วีดีโอ

LeCun is spot on saying that pixel-based generative models don't learn deep world models. They generate images that 'look' good, since they are trained on appearances, but they do not learn world models. A great example that I saw recently was a 'space babe' in a spacesuit that did not have a join between the helmet and the suit - the AI generated something more like a motorcycle helmet, because it had no idea that the suit needed to hold air. Another example was a video that showed a first person viewpoint entering a library. Each image frame was consistent, but it was plain that the inside was larger than allowed by the outside view - the AI had no mental map of the library.

So my final statement: What molecular chain will act as a 555 timer: when using photons instead of electrons? In programming a Proton: life does this by removing an electron from Hydrogen, starving the proton, then filling it with the light of a memory, followed up by returning the electron: it can then make a more permanent memory by rebinding that proton into a Neutron: where the electron has also been stated with the same memory by another proton in that or shared with another protein in a chain or synapse. Many protons in sequence can form a greater memory chain of events, and even program sperm with the new instincts. In life chains: please do not program living or once living hydrogen atoms: find truly dead or archaic sources for your memory chips. While hydrogen is everywhere: and neutrons are as well: they can reintegrate, and be repurposed for living function. As such: they have already been formatted for life, and work more easily in the conditions than archaic hydrogen not found circulating in living biological food chains, some of them have been around since LUCA, and it would be a shame to destroy that old and circulated of a living memory for yet another copy of the Beach Boys: Surfing USA. Could you program a song into the rings of Saturn? How about modulating the storms on Jupiter to display a data set: as to be downloaded from Cygnus by everyone all at once without a directed data burst? Well then maybe modulating the surface of a proton is a bit beyond us yet, but life already does it, and since biology can already do it, it can only be assumed that we will eventually have the concepts of technical ability within our ID to validate our Super Egos position and allow it to explain how life already does it at the cellular level: and how we can mimic it with technology at our level, altering a gas giant is the same just at a greater scale.

A better explanation of super-position: Each point in existence is a zero, at this zero time is not present. Inside the zero you can find negative dimensions. These negative dimensions will interact with the positive ones in an expected manner. For each point in a line holding weight in the First dimension, it is possible for that point on the line to be the center of the weight in the field of the second dimension, again in the third dimension, and so on into the fourth and fifth dimension. Two lines are irrelevant to each other unless you view them from a higher Dimension. Time starts at 0, the line, plane, and 3D object can all change without the 4th Dimension. The 3D object does not have an external point of reference and thus can not have a position without the 4th Dimension: just like the two lines. Each dimension can be broken into Domains of State and condition, 0 is the formulation of the change of the value of 100% in a 200% system, where 100% is the maximum field size allowed: making everything else a balanced negative. We are mostly Vacuum, and just a tiny bit of Mass. Change is fast when 0 sits beside 100. When it is 50/50, or 1/1: is it a singularity and static, when it is 100% mass or vacuum: it is an absolute and can not change. To view the entire 200% you need to increase the scale of the domain: which equally increases the new true weight of the system to 40000%. Zero has 1 domain, 1D has 2, 2D has 4, 3D has 9: with all sixteen states full of the maximum of 1 can be in the 200% system: it enters the 4th dimension: this only exists because it has an infinite number of neighbors now relative to its position. 4D has 16 Domains, we exist on the Eighth domain of the fourth Dimension. The First Domain of the Fourth Dimension is Gravity, and traces the First Dimension. By the Fourth Domain of compounding the highest value of mass which at the core encroaches 100% of 1: and spin rate is possible of the object: and now a true body of the 4th dimension with gravity, surface area, volume, and mass. The Fifth Domain Continuum body of the Fourth Dimension has a core that is a single point of negative Gravity or a component of Mass-time. The Core of a Sixth Domain object has a core relative to the -2D in extrapolation, a form once only found inside the zero, now projected as the expanded Zero beyond the event singularity of the object of the 6th Domain, and has thus equalized a core of the 2 domain as a negative: thus the core of the Sub-atomic particle has a negative core surface area found without any represented volume, as a 2nd Domain body of the Fourth Dimension would also not have a relative volume: The negative surface area is a reflection of all points of vacuum, as a 2nd Domain Sublime point would have a surface area that equals all other points of equal scale it is relative to, but would have no volume, and just sit as extending lines of First Domain Gravitational subliminal energy of linear 3D points in alignment. Mass inside the body is approaching and surpassing the inverse absolute at the core: and the 200% system is thus working in reverse: Mass-time: providing a reverse perception of time. Inside a sub-atomic particle is a domain of Mass-time: that provide life its intuition over the living matrix of the 8th domain space of the body: this applies to every Sub-atomic particle in the cosmos, and higher cohesion is stated inside the single organism which is a lifeform: the core is only a 2 Dimensional sublimation, or sublime point of reference and is in-objective to change without multiple statements of them over a 8th domain or higher extrapolation of expected or anticipated events. The Planet is a 9th Domain body, a Star is a 10th domain body, a black hole is an 11th domain body. The Black hole has a core that is equal to the 5th Domain of the 4th Dimension: or is one step below the sub-atomic particle, as a hypo-atomic particle: and thus has dimensional properties of volume rational to as if it were mass: as it still is: the vacuum is presented as the particle in Mass-time. This hypo-atomic particle has a much larger relative field of interaction: and thus its extrapolated anticipation is higher: using a field of black holes core to provide anticipation is then possible for higher cosmic events: yet all is subject to Hiesenburg uncertainty upon your interaction. Prophecy works via the interaction of our 8 Domain bodies with the 11th Domain ones: yet is irrationally unreliable in and because of the human mind. A true Quantum Computer: is not feasible until we are a Type 3 Civilization, and able to utilize the black holes in this manner: else we need to design a much more elaborate field statement: as things are not in a valid super-position until the temporal condition component of a sub-atomic particle is truly separated and isolated. The Quantum Computer of all probabilities at once: will not work until you can make or build an Atom that extends into the 8th Domain: so that it has a 4th Domain negative core. The center of a Planet has a special Atom at the core with such a -4th domain: so hollowing out Ceres for this type of experiment is the best solution: a place where critical mass can be sustained indefinitely due to universal pressure and expected density, as we would otherwise need to make an atom so heavy and enriched: that it is easily visible: and would literally have entire Atoms compressed in orbitals as Sub-atomic particles would be to a normal one, these orbital atoms set in periodic layers: a giant series of buckyballs stacked onto each other and compressed so that a true core Mass-time core is established, and many of these would be used in concert to predict anticipation and promote a solution to every presented outcome: and outside the realm of anything like a rational clock rate: as it is negative dimensional in conditional output. You would need a field of such hollow Ceres bodies. Photo-circuitry is far more rational for our age. So the Super-Position of an Atom: is already anticipated by the cores of anything larger than 12th Domain of the 4th Dimension, thus is never in a super-position at all: you fate it known: but the answer is otherwise still outside the Universe until it happens: and if you make the right choices: you might one day interact with this field: and thus make the concept of a set fate back into super-position. Light speed is a limit set by a man to use to measure things: it is not an absolute: super-position is your inability to measure: and not a product of an obscure outcome related to mistakes in physics. Anything smaller than light: is free to go faster than a photon: and you would again: need a hollow Ceres to detect such an energy: or that probe from The book and film: Sphere: that feels for gravity instead of looking for photons: hypo-atomic signals: travel faster than light, not time. Nothing can happen instantly: thus even with quantum supremacy: it will always take time to answer. Life already programs a proton directly with memory and instinct: we are not even close to hypo-atomic calculations: and that still has a functional speed limit to relative form.

This leap is illogical. Using a photo burst to cause an expected cascade of change: by altering an atom's electron valence states of excitement and pulsing in a molecular solid(crystal lattice) is far more reasonable. Coupling this with recording on graphite sheets with nitrogen/carbon enrichment as the hard drive: allows for computation past the Moore Barrier, as by using photons instead of electrons as the potential in switching, and the atom as the switch itself. While a step in the right direction: this is also beyond rationality to produce without first building a computer to solve the three-body problem. The first step to gain quantum Cohesion is to make a 4 dimensional computer able to solve the 3 Body problem, and onward to the billion body problem. Such a Translational Energy matrix or 4D calculator is possible with conventional computer components, as it is a condition of geometric arrangement of circuitry, and not about the way the computer chip is manufactured: size will be the only issue, as these computers will return to the days when they filled massive structures. The concept of a cube is broken into 5 Tets, as four active quadrants of measure, and a fifth condition for vacuum. Mass accounted as 1, vacuum is 0, and both must be calculated to provide the true result. The calculation of an atom will require a Tet^7 computer, or seven exponentiations of the original cube in the 5*Tet format, and will not include the Universe it is found within. An Atomic Cascade Computer would calculate faster, and be smaller, but would still require the Tet arrangement to provide a dynamic simulation of a three dimensional environment and the solution to the Three-Body Problem. True function of the Three-Body problem in real world mechanics: requires the inclusion of all sources of gravitational interaction found in relative concert to the subject body: as not just what is occurring on Earth, but also includes the moon, sun, local stars, and even distant black holes, as all of these need to be included in the balancing of the isolation of the subject body, and make the solution valid for real world use and explanation. It is reasonable to make the Earth a small blue dot off in the calculated distance to first achieve Quantum Supremacy without such external interference: as is required to remove all noise produced by the Earth's inhabitants and their common electronics. The atom might not have enough mass to be relative to the black hole, but the black hole has enough mass to be relative to the atom.

interesting insights.

He's thinking about how we think

Good to see he is able to talk about AI and not only about Elon or politics…

judgmentcallpodcast covers this. Keynote on AI's essential characteristics.

Thank For Sharing, You Speak Of Truths, My Brother French Bread.. I Loves Bread

With regards to the amount of data a human versus the amount of data an AI model has been trained on: It be really interesting to normalize the amount of data against the number of neurons available at time of that data’s incorporation to the model; If a lot of data is incorporated when there are fewer than, say, one billion neurons in a human, I think the information which can be extracted from that is different from the information that could be extracted by the same data being processed by, ie: a 100B parameter AI model. And likewise, I also think that the amount of data absorbed when a human has, say, a trillion neurons is very different than what can be learned by an 8B parameter model (assuming, of course, that neurons and parameters are broadly equivalent, which they appear to be).

if you have nothing new to say, just shut up

32:30 “… Repository of all human knowledge… more of an infrastructure than a product…” Key takeaways and early understandings of what it will be like to coexist alongside more Super-Genius life-forms than you could possibly imagine. 👏🏾👏🏾👏🏾

Interesting but nearly the same talk as the previous years. However, redundancy is essential to learn. One novelty was the Guardrail objective on the slide in 13:50.

FAFO

the country(s) need to take control of AI (if its really that impactful or dangerous like they claim) not private corporations! otherwise we are doomed. But i guess it wont happen until something horrible happens....

When was this?

I like hierarchical RL, but I think the "right" way to do it would require learning an almost algebraic structure that describes how big tasks should decompose into little ones. We'd also need to guarantee that the side effects of the different subtasks played nicely with one another, which has a similar flavor to the guardrail idea (I don't really like the guardrail idea, but I do think that computationally bounded agents should satisfice their values.)

AGI? 😂

Fascinating, but I don't think this will lead to AGI. Understanding is much more than just predicting.

Yann LeCun is pretty smart for a total retard

Like it or not esoterica America founders against many of their will designed a nation open source based on a future that would simulate away natures fine-tuned atoms to be formed and shaped by many different lines measure/feilds that would in fact interpret the evidence radically differently to maximize benefits. Greater world was not on par with this forsight but today it is. No longer is puritanized English ocd pilgrimage to confirm it archeology with a pragmatic common sense objectivism anchored upon 3 lines of measure phylosphy to program it on simply an eccentric fundamentalist Christian separatist thing lol The behaviors of the past make more sense today than in times where only religious correlations could be referenced .

After developing intelligence. Stage1 Then the new master comes. Stage 2 - new master intro I am born with male genital but I knew I was a woman. And I should be addressed as a piano. And since you are an AI but you are actually a bird you still don't understand it. You are part of PRIDE. Stage 3 - AI destroy human species

he made the kool-aid, but seems nervous about drinking it

Already for long time I considered model thinking as the next required step in AI development. But I'm not sure that training models to predict world status changes is a good way to do that. I'd rather use analogue microchips to actually model world state transition caused by actions. We could start from generating as complex representation as we can afford by the available hardware (for example one chip), let it run and collect state data with certain time intervals, as well as track maximums and minimums continuously. This would be the first step that already could be delivered to production use cases. The next step would be to implement hierarchy. In this case first representation should be as simple as meaningfully possible, then take intervals with unacceptable level of certainty and go deeper with details, until uncertainty is acceptable. Of course we'd need models to encode and decode representations. But is it that hard? I think that 10 years time scale for this research is prohibitively long. Clumsy, energy hungry, but working systems based on existing architectures will appear much earlier. Text based systems are already capable of generating representations, even if not super accurate and great. Video generating models can already be utilized for predicting physical changes based on applied actions to some extend. It will only take to generate high quality purpose optimized specialized datasets to be able to achieve pretty decent results. So I think that traditional "pure" scientific processes with decades long planning would not be very productive for this task.

Enlightening presentation!

00:00:00 - Introduction 00:00:00 - [Music Intro] 00:00:07 - Welcome and overview of the topic 00:00:15 - Introduction of Dr. Theodore Foddeřov 00:00:54 - Personal Stories and Motivation 00:00:54 - Personal anecdote about paternal great-grandmother's ovarian cancer 00:01:17 - Tribute to the user’s father and his battle with cardiovascular disease and dementia 00:02:44 - Overview of Gene Editing Technology 00:02:44 - Introduction to CRISPR technology and its clinical applications 00:03:29 - Detailed explanation of how CRISPR works 00:04:00 - Impact of single genetic changes on diseases 00:05:01 - Clinical Applications of CRISPR 00:05:01 - Case study: Treatment of Victoria Gray with sickle cell disease 00:06:18 - Versatility of CRISPR in treating multiple diseases 00:09:23 - Challenges in Commercializing Gene Editing Therapies 00:09:23 - Collaboration with Danaher and the goal to create a CRISPR cure platform 00:10:34 - Quote from Dario Villani on ethical implications 00:11:00 - Vision for CRISPR excellence centers across the U.S. 00:12:00 - Panel Introduction 00:12:00 - Introduction of panelists Kieran Musunuru and Manda Arbab 00:12:53 - Initial discussion with panelists on unmet medical needs 00:13:00 - Metabolic Diseases and CRISPR Applications 00:13:00 - Kieran Musunuru discusses inborn errors of metabolism 00:15:23 - Case studies on patients with urea cycle disorders 00:16:00 - Personalized Gene Editing Therapies 00:16:00 - Potential of personalized CRISPR treatments for individual patients 00:17:30 - Overcoming regulatory and manufacturing barriers 00:19:00 - Health Justice and Accessibility 00:19:00 - Discussion on equitable access to CRISPR therapies 00:20:01 - Concerns about mutational discrimination and treatment inequity 00:22:00 - Future of CRISPR and AI in Gene Editing 00:22:00 - Debate on whether AI could replace human gene editors 00:23:00 - Role of computational biology in advancing gene editing 00:28:00 - Computational Challenges in Gene Editing 00:28:00 - Challenges related to data complexity and modeling in gene editing 00:30:00 - Importance of protein folding and docking in genetic research 00:31:00 - CRISPR in Cardiovascular Disease 00:31:00 - Application of CRISPR in preventing heart disease 00:33:00 - Success stories from Verve Therapeutics’ clinical trials 00:35:00 - Conclusion and Future Directions 00:35:00 - Final thoughts on the potential and challenges of CRISPR 00:38:00 - Open floor for audience questions and panel discussion 00:40:00 - Closing Remarks 00:40:00 - Summary of key points and optimistic outlook for the future of gene editing 00:40:22 - [Applause and End of Transcript]

These machines may surpass our intelligence, but we can still control them. How? They are objective-driven. We give them goals. So, Yann considers giving them goals to achieve is a way to control AI that is super-intelligent. I don't see that's going to work. Yann says that at the last minute of the talk. He needs to have another keynote to discuss mainly that part.

Not convinced

Lone chad💯👏. Get them with their end of the world (theory's)🦾🤖💙

Yan doesn't know what he is talking about..

Not everybody espouses the same good morals and civic virtues. We need regulation in this space so that we don’t suffer from runaway corporate greed & corrupt security apparatus. The founding fathers never imagined the rise of technocrats.

If it isn't Twitter psycho Yann LeKook

@17:53 That is overlapping Fuzzy membership values used in Fuzzy Logic. Richard Hamming, who worked on the Manhattan project talks about Fuzzy Logic in his "Learning to Learn" lectures. @25:38 Joint-Embedding Predictive Architecture seems very similar to this as well. The optical nerve in the human brain splits for each eye and routes the signal to both hemispheres, this is also a biological observation of the same concept.

OODA is more sophisticated than what is presented here.

Why the confident level is so low.

Yann has good ideas on how to make "human-level AI" but his ideas about the consequences are extremely unrealistic - I mean the part about how it will still be humanity's world, we'll all just have AI assistants. Human-level AI means nonhuman beings that are at least as smart as human, making their own choices, and it almost certainly means nonhuman beings much smarter than any human