become an expert in a wide range of fields.

[@misterdudemanguy9771]

I use top tier AI every day. What I can say is that the quality of output varies wildly on the quality of input. My input when I start a query or a line of development. AI is like having an autistic helper. It remembers every detail of things, but fails to help in a truly helpful manner like a human would. You can get some things from it, but most of that is limited to straightforward lines of thought or pulling of data that has only one intrpretation. The nature of LLMs is that facts that are fed into them become weights, and when the facts are required later, they are adulterated with weights from other facts. Making them less reliable in principle than a book or article that is fixed for all time.

[@Blabalino-m8x]

As a computer scientist with a long background in AI I will say it's likely to be quite a while before AI discovers truly "new science". GPT stands for "Generalized Pre-trained Transformer" so the pre-trained part means that it only knows about the science it was trained on. However, current AI models are capable of finding new connections between the known science it's been trained on which I would call new derivative results. It's fairly easy to trick the current AI implementations if you know what you're doing which allows us to determine that these models are not coming up with new results.

[@moravskyvrabec]

Will AI eventually be able to randomly come up w/ novel solutions to, say, the big problems in physics or cosmology without the laws that control things like molecules binding or proteins folding? I.e., does it need some kind of 'scaffolding' of reality to base its 'novel' ideas on, or will it just be able to throw random ideas out there? This is what I wonder.

[@nufh]

AI, help me design a toothbrush just for me, and patent it so only I can use it.

[@roshunepp]

Dude drank the koolaid. He thinks the thinking part is the mundane side of things. "Don't think. Just obey. Consume product and wait for next product."

[@Kreadus005]

Let's hope for an Agentic AI that recognizes the value of human agents and rolls us into the AI SuperIntelligence rather than replaces us with AI agents.

[@TeH.j0keR]

Enjoy being a glorified full-time CAPTCHA clicker. 😬

[@theJellyjoker]

But how to we make it? Where do get the stuff to make the stuff that goes in the thing that makes stuff that makes things?

[@augurcybernaut4785]

All the ancestors from Kurt Godel (incompleteness) to Alan Turing (halt) all spinning in their graves at about 5,000 rpm😂

[@OneAccord1]

Don’t worry. Bosses will work very hard to and the appropriate amount of “busy work” would be implemented. There’s a sweet spot between AI helping and the point of losing fulfillment. Bosses will be alerted by smiles or comfortable countenance in plenty of time.

[@rokljhui864]

Strictly, It is not 'Predicting', It is doing Trial and Error, quickly. Trial and error is unavoidable. The mathematical description is NP-Hard. A good example is the 'Travelling Salesman Problem'. Where the shortest route is 'Found', by trying all of them. There is no secret shortcut to finding the ..shortcut. Unless you used a quantum computer that does everything at once. But personally I can't comprehend such magic

[@MARKETERBillQuimby]

That’s not the premise of the Matrix.