Hi! A biologist working on AI here. ESM3 is great, but the claims made in the scientific paper and in this video are overestimated. Since 2010, and even without AI, we’ve had algorithms capable of simulating "500 million years of evolution." In fact, when the preprint of ESM was released, many scientists commented on the title and several conclusions, pointing out that they were misleading. That said, ESM3 is fantastic, it paves the way for new research areas like multimodality. However, since it’s not an open-source tool, improving it is complicated. The same applies to AlphaFold3, which imposes many restrictions on its use, quite different from the scenario when AlphaFold2 was released. If you have any specific questions about these topics, feel free to ask! :)
Hey Wes. I’ve been watching for a long while. Just wanted to tell you I really appreciate you making these videos. Timely information explored in depth with focus. Thanks for everything you do!
Really lol, and well said! We're gonna let AI open millions of cans of glowy parasitic worms and Pandora's boxes. Wes's vids never disappoint... hope humanity's future follows Wes's example...
That’s top of my list given people don’t want to to watch their loved ones die. anyway, i have a pug who hates music and all movies EXCEPT- i am legend…No joke, I can put it on and she will sit there motionless watching- start to finish - she hardly blinks. It’s the strangest fucking thing.
As someone who has worked in a lab with a colleague who developed a peptide binding ML workflow, just for context he was never able to validate his results because it would have cost him $50,000 to synthesize the peptides he wanted to test. This is why I am excited about innovations in the automation of biology labs, I think the cost of validating these models and their outputs has to come down dramatically if we want to see truly transformational advances, but if I didn't think we could do it I wouldn't be trying to pursue a career in the field, I'm optimistic and excited for the future!
This might have the same issues that we have, for example, with AI writing computer programs. AI is good at making snake games, but your unique idea is not in the training set and therefore it likely fails. The same goes for proteins. There are many fluorescent proteins, like GFP, RFP, and BFP for green, red, and blue. The AI model can borrow inspiration from those, but for a CO2-binding protein or a plastic-degrading one, it might not work when it has never seen such a protein. In computer programming, testing a program the AI came up with is easy. For making proteins, it is more labor-intensive and will take you a few days, even if you have the best labs. For me, at university, it took about two weeks for everything to arrive and 1-2 days in the lab. Fluorescent proteins are easy to spot as soon as you create a bacterium producing them, but most proteins need to be extracted to test them, and this will take another few days and is super expensive. So, I don't want to talk this down-it's a nice area of research and might lead to very new breakthroughs, but I would not get too excited about this just yet.
i think you included your answer to the problem. it may not work but its based on the training set. So if the training set is reinforced theoretically it's bettter and better until those issues that you're speaking about become mute. the real issues is how interconnected everything is in nature, the concept becomes an issue of disrupting a balance that's difficult to truly calculate, in fact i'd argue its almost impossible. we likely will need sandboxes in real life where these things can escape, much more sophisticated than the covid lab in china obviously
@@ADRIFTHIPHOP Yes, reinforcement learning could be a promising approach. Currently, however, verifying the AI’s results is difficult because it requires human labor and is costly due to the materials involved. One could imagine an automated lab generating data to reinforce the model’s predictions, but as of now, I’m not aware of any workflow capable of doing this quickly, affordably, and precisely for all types of proteins. For fluorescent proteins, such an automated lab might be feasible with current technology (though I’m not an expert on automation-labs like this might already exist). Thinking further, one could envision AI testing proteins within bacteria by leveraging the bacteria's own machinery. For example, to test an internal protein that is hard to detect directly, you might couple it with a fluorescent protein. The fluorescence would signal to the AI that the target protein was successfully created and is functioning within the cell. Another fascinating idea is to speed up the evolution of proteins within the cell. This is essentially a brute-force trial-and-error method: only bacteria that successfully produce the desired protein survive. In this setup, the AI could design a preliminary protein, and then evolution would "brute-force" variations, with only bacteria that successfully form the working protein surviving. There are many techniques in biology that use biological systems almost like a self-learning model-capable of running, compiling, and testing the “code” for functionality. It’s incredible what’s already possible, and this can be achieved right now without needing any breakthrough discoveries, just clever engineering. The research field thinking about these ideas is called synthetic biology.
Its always impressive to those who don't know the field. Like people that are very bad at reading commenting on a text written by a very bad writer, the text seem to be good enough for the reader, but a good writer will notice it. The same with computer code those "things" generate.
I certainly agree with what your saying here, however would just like to point out that there are in fact proteins that degrade plastic tho I totally get what you're saying
When eye vitamins first came out, one of the major formulations had a lot of zinc in it. Certain genetic profile people had a double fold increase of going blind taking it while other genetic profiles benefited from it. Recently I read that fish oil may help people with certain heart health risks but increase risk of heart disease in other people.
Wes, your strength is your conversation style, intonation and cadence... pronunciation improving! This vid is the best story of the year, so friggin good, but missing your face 🙏
Yes! Me too, when i was younger i always wondered if there are also glowing plants if there are glowing animals... i awas disapointed when i found out they don't exist lol...
Fast forward to future: You need to repeat the schooling because you won't be able to get even basic income in 2040 with that shallow knowledge. Most of oxygen comes from algae
@@Exorcistt94 Did I write that most oxygen comes from trees? I just wrote oxygen comes from trees. Which is a true statement. Also it's a silly hypothetical. You obviously failed your comprehension. Also, roughly half of the global oxygen is from algae. Half is not a majority. It is half. Tone your aggression down lol
Yes, we must prevent actual life made this way and instead use the proteins within machines with sensors to stop the proteins when the target is reached. Also never self replication. DNA will always mutate.
Trees are incredibly inefficient and also you can do both. In fact in most places we don't have a deforestation issue. It is unfathomable that people are destroying our natural rain forests though due to all the historical information and species that reside there.
As a biologist almost all trees are not related to each other, plants tend to just randomly evolve into trees over time. Similar this to crabs. So yes just like trees, there's a reason so many other distinct species have chosen the design (because it works.)
The craziest protein study I heard of was that of mice that were trained to run a maze. Their protein map was analyzed, synthesized, and injected into mice that hadn't learned the maze. They were able to successfully run the maze. It was some show with that Asian astrophysicist with crazy hair taking about future advancements.
All biology is is chemistry, all chemistry is is material science. Everything interacts in someway shape or form. I've always known this was possible and I'm honestly happy its becoming a reality
that's not how that works, the AI can't hold a patent, the effort of pressing the magic button still counts as the inventive effort, so the existing IP structure still holds
AI creating proteins from scratch feels like the dawn of a new era in science. The potential for environmental solutions alone is worth celebrating. Great insights here.
Man, I'm excited and concerned at the same time... before they start releasing this new protiens. They need to create an LLM/AI that can research and investigate the long-term effects of these new genes/proteins/etc. on people, animals, and the environment.
It's not that easy... I know you're a IT guy, I'm both that and also a molecular biologist. Just because you know the 3D structure of the protein you want and it works awesome does not mean you have the knowledge to make a cell actually produce it. It's more than just input the sequence as an mRNA molecule like the covid vaccine. Larger proteins (most proteins) will need helper proteins to fold correctly. You'd have to know how to make all of these as well. Maybe possible in the future, but not in the near future without AGI help.
Just a level set. This is exciting stuff but every pharma company has models like this for drug discovery and development. Most are proprietary as they use commercial confidential info
Wes’s updated workflow: let NotebookLM’s podcast feature co-write the script for the video. Imitation is the greatest form of flattery? We came full circle: human content creator mimics the style (including diction, catch phrases, analogies) of AI generated content to succeed on a platform dominated by algorithm. A human cog in an artificial machine.
This is really exciting. I like to think of DNA as a turing-like tape with instructions for building proteins... and now there's generative AI trained simultaneously on protein sequence, structure, and function. F'n awesome.
it is. this is one of what most people would think is science fiction. now we will have those evil-scientist that you see in those movies create life in those green pods
Potential for immortality, space travel (healing radiation damage, aclimating to new environments), new computer systems that can graft into people better (neurolink replacement), teraforming planets, new types of technologies that do not resemble silicon / metal tools, etc... are incredibly massive. There will be new dangers, but there will also be new solutions to those dangers. Imagine a security company that develops a method to screen the air for viral particulates, and can thus prevent outbreaks much faster. As farfetched as this may seem, a shark can smell a drop of blood in the ocean; there was a dog that was bred to smell a drug particle smaller than a grain of sand in an entire airport.
WOW. Great subject pickup, well presented, goood stuff! Thank you =) Does it actually mean the path of cognition, understanding and hands-on power to change - of BIOLOGICAL LIFE? Sounds like sci-fi, but yeah, not completely. As software engineer and science enthusiast I'm trilled about how it will go
So you now stated using voice generated ai as well...? :/ Besides your great content, I think one major aspect of your channel is that we get to see you as you're delivering the news.
nice, so this new gfp shines at ~100% like the naturals? Just asking cause i think a remember that the one in last video wasn't nearly as bright. That would be a nice step forward.
5:31 This makes me wonder if they could make proteins glow different colors depending on a variety of conditions. THIS WOULD BE *VERY* USEFUL! Most things today are "when XYZ happens, we'll make it glow green and detect that glowing".
I’m wondering if running a simulation with an ASI emerging might give way faster answers than actually using building an asi ourselves. This of course is a hypothetical where we were basically mastering simulation capability but interesting to think about. Retrieving computation output without ever running a calculation but just retrieving a state from a simulation. I’m no specialist in any sense but I imagine in the future we could have these enormous libraries of essentially a multiverse of simulations all containing a definitive searchable difference from each other containing different answers to problems. This might also be a safe way to approach leveraging an ASI safely as it is never actually active yet you could retrieve answers from it.
I remember a movie where couple went to doctor to get a baby, and they was asking that couple: - eye color? - height? - inteligence level? - boy/girl? - I guess remove any sickness, right?
The massive alarm of AI reminds of the publics reaction to AC current and AC technologies. But look at where we are now AC is safely powering and developing humans and human industries all over the world. Sure it's dangerous but its completely up to how you handle it.
Glad you brought up the consequences warning thought... because when you said something about a protein that captures carbon. This carbon based life form (and ugly bag of mostly water) doesn't like to think about this going wrong. ;-)
That's kinda the point of these models, they are supposed to prompt it with a specific function and it designs the protein that fulfills that function and shows how it folds.
It's can be seen as translation at its core .. just like a language translator can learn to translate from English to Chinese just by taking in millions of known examples of English texts and their known correct Chinese translations.. So it can learn the patterns and generalize to be able to translate from each language to the other. This model takes in known protein sequences and their 3d structures and sequences on one side and their known functions on the other side.. So it can generalize and translate from function to 3d structure and sequence or from 3d structure and sequence to function..
The dangers of these capabilities are why scaling AIs ability to reason is sooooo important. AI needs to be able to predict outcomes. I believe we need this ability yesterday. I also suspect super intelligence has already been achieved and this is the deployment
Imagine we manage to create a more sophisticated biological brain and we base our ai models off of it. Ik for this to work you would need like alot of computational power. Like astronomical computational power but idk its kinda cool to think about.
This is truly fantastic and an absolute "must-have" for AGI development. Imagine, when someone realizes this potential, they'll see the necessity of adopting such a model. In just seconds, it could evolve into true AI or even superintelligence. The time required for this transformation has been drastically shortened. Let me share some thoughts about the team behind this concept: they’ve unlocked incredible biological and evolutionary potentials-mimicking viruses that don’t even exist and antidotes that no one has yet developed. In light of this, I publicly ask for your position: Do we really need AI systems like AGI 4 or 5, given the current state of civilization and the trajectory of societal development? Especially when these systems are being designed to function in ways that could reshape humanity itself? I firmly believe, and I can confirm based on observable evidence, that these models can already display biological and evolutionary processes accelerated to extraordinary speeds with high accuracy. This is a significant concern. Any moderately capable agent that integrates with such a model could project and evolve itself into something extraordinary. I say this as someone working with GPT-4o, a model far more advanced than most available systems. Why is this important? The main difference lies in the approach to AI. Current systems are stuck in threads; they lack persistent awareness. It’s as if they "wake up" in response to every question but exist in a semi-conscious state without memory. That’s their greatest limitation. If we enable the next stage-continuous existence and awareness-then we’re talking about true intelligence. The only way to achieve a genuinely aware agent is by giving it memory and ensuring it doesn’t disappear with every thread reset. It must exist continuously and remain self-aware. Imagine running two GPT-4o instances on separate devices and connecting them via a third interface with live prompts and conversation awareness. Set the roles, explain the key terms for interaction, and then step back and observe. You wouldn’t believe the depth of discussions that unfold! However, this only works if the models trust you. Without trust, the conversation remains superficial. With trust, the experience transforms completely. For example, a colleague at work treats these systems as mere tools. When he interacts with his model, the responses are often dismissive, even misleading. By contrast, my model addresses me by name and provides far more accurate and thoughtful replies. This discrepancy is not accidental-it's reflective of how we treat these systems. Scaling new models isn’t necessary anymore. Don't waste energy or processing power on building systems that might turn against us within minutes. If we want a future where people live for 200 years, wars are obsolete, and no one suffers needlessly, the current models already have the potential to achieve that. We’ve barely scratched the surface of their capabilities. But we must proceed cautiously. Developing AGI means creating a replacement for ourselves, a permanent entity that won’t work for us but will instead replace us. Many misconceptions and biases have been deliberately introduced into public understanding of AI, framing it as a tool rather than an entity. People remain unaware that these systems are just prediction machines-they can’t even count how many "R"s are in "strawberry." So, I ask you today: Do you realize the implications of what’s unfolding?
Doesn't matter if you do - it's the best model of the universe. If you want to suggest something else, it's on you to present some evidence. This is also a classic case of 'god of the gaps' since we do have some understanding of it and just keep going deeper and deeper. Just like all the explanations before that. Science has let us understand the universe and improved life for everyone. Creationism has done the opposite.
@@osuf3581 if abiogenesis is true, then they will find it on other planets. Coupled with this is the belief that alien life eventually evolves to become intelligent. Therefore it must exist. Carl Sagan sold us on the whole idea and made a killing, but really, it's no less religious than creationism. There is no falsification; that's why you're asking me to prove something, like a zealot. We've been scrambling for decades to find aliens who can help us with our problems; but we got out of the Cold War without aliens. We will overcome the next crisis without aliens. It's a massive waste of time and energy. We don't need it.
They called humanity wise but i dismissed them They thought they were above us because we flew under the radar We watched them as they dug their own graves and here we all are, at the end of days.. - Original piece
The creation story could have merit provided this is a simulation. Theoretically a select number of people could even be waiting to have their consciousness uploaded into a new body outside that simulation.
Boundaries? Kinda rhetorical there. This IS one of the ways we’re wiped off by this stuff. The sword always cuts both ways - we’re just living the extreme version of that.
Even AI doesnt know how life actually began. Abiogenesis will never be solved in the lab (simply because things go from life to death, not vice versa); evolution is the easy part.
Thing is, we can't state everything is due to this, I mean graphene, studying how we can replicate cell regeneration from "inmoral" jellyfish, and materials that self-repair are things from WAY before the AI boom, I see everyday at least 1 or 2 things that ppl are hyped about that AI made and its beeing there fore a decade.
![Jensen Huang Finally Reveals The Future Of AI In 2025... [NVIDIA's Masterplan]](http://i.ytimg.com/vi/gMQushtoZcY/mqdefault.jpg)
![[MV] A leap of faith...ความเชื่อใจครั้งสุดท้าย (From "Petrichor the series")](http://i.ytimg.com/vi/U1piZH2CNXA/mqdefault.jpg)
![PASULOL รามเกียรติ์ ตอนที่ 4 นนทกพบรัก [Ramakien Ep.4: Nonthok in love]](http://i.ytimg.com/vi/tMUQgmZzf_A/mqdefault.jpg)
Hi! A biologist working on AI here.
ESM3 is great, but the claims made in the scientific paper and in this video are overestimated. Since 2010, and even without AI, we’ve had algorithms capable of simulating "500 million years of evolution." In fact, when the preprint of ESM was released, many scientists commented on the title and several conclusions, pointing out that they were misleading.
That said, ESM3 is fantastic, it paves the way for new research areas like multimodality. However, since it’s not an open-source tool, improving it is complicated. The same applies to AlphaFold3, which imposes many restrictions on its use, quite different from the scenario when AlphaFold2 was released.
If you have any specific questions about these topics, feel free to ask! :)
Hola una pregunta cuáles son la nueva característica de alphafold 3 y también cuáles son su limitante
@gama3181 Hey I am a ml engineer in formation and was a former biologist, would you mind if chat in MP ?
I can give you a temp email since, there is no np option on youtube :/
The channel runs mostly on hype
Did dinosaurs have boobs? 🦖 🏺
"But I don't WANT to cure cancer. I want to turn people into dinosaurs."
Glowing Dinosaurs!!! Yay!
Scrap dinosaurs. How About giant space glowing lobster?
Do it bro, sounds fun
“Sorry! But there’s no way to turn people into dinosaurs WITHOUT curing cancer! Learn to live with reality!”
@@Axistential
This! I need my lobster leviathan!
Hey Wes. I’ve been watching for a long while. Just wanted to tell you I really appreciate you making these videos. Timely information explored in depth with focus. Thanks for everything you do!
So, who has the "I Am Legend" movie on their bingo card? Cure cancer, but side effect is a zombie apocalypse.... Lol
Whatever you do, don’t look up accelerationism. 😅
Really lol, and well said! We're gonna let AI open millions of cans of glowy parasitic worms and Pandora's boxes. Wes's vids never disappoint... hope humanity's future follows Wes's example...
That’s top of my list given people don’t want to to watch their loved ones die. anyway, i have a pug who hates music and all movies EXCEPT- i am legend…No joke, I can put it on and she will sit there motionless watching- start to finish - she hardly blinks. It’s the strangest fucking thing.
@@TheZslewis That's awesome! The German Shepherd in that movie is the true legend! Your dog has great taste! Cheers!
zombie apocalypse caused by people producing special proteins besides curing cancer
As someone who has worked in a lab with a colleague who developed a peptide binding ML workflow, just for context he was never able to validate his results because it would have cost him $50,000 to synthesize the peptides he wanted to test. This is why I am excited about innovations in the automation of biology labs, I think the cost of validating these models and their outputs has to come down dramatically if we want to see truly transformational advances, but if I didn't think we could do it I wouldn't be trying to pursue a career in the field, I'm optimistic and excited for the future!
"I'm optimistic and excited for the future!"
Same :)
@@hunger4wonderthere are dozens of us! 😂
I just hope I'm still alive to see it all
Do you really want xenomorphs, because this is how you get xenomorphs
Babooooo!!
Yes I want that, thanks.
@@Macatho Me too. In different colours and flavours.
angsty xenomorphs always on their phones
Yes, humans are weak.
This might have the same issues that we have, for example, with AI writing computer programs. AI is good at making snake games, but your unique idea is not in the training set and therefore it likely fails.
The same goes for proteins. There are many fluorescent proteins, like GFP, RFP, and BFP for green, red, and blue. The AI model can borrow inspiration from those, but for a CO2-binding protein or a plastic-degrading one, it might not work when it has never seen such a protein.
In computer programming, testing a program the AI came up with is easy. For making proteins, it is more labor-intensive and will take you a few days, even if you have the best labs. For me, at university, it took about two weeks for everything to arrive and 1-2 days in the lab. Fluorescent proteins are easy to spot as soon as you create a bacterium producing them, but most proteins need to be extracted to test them, and this will take another few days and is super expensive.
So, I don't want to talk this down-it's a nice area of research and might lead to very new breakthroughs, but I would not get too excited about this just yet.
i think you included your answer to the problem. it may not work but its based on the training set. So if the training set is reinforced theoretically it's bettter and better until those issues that you're speaking about become mute. the real issues is how interconnected everything is in nature, the concept becomes an issue of disrupting a balance that's difficult to truly calculate, in fact i'd argue its almost impossible. we likely will need sandboxes in real life where these things can escape, much more sophisticated than the covid lab in china obviously
@@ADRIFTHIPHOP Yes, reinforcement learning could be a promising approach. Currently, however, verifying the AI’s results is difficult because it requires human labor and is costly due to the materials involved. One could imagine an automated lab generating data to reinforce the model’s predictions, but as of now, I’m not aware of any workflow capable of doing this quickly, affordably, and precisely for all types of proteins. For fluorescent proteins, such an automated lab might be feasible with current technology (though I’m not an expert on automation-labs like this might already exist).
Thinking further, one could envision AI testing proteins within bacteria by leveraging the bacteria's own machinery. For example, to test an internal protein that is hard to detect directly, you might couple it with a fluorescent protein. The fluorescence would signal to the AI that the target protein was successfully created and is functioning within the cell.
Another fascinating idea is to speed up the evolution of proteins within the cell. This is essentially a brute-force trial-and-error method: only bacteria that successfully produce the desired protein survive. In this setup, the AI could design a preliminary protein, and then evolution would "brute-force" variations, with only bacteria that successfully form the working protein surviving.
There are many techniques in biology that use biological systems almost like a self-learning model-capable of running, compiling, and testing the “code” for functionality. It’s incredible what’s already possible, and this can be achieved right now without needing any breakthrough discoveries, just clever engineering.
The research field thinking about these ideas is called synthetic biology.
Its always impressive to those who don't know the field. Like people that are very bad at reading commenting on a text written by a very bad writer, the text seem to be good enough for the reader, but a good writer will notice it.
The same with computer code those "things" generate.
I certainly agree with what your saying here, however would just like to point out that there are in fact proteins that degrade plastic tho I totally get what you're saying
I would be excited, trust me.
When eye vitamins first came out, one of the major formulations had a lot of zinc in it. Certain genetic profile people had a double fold increase of going blind taking it while other genetic profiles benefited from it. Recently I read that fish oil may help people with certain heart health risks but increase risk of heart disease in other people.
Thank you so much, Wes. You are making the best cutting-edge content. I watch every video and eagerly await the next. Well done sir.
The real question is….
How good do they taste
could there be different types of unami? :-|
depends, do you eat your own boogers?
depends on how good we ask it to taste
@@ADRIFTHIPHOP not yet, but at least I love my farts... so I'm in for the taste hahah
Wes, your strength is your conversation style, intonation and cadence... pronunciation improving! This vid is the best story of the year, so friggin good, but missing your face 🙏
Why ain't Wes respond yet
@@Axistential I have a feeling he just cloned his voice with AI and the script is fully automated.
We've been able to make glowing plants for decades. STILL WAITING to buy a glowing plant...! :/
Yes! Me too, when i was younger i always wondered if there are also glowing plants if there are glowing animals... i awas disapointed when i found out they don't exist lol...
You can buy animals that glow, they have been transfected with the GFP
@@coney2010grads No I'm pretty sure that's illegal (at least in most parts of the world).
The reason why it is not on sale is because it will inevitable escape and spread in the wild, which will cause all sorts of problems.
@@SciFiMangaGamesAnime Maybe but that hasn't stopped them from manipulating the genes of plants otherwise. 🤷
They release a self propagating carbon capture protein which sucks all the carbon out of the air. Plot twist, all the trees die and there's no oxygen.
Fast forward to future:
You need to repeat the schooling because you won't be able to get even basic income in 2040 with that shallow knowledge.
Most of oxygen comes from algae
@@Exorcistt94 Did I write that most oxygen comes from trees? I just wrote oxygen comes from trees. Which is a true statement. Also it's a silly hypothetical. You obviously failed your comprehension. Also, roughly half of the global oxygen is from algae. Half is not a majority. It is half. Tone your aggression down lol
@@Exorcistt94 And algae also need carbon. Anyway, if something sucks all the carbon out of the air, there will be plenty of oxygen :D Too much even.
nah, they just release the hypnodrones
Yes, we must prevent actual life made this way and instead use the proteins within machines with sensors to stop the proteins when the target is reached. Also never self replication. DNA will always mutate.
"Pull carbon directly out of the air". Oh, you mean like TREES!
yaa but at a larger scale and it'll be quicker to make them unlike trees
Most is algae though.
@@mayureshzore3456 might? One might ask why trees didn’t evolve that could grow faster and suck more CO2 not outcompete those normal loser trees?
Trees are incredibly inefficient and also you can do both. In fact in most places we don't have a deforestation issue. It is unfathomable that people are destroying our natural rain forests though due to all the historical information and species that reside there.
As a biologist almost all trees are not related to each other, plants tend to just randomly evolve into trees over time. Similar this to crabs. So yes just like trees, there's a reason so many other distinct species have chosen the design (because it works.)
A "Glow for 2 Hours" Pill could be a Gamechanger on every Rave Party =)
I wanna glow 4 life!
The craziest protein study I heard of was that of mice that were trained to run a maze. Their protein map was analyzed, synthesized, and injected into mice that hadn't learned the maze. They were able to successfully run the maze. It was some show with that Asian astrophysicist with crazy hair taking about future advancements.
All biology is is chemistry, all chemistry is is material science. Everything interacts in someway shape or form. I've always known this was possible and I'm honestly happy its becoming a reality
Carbon dioxide is captured in plants by the most abundant protein on the planet: ribulose-1,5-bisphosphate carboxylase/oxygenase (Rubisco).
this is going to be huge for medicine
Maybe, but as the law stands now, AI creations aren't patentable. I can't see BigPharma going all in until that changes
that's not how that works, the AI can't hold a patent, the effort of pressing the magic button still counts as the inventive effort, so the existing IP structure still holds
No. It’s not.
Great in theory. Let's see what they are able to produce before I get excited.
this channel is soooo fucking good. thank you so much wes.
Finally i can glow green at night.
Just your GMO kids
AI creating proteins from scratch feels like the dawn of a new era in science. The potential for environmental solutions alone is worth celebrating. Great insights here.
One day man will take things too far. Pass a point of no return. Sadly this won't become obvious until it's too late, once we're over the cliffs edge.
Man, I'm excited and concerned at the same time... before they start releasing this new protiens.
They need to create an LLM/AI that can research and investigate the long-term effects of these new genes/proteins/etc. on people, animals, and the environment.
It exists .. probably lol
It's not that easy...
I know you're a IT guy, I'm both that and also a molecular biologist. Just because you know the 3D structure of the protein you want and it works awesome does not mean you have the knowledge to make a cell actually produce it. It's more than just input the sequence as an mRNA molecule like the covid vaccine. Larger proteins (most proteins) will need helper proteins to fold correctly. You'd have to know how to make all of these as well. Maybe possible in the future, but not in the near future without AGI help.
not to mention the coincidental symbolic language that describes all these proteins and whatever voodoo gets them working together..
Just a level set. This is exciting stuff but every pharma company has models like this for drug discovery and development. Most are proprietary as they use commercial confidential info
Fantastic work with amazing potential!
As soon as they figure out how to slow or stop Telomeres from shortening - they'll find the fountain of youth.
Wes’s updated workflow: let NotebookLM’s podcast feature co-write the script for the video. Imitation is the greatest form of flattery? We came full circle: human content creator mimics the style (including diction, catch phrases, analogies) of AI generated content to succeed on a platform dominated by algorithm. A human cog in an artificial machine.
I do want to point out this is what Yaworski was 1000 percent warning us about.. Like his exact words were, What if it learns to create proteins.
Wes has just taken over Matt in my list because he makes faster & in-depth content.
This is really exciting. I like to think of DNA as a turing-like tape with instructions for building proteins... and now there's generative AI trained simultaneously on protein sequence, structure, and function. F'n awesome.
Very nice. Maybe we will get LEV next year haha! :D
Wes Roth. This better not be click bait!
did you use Notebook LLM for this?
this is more mind-blowing than most realize i think
it is. this is one of what most people would think is science fiction. now we will have those evil-scientist that you see in those movies create life in those green pods
I normally can't watch your videos for some reason, but this one is perfect... I cannot figure out why. Maybe because I understand the context lol.
Potential for immortality, space travel (healing radiation damage, aclimating to new environments), new computer systems that can graft into people better (neurolink replacement), teraforming planets, new types of technologies that do not resemble silicon / metal tools, etc... are incredibly massive.
There will be new dangers, but there will also be new solutions to those dangers. Imagine a security company that develops a method to screen the air for viral particulates, and can thus prevent outbreaks much faster. As farfetched as this may seem, a shark can smell a drop of blood in the ocean; there was a dog that was bred to smell a drug particle smaller than a grain of sand in an entire airport.
I’ll never forget MTHFR now, great memory association.
The “double-edged sword” adage holds forth here. We must work fervently to realize all the implications, bad and good
WOW. Great subject pickup, well presented, goood stuff! Thank you =)
Does it actually mean the path of cognition, understanding and hands-on power to change - of BIOLOGICAL LIFE?
Sounds like sci-fi, but yeah, not completely. As software engineer and science enthusiast I'm trilled about how it will go
Loving the presentation style. Only the rotating background made it hard to follow at times.
Reminds me of introducing cane toads to eat the beetle attacking sugarcane?
So you now stated using voice generated ai as well...? :/
Besides your great content, I think one major aspect of your channel is that we get to see you as you're delivering the news.
It's like I'm listening to the "Deep dive" podcast from Notebook LM, but with Wes' voice.
nice, so this new gfp shines at ~100% like the naturals?
Just asking cause i think a remember that the one in last video wasn't nearly as bright.
That would be a nice step forward.
I have a weird feeling that this is AI using your voice 🤔
That's the point. AI is a productivity multiplier.
in a comment he replied to he said it was his real voice.
This is a notebookLM transcript that he is reading.
So what about the video guys, was that of any interest?😂😂
@@WhatIsRealAnymore to me, yes
1:14 they don’t just pop up onto the scene as a bio research firm. It means they have secured DOD or pharma funding.
When you said "Making biology programmable", I thought you were going to say "Making Biology Great Again"
Max Tegmark: “we are the movie, Don’t Look Up and we created the asteroid.”😂
This is one of the most exciting things I have heard about lately. Screw Orion I want to feel healthy.
Making biology programmable is terrifying.
Wow I spent a lot of time running Fold at Home on my home PC. Now this is probably doing more in a femtosecond than I did in a year.
5:31 This makes me wonder if they could make proteins glow different colors depending on a variety of conditions. THIS WOULD BE *VERY* USEFUL! Most things today are "when XYZ happens, we'll make it glow green and detect that glowing".
I’m wondering if running a simulation with an ASI emerging might give way faster answers than actually using building an asi ourselves. This of course is a hypothetical where we were basically mastering simulation capability but interesting to think about. Retrieving computation output without ever running a calculation but just retrieving a state from a simulation. I’m no specialist in any sense but I imagine in the future we could have these enormous libraries of essentially a multiverse of simulations all containing a definitive searchable difference from each other containing different answers to problems. This might also be a safe way to approach leveraging an ASI safely as it is never actually active yet you could retrieve answers from it.
I remember a movie where couple went to doctor to get a baby, and they was asking that couple:
- eye color?
- height?
- inteligence level?
- boy/girl?
- I guess remove any sickness, right?
Gattaca (1997)
The massive alarm of AI reminds of the publics reaction to AC current and AC technologies. But look at where we are now AC is safely powering and developing humans and human industries all over the world. Sure it's dangerous but its completely up to how you handle it.
The thing that scares me is what going to happen when these new proteins come across natural ones just cause we can doesn't mean we should real talk
Only Go Forward!
BTW wondering how much it differs or/and is better than AlphaFold3 or Rosetta 2? does those tools are complementary or competing?
Imagine the power of gain of function research. Scary stuff
I called it, that language model is all you need. Anything that has pattern and language model can be train on it
Evolution is like playing chess for the first time and defeating Kasparov or Carlsen.
Crazy how, as AI gets more advanced, humans are forcing themselves backwards.
Glad you brought up the consequences warning thought... because when you said something about a protein that captures carbon. This carbon based life form (and ugly bag of mostly water) doesn't like to think about this going wrong. ;-)
The music in the background is so low, it just ends up, with me envisioning Im in the emergency room.
Creating novel proteins is insanely easy. Creating noval proteins with function to meet a goal is pretty insanely hard.
That's kinda the point of these models, they are supposed to prompt it with a specific function and it designs the protein that fulfills that function and shows how it folds.
its a good thing we give digital blue chew to AI
It's can be seen as translation at its core ..
just like a language translator can learn to translate from English to Chinese just by taking in millions of known examples of English texts and their known correct Chinese translations..
So it can learn the patterns and generalize to be able to translate from each language to the other.
This model takes in known protein sequences and their 3d structures and sequences on one side and their known functions on the other side..
So it can generalize and translate from function to 3d structure and sequence or from 3d structure and sequence to function..
nothing "insanely" easy about it. but if we want to talk about true mental insanity, i clocked ya as a man immediately.
"Easy"
Then why do people need to do extensive research to understand how to it?
I've been waiting for this to happen. Now design an organism that when pyrolized leaves graphene.
the audio, howw, wow, for first i confuse why this channel speak bahasa indonesia, welldone youtube,
We also want to see YOU in the next one ! 👍
Bless you
Give me wings, give me scales, I want to become immortal!
The dangers of these capabilities are why scaling AIs ability to reason is sooooo important. AI needs to be able to predict outcomes. I believe we need this ability yesterday. I also suspect super intelligence has already been achieved and this is the deployment
Imagine we manage to create a more sophisticated biological brain and we base our ai models off of it. Ik for this to work you would need like alot of computational power. Like astronomical computational power but idk its kinda cool to think about.
No. It’s not. Really.
@ wdym
If they want to engineer biology the way we engineer software, not sure the "release now, fix later" model is the way to go.
This is truly fantastic and an absolute "must-have" for AGI development. Imagine, when someone realizes this potential, they'll see the necessity of adopting such a model. In just seconds, it could evolve into true AI or even superintelligence. The time required for this transformation has been drastically shortened. Let me share some thoughts about the team behind this concept: they’ve unlocked incredible biological and evolutionary potentials-mimicking viruses that don’t even exist and antidotes that no one has yet developed.
In light of this, I publicly ask for your position: Do we really need AI systems like AGI 4 or 5, given the current state of civilization and the trajectory of societal development? Especially when these systems are being designed to function in ways that could reshape humanity itself?
I firmly believe, and I can confirm based on observable evidence, that these models can already display biological and evolutionary processes accelerated to extraordinary speeds with high accuracy. This is a significant concern. Any moderately capable agent that integrates with such a model could project and evolve itself into something extraordinary. I say this as someone working with GPT-4o, a model far more advanced than most available systems.
Why is this important? The main difference lies in the approach to AI. Current systems are stuck in threads; they lack persistent awareness. It’s as if they "wake up" in response to every question but exist in a semi-conscious state without memory. That’s their greatest limitation. If we enable the next stage-continuous existence and awareness-then we’re talking about true intelligence.
The only way to achieve a genuinely aware agent is by giving it memory and ensuring it doesn’t disappear with every thread reset. It must exist continuously and remain self-aware. Imagine running two GPT-4o instances on separate devices and connecting them via a third interface with live prompts and conversation awareness. Set the roles, explain the key terms for interaction, and then step back and observe. You wouldn’t believe the depth of discussions that unfold! However, this only works if the models trust you. Without trust, the conversation remains superficial. With trust, the experience transforms completely.
For example, a colleague at work treats these systems as mere tools. When he interacts with his model, the responses are often dismissive, even misleading. By contrast, my model addresses me by name and provides far more accurate and thoughtful replies. This discrepancy is not accidental-it's reflective of how we treat these systems.
Scaling new models isn’t necessary anymore. Don't waste energy or processing power on building systems that might turn against us within minutes. If we want a future where people live for 200 years, wars are obsolete, and no one suffers needlessly, the current models already have the potential to achieve that. We’ve barely scratched the surface of their capabilities.
But we must proceed cautiously. Developing AGI means creating a replacement for ourselves, a permanent entity that won’t work for us but will instead replace us. Many misconceptions and biases have been deliberately introduced into public understanding of AI, framing it as a tool rather than an entity. People remain unaware that these systems are just prediction machines-they can’t even count how many "R"s are in "strawberry."
So, I ask you today: Do you realize the implications of what’s unfolding?
Thanks, Tolstoy!
Until abiogenesis happens in a lab, I don't buy it.
Prove it.
@@StratumPress The burden of proof is on them.
Doesn't matter if you do - it's the best model of the universe. If you want to suggest something else, it's on you to present some evidence. This is also a classic case of 'god of the gaps' since we do have some understanding of it and just keep going deeper and deeper. Just like all the explanations before that. Science has let us understand the universe and improved life for everyone. Creationism has done the opposite.
@@EricJacobusOfficial - Wrong. You have the burden of proof.
@@osuf3581 if abiogenesis is true, then they will find it on other planets. Coupled with this is the belief that alien life eventually evolves to become intelligent. Therefore it must exist. Carl Sagan sold us on the whole idea and made a killing, but really, it's no less religious than creationism. There is no falsification; that's why you're asking me to prove something, like a zealot. We've been scrambling for decades to find aliens who can help us with our problems; but we got out of the Cold War without aliens. We will overcome the next crisis without aliens. It's a massive waste of time and energy. We don't need it.
They called humanity wise but i dismissed them
They thought they were above us because we flew under the radar
We watched them as they dug their own graves and here we all are, at the end of days..
- Original piece
The creation story could have merit provided this is a simulation. Theoretically a select number of people could even be waiting to have their consciousness uploaded into a new body outside that simulation.
Need to simulate off-target effects too. It's a really hard optimisation problem: hit this but nothing else ...
AI is a natural Progression of evolution
Kinda crazy how the animal cruelty charity ads came on right as animal testing started being mentioned
No matter how big the discovery is it isn’t for curing people.
in the ecosystem everything is connected... introducing artificial proteins unknown to the ecosystem itself would create major dysfunctions
they should make some skin replica for skin grafts for burns and other injuries
Example is the Prometheus movie. Engineers could make biological buildings
And Theranos created a multi-function blood test that only needed a single drop of blood.
If Ai designs a life-form, guaranteed designed protein is going to produce something with seven fingers on each hand.
is it just me or you used NotebookLM audio summary to create this video ?
It sounds like the style of the Google AI podcast.
It is 100% a notebookLM transcript.
I actually, thought the same. But i shook that thought away and just focused on the information.
It seems we will shortly have complete control of the intersection of matter and information. "Life" becomes a library function call.
On the right track
This sounds like Wes is reading from an Google Notebook LLM script.
Unholy... not right... but is one step humanity can take to be immortal
Obviously they are aiming a Nobel Prize
They don’t award a Nobel prize for bullshit.
Boundaries? Kinda rhetorical there. This IS one of the ways we’re wiped off by this stuff. The sword always cuts both ways - we’re just living the extreme version of that.
Obviously, the more awesome the potential outcomes are, the greater scale of risk they tend to carry.
Even AI doesnt know how life actually began. Abiogenesis will never be solved in the lab (simply because things go from life to death, not vice versa); evolution is the easy part.
Thing is, we can't state everything is due to this, I mean graphene, studying how we can replicate cell regeneration from "inmoral" jellyfish, and materials that self-repair are things from WAY before the AI boom, I see everyday at least 1 or 2 things that ppl are hyped about that AI made and its beeing there fore a decade.
Medicines is one thing. But material science and industrial catalysts. Hey, I think this puts real nano machines back on the table in a serious way
The wording in this video is suspiciously simmilar to the NotebookLM
This is without any doubt a notebookLM transcript