The acceleration is happening right now. This isn't just sci-fi anymore. AI is already coming up with huge and completely new scientific discoveries,
and it's getting faster and faster. In the same week, we had not one but two papers published in Nature, which is like the most prestigious scientific
journal in the world. Both of them involve using AI to autonomously do scientific research. And
this isn't just theory. These AI agents literally came up with new treatments for cancer, blindness, antibiotic resistance, and other diseases that
are currently really difficult to treat. And these breakthroughs are completely new. AI was able to come up with a ton of discoveries that human
experts never thought of before. Most importantly, they actually work in real life. So, in this video,
we're going to go over these two papers and why they're such a big deal. Now, this is super technical, but as always, I'm going to simplify this to
make it easy to understand for everyone. Let's jump right in. All right, so the first paper is called accelerating scientific discovery with
co-scientist. And this is built by Google. This isn't just a regular AI chatbot. Co-scientist is fundamentally
different. And here's how it works. Instead of one huge model, it's actually an entire ecosystem of specialized AI agents working together. In fact,
you can think of it as a virtual lab. Here's how the agentic framework is organized. At the very top, you have the supervisor agent. This is the boss
and its job is to take the goal specified by the human scientist. Say a prompt like find a new way to
treat this specific type of leukemia. So this supervisor agent parses that goal and then allocates tasks to the underlying agents. It's important to
note that the supervisor is purely administrative. Its job is to ensure the system stays on task. The actual creative ideation begins with the
generation agent. This agent is tasked with brainstorming initial ideas. It goes out, explores the
scientific literature using web search to pull in the latest and most relevant data. Then it synthesizes existing findings into new proposed
directions. Afterwards, it generates the initial batch of hypotheses or ideas. Now, usually when you're brainstorming, especially in a complex field,
the first few ideas you throw on the whiteboard are usually well terrible or highly flawed. So, this brings
us to the next most critical component of the entire system, which is the reflection agent. Think of this agent as the most brutal, rigorous reviewer
you can possibly imagine. Its whole purpose is to try to destroy the hypotheses generated by the generation agent. Just tear them apart. It critically
examines every proposed idea looking for flaws. It fact checks to make sure everything is correct.
It also checks for novelty, ensuring the idea isn't just a slight rephrasing of an existing discovery, but is actually a new idea. So, the generation
agent comes up with ideas and the reflection agent tries to tear them apart. The friction between these two agents is what elevates the system to
produce really highquality ideas. And then working quietly in the background is the proximity agent. Its
job is to map all these ideas out into a highdimensional space to group similar ones together. So if the generation comes up with several slightly
different variations of pretty much the same thing, this proximity agent is able to detect this and then tell the system, hey, these are all
essentially the same idea. Stop wasting compute and resources and move on to a different concept. And the
process doesn't stop there. Co-scientist also uses an evolution agent. This agent takes the harsh critiques from the reflection agent, looks at
surviving ideas and then iteratively refineses them. For example, it might take two good surviving ideas and mash them together to create a brilliant
one. It tries to bridge logical gaps. Now, here's where it gets fascinating. How exactly does the system
decide which of these surviving ideas is the best one? This is the genius part. The system uses a ranking agent to host an actual automated tournament
of ideas. This is perhaps the most innovative aspect of the entire co-scientist architecture. The ranking agent uses an ELO rating system. Just like
in chess or in competitive video games where if a lowerranked player manages to beat a grandmaster,
they get a huge chunk of their points. But if two equally ranked players tie, their scores barely move. It's the exact same mathematical principle,
but applied to the rigorous evaluation of scientific ideas. And this co-scientist system actually simulates head-to-head debates between pairs of
ideas. So, hypothesis A argues why it's more new or plausible or testable than hypothesis B. And
hypothesis B tries to argue back, pointing out flaws in hypothesis A. And then we have a separate AI model that acts as a judge, evaluating the
debate. Winners gain ELO points, losers lose them. And over hundreds or even thousands of these automated simulated debates, the strongest and most
robust ideas naturally rise to the top of the leaderboard. Now, if you've been playing around with AI,
you'll probably know that large language models are notorious for hallucinating and sometimes just making stuff up while sounding very confident. So,
how do we know that co-scientist isn't just making up these ideas? How do we know it's actually producing better and legit scientific hypothesis?
Well, the researchers gave co-scientist 15 incredibly challenging unsolved biomedical goals written by
PhD level scientists and they asked it to come up with the best solutions. They then asked human experts to provide their absolute best guess
solutions as well. They also gave the same goals to other state-of-the-art AI models at that time and they asked independent human experts to judge
which ideas were better. Now this is a blind test. So the human experts couldn't see if the ideas were
generated by other humans or by which AI models and the results were remarkable. You can see that co-scientists after giving it enough time to think
and iterate eventually beat all the AI models and the human experts. These independent human judges rated the ideas from co-scientists as
significantly higher in novelty, plausibility, and impact compared to the best solutions from human experts. So
this AI system, it's not just making stuff up. This is proof that it can actually create new innovative ideas which even independent human experts
approve of. Now, generating solid ideas is one thing, but does this actually work in real life? That's the ultimate test. Well, next the researchers
did exactly that. They took the best outputs from co-scientist on some very different and incredibly
complex problems that are not solved yet and the results were astounding. Let's dive into each one. The first one involves acute myoid leukemia or AML
for short. This is a very tough disease and here's why it's so hard to cure. You see, acute myoid leukemia is a devastating and highly aggressive
blood cancer that starts in the bone marrow. Now, the major challenge with AML isn't necessarily
killing the cancer cells at the start. We actually have chemotherapies that can already do this quite well. The problem is that relapse is incredibly
common. And when the disease does come back, it's highly resistant to the therapy. And it comes back really aggressively because of a specific
subpopul of the cells which are called leukemia stem cells. These are deeply dormant cells that are like
the roots of a weed. So you can mow the lawn and kill the bulk of the cancer cells, but if you leave the roots, then the weeds grow back and often
stronger. So these stem cells hide in the bone marrow and they evade standard chemotherapy and they could eventually cause a relapse. There's
currently no good therapeutic solution for killing these dormant stem cells. Now, here's what the researchers
did. They handed co-scientist a massive data set of 2,300 drugs that are already FDA approved for human use. They're just currently used for entirely
different purposes. The prompt for the AI was essentially which of these 2,300 existing drugs could be repurposed to specifically fight acute mileoid
leukemia. Now, this is basically the entire field of drug repurposing. Designing a new drug from
scratch would almost take a decade and millions of dollars and the failure rate is super high. But if we can look at existing drugs and discover new
uses for them, then it's a lot more efficient. Well, co-scientist actually found several strong candidates. It identified drugs like binometanib,
pritinib, and surivatin. Let's take one example, binameanib. This is currently approved and used for skin
cancer, not leukemia. But they took the advice from co-scientist and tested benametanib on leukemia cells in the lab and it actually turned out to be
extremely effective. It had an IC50 of two nanomars against acute myoid leukemia. Let's quickly clarify what this IC50 means. It's kind of a gold
standard metric. So this is a measure of a drug's potency. Specifically, how much of a drug is needed to
inhibit a biological process by exactly half. In this case, the process is the survival of the cancer cells. A lower number means the drug is very
effective. You need very little of it to do massive damage. And in this case, two nanomars is an extremely low concentration. So this indicates a very
strong efficacy against those cancer cells. This shows that co-scientist isn't just a theoretical
framework anymore. They've already gotten it to find a new drug that works very well on a very problematic cancer that previously had no good
therapies. Now, this benametanib is previously known to work for skin cancer. So, maybe there's a connection there. But here's another discovery
that's even crazier. They told co-scientists, "Don't just give us the obvious repurposing candidates. Give us
some drugs that have no prior evidence linked to leukemia or cancer at all." They wanted completely new ideas that no human has ever published on this
subject. And lo and behold, after a lot of thinking and internal debating and ranking different ideas, co-scientists generated a hypothesis suggesting
a drug called cur 6. Now cur 6 is a fascinating molecule. It inhibits an enzyme called I1 alpha
which manages cell stress. So this decides how cells clear out misfolded or broken proteins. So co-scientist after reading and synthesizing stuff
across entirely different fields of biology hypothesized that targeting this specific stress pathway could selectively kill these leukemia cells.
Humans have never thought of this before. Now let's dig into this a bit more. Why did the AI suggest there?
Well, if you look into its hypothesis, the AI recognize that cancer cells by their very nature of fast uncontrolled division are also under huge
internal stress because they're growing so fast. Their protein manufacturing machinery is running in overdrive and constantly breaking down. So, they
rely heavily on this stress response pathway like I1 alpha just to survive their own chaos. It needs to
constantly clean up a lot of broken proteins just to stay alive. Otherwise, the cell would just be filled with garbage. Co-scientist deduced that if
you inhibit this specific stress pathway with a drug cur 6, this would mess up the cancer cells. Now, you might be wondering, what about normal cells?
Well, it turns out that a normal healthy cell can tolerate this drug because it's not stressed. It
doesn't really need to rely on this garbage pathway as often. But to the cancer cell, which is constantly dividing and under immense stress, this drug
is very potent. At least that's what the AI suggested. Of course, the next logical step is to test this in real life. So, the scientists applied this
Curac 6 drug to these leukemia cells. And the results were shocking. It wasn't just good, it was
remarkably effective. It turns out that Curac 6 was 18 times more effective at killing these leukemia stem cells. Again, these are the dreaded cells
that caused the relapse of the cancer. These are the ones that previously had no good treatment, and they were very difficult to kill. But
co-scientists found that this drug was 18 times more effective at killing these types of cells compared to a
normal healthy cell, which means it's a really effective and selective drug. Again, this is solid proof that AI isn't just making up ideas or spitting
out known information anymore. Here, it actually output a new way to repurpose a drug to treat a very difficult cancer. And this is a new method that
human researchers have not thought of before. This is a genuine medical discovery. Now, beyond
single drugs, the AI coscientist was also able to tackle combinations of drugs, which is actually really difficult. You see, one drug is just a single
variable. But combining two drugs creates thousands of potential interactions to test for, including effects and toxicity and more. Introducing a
third drug increases the possible combinations into the millions. Lab testing all these different
components of drugs is pretty much impossible. But because this co-scientist was able to synthesize all the information from the scientific literature
out there, it's able to understand and predict these combinations a lot better than even the top human experts. So again, targeting this specific type
of leukemia, the AI suggested this combination of three drugs, JQ1, Olarib, and MSA2. That was
just its hypothesis. So again, the researchers put this to the test in real life. And what they found is that this specific combination was indeed
very effective against the cancer. Specifically, it works synergistically to attack the cancer from three different angles. The AI was correct in
saying that these three drugs together were significantly better at stopping the cancer than any of them
were on their own. This is a huge deal because testing so many drug combinations is just economically and physically impossible. There are millions of
potential combinations which is way too many for humans to test in real life through trial and error. But this breakthrough shows that we don't need
millions of expensive lab experiments. This AI can just synthesize all known information and come up
with the best combinations on its own that actually work. This makes drug discovery way more efficient, especially for combinations of drugs. So,
these are just a few examples of how you can get this AI agent system to find new treatments for a type of cancer that was very difficult to treat.
But it doesn't just do cancer. If you want to supercharge your content, definitely check out Higsfield,
the sponsor of this video. Think of it as an all-in-one AI creation platform built specifically for creators. Instead of jumping between a bunch of
different tools, Higsfield gives you access to some of the world's leading models in one place, including Seed Dance 2.0, Nano Banana, GPT Image 2,
and more. What makes it really interesting is that Higsfield also has its own tools built for full
creative workflows. For example, there's Higsfield Supercomputer, which is basically a general purpose AI super agent for content creation. You can
give it one prompt and it can help with the whole process. finding an idea, creating the product concept, writing the brand direction, generating the
visuals, cutting the launch video, and even helping with distribution. And for actual video
generation, Higsfield has Seed Dance 2.0, which can generate 1080p videos with synced audio. It also supports multiple inputs, so you can combine
text, images, video clips, and audio references to control the final results more precisely. They also have marketing studio, which is really useful
for making ads. You can paste a product link or upload a product image and it can generate multiple ad
formats in one workflow like UGC videos, tutorials, unboxings, product reviews, and more. One product can instantly turn into a whole campaign. They
also have Cinema Studio 3.5, which is built as a full end-to-end film making pipeline. This is especially useful if you want more cinematic control.
Instead of just typing a prompt and hoping it looks good, Cinema Studio is built around actual shot
level filmmaking. You can plan the scenes, control the camera, set the mood, add characters, and keep everything consistent across the whole project.
The big idea is simple. From idea to final output, Hakesfield gives you more control over the whole creative process. Whether you're making ads,
social videos, product launches, cinematic clips, or full marketing campaigns, this is one of the easiest
platforms to start creating with AI. Try Higsfield today using the link in the description below. The paper also goes through how AI was used for
liver fibrosis. This is a condition where the liver develops excessive scar tissue usually from inflammation and it can eventually lead to liver
failure. The researchers asked the co-scientist to find new epigenetic targets for this disease. In simple
terms, epigenetics refers to the mechanisms that change how genes are turned on or off. Identifying these targets is like finding which specific
switches in the DNA tell the liver cells to produce excessive collagen and scar tissue and other stuff which cause the fibrosis. So once we identify
these targets, it's easy to develop therapies that flip these switches back into a healthy state. So
again, the AI system took this task and came up with a ton of hypotheses, debated with itself internally until it came up with a leaderboard of the
best solutions and it ultimately was able to identify new epigenetic targets and proposed the drugs to hit them, including a drug called Vorinostat.
The breakthrough here is that Voronostat is already an FDA approved drug, but it's actually used to
treat a rare type of lymphoma cancer, not liver disease. So again, the scientists tested the AI's theory in real life, and indeed they found that
fenoat actually successfully reduced liver scarring without being toxic to human liver cells. This is yet another example of a breakthrough discovery
from this AI system. It shows that AI can accelerate the search for treatments by finding hidden
potential in drugs we already know about, drugs that are currently used for completely different purposes. And it gets even crazier. So let's discuss
the third use case for this which is antimicrobial resistance or AMR for short. This is arguably one of the most severe global health threats of our
century. Many common diseases like pneumonia, tuberculosis or UTI are caused by bacteria, fungi and
parasites. Now we use drugs called antibiotics to kill them. But the problem is that these can evolve over time to survive these drugs. Basically, the
bacteria can adapt and become resistant to antibiotics and other treatments. Meaning medicines designed today gradually lose their effectiveness.
Well, in this paper, by using co-scientist, they were able to uncover a major discovery in how bacteria
spread this antibiotic resistance. Here's more background on the problem. The researchers were trying to understand how antibiotic resistance spreads
so quickly between completely different species of bacteria. So far we know that resistance is often carried in mobile genetic elements called
cfpicis. These are like specialized packages that move from one bacterium to another. But usually these
packages are very picky about which bacteria they enter. And what scientists don't fully understand is how these CFPICs were able to jump between
different species so easily. Well, using co-scientist, researchers tasked the system to find the mechanism behind this. They just said, "Hey,
co-scientist, here's a weird genetic element. Figure out how this bacterial gene transfer actually works." And
in just 2 days of autonomous literature review and reflection and internal debate, co-scientist generated a topranked hypothesis. It predicted that
these CFPICs actively interact with diverse fage tails. In really simple terms, these fage tails are essentially virus-like docking tools that latch
onto and enter a host bacteria. The AI hypothesized that these CFPICI genetic elements hijack the fage
tails to dramatically expand the range of bacteria species it can infect. And that's why antimicrobial resistance can spread so quickly between
different bacteria species. Again, keep in mind we didn't have this knowledge before. This is a completely new idea that the AI deduced from its
understanding of scientific literature. The remarkable thing is that the AI's theory actually perfectly matched
the findings of an independent team of scientists who just figured it out in the lab after months of work. They haven't even published anything on it
yet. In contrast, the AI was able to deduce all of this in just 2 days, which is crazy. This proves that the AI isn't just regurgitating information.
It's actually coming up with completely new scientific ideas that work. Now, so far co-scientist is
pretty impressive, but it hits a limitation pretty quickly. So far, this can only synthesize information and come up with hypothesis. The next major
step in AI science involves a system that can actually analyze the results once these ideas are tested in a lab and then iterate again and again. We
need an AI system that can work alongside scientists as they do experiments and help them iterate
further. Currently, scientists have to do a lot of the grunt lab work. It's messy and intensely physical. You need to gather evidence, pipet liquids,
culture the cells, and also write code to analyze a ton of raw data. All of this takes an immense amount of time, which is why most biology scientific
papers takes months to years to publish. How can we get AI to automate these steps as well? So
that's where this next paper comes in. So this is called a multi-agent system for automating scientific discovery and this was also published in
nature in the same week. Now this is also an agentic system called Robin. But here's where it differs from Google's co-scientist. Co-scientist only
focuses on the thinking part of discovery. But Robin is what they call a closed loop system. It not only
comes up with new ideas but also interprets raw data from lab experiments to refine them in real time. So this is an AI system that can manage the
entire scientific cycle from start to finish and keep looping and iterating again and again to come up with new directions or dig deeper. Let's look
under the hood of Robin. So it relies on this system of specialized AI agents, but they each have very
different roles. First we have Crow and its primary function is to perform accurate concise literature review. So it's like the researcher. If the
human user asks a question about a disease or something, Crow scans thousands of scientific papers to find quick summaries of how the disease works
and which lab tests should be used to study it. Once a specific treatment or a drug is suggested, the
information is sent to the next agent called Falcon. It does a deep dive into the literature and writes a comprehensive report on that drug. It looks
at its safety, how it works, and potential risks. But the true star of the Robin ecosystem and the agent that fundamentally distinguishes it from
co-scientist is the third agent called Finch. This is like the data analysis agent or the lab partner.
After a human scientist tests the drugs in a lab, they just give the raw data to Finch. And Finch can then write its own code to analyze these numbers
and create charts to see if the treatment actually worked. It autonomously writes, debugs, and executes code to analyze the raw unstructured data that
comes straight out from lab results, which is often the most brutal part of doing science. This is
a huge distinction that we need to emphasize. It's one thing for AI to read a published paper which is already well formatted with graphs and clearly
written sections. It's entirely different and way more difficult to look at raw data from science experiments and figure out what the hell is going
on. It's notoriously messy and noisy and it could be highly subjective. Two human scientists might
look at the same data set and draw different conclusions which makes it very prone to human error and subjective bias. So how does Finch handle that?
Well, the researchers building Robin knew that if they just asked a single instance of an AI model to analyze this messy data, it might fail or
hallucinate something or make a bizarre subjective choice. So to ensure accuracy, they gave Finch a
fascinating architectural design. When it receives raw data, it doesn't just run Finch once. It actually launches eight independent instances of the
Finch agent to analyze the exact same data in parallel. It's like asking eight different research teams and they all write their own code. They all
clean the data their own way and they all reach their own conclusions. Then the findings of these eight
Finch agents are analyzed together using a consensus mechanism. In simple terms, the idea only gets accepted if the majority of the agents agree on
it. In other words, if at least 50% of these independent ideas reached the same conclusion. This process is super important because scientific data
can be very messy and hard to interpret objectively. But by requiring a consensus, the system can output
highly consistent and trustworthy results. The really cool thing about the design is that it's an endless loop. So the findings from Finch can be fed
back to the first CRO agent to generate new hypotheses and figure out what to do next. It's an endless cycle. You can keep iterating again and again
based on lab findings to eventually narrow down a solution for your specific problem. Now, this
sounds great and all that, but the next obvious question is, of course, how well does this actually work in real life? Let's go over some of its
impressive findings. To prove the system works end to end, the researchers gave Robin a massive complex target, dry age- related macular degeneration,
or DAMD for short. Now, this is a devastating widespread disease. It's the leading cause of irreversible
sight loss in the developed world and it's incredibly common as people age. Now, currently the therapeutic options are very limited. There's no highly
effective treatment for this. So, we are in desperate need of a breakthrough. So, the researchers hand the wheel to Robin and it starts round one of
its iterative cycle. The crow agent scans the vast literature on the eye and Robin hypothesizes that
a key cellular mechanism to target is something called RP faggoytosis. The RP or retinal pigment epithelium is a single layer of specialized cell
located at the back of the eye and one of the most important daily jobs is to act as a garbage disposal for the eye which is the faggoytosis step.
Basically, it cleans up waste products to keep the eye healthy. The logic from Robins seems
straightforward. If the RPG garbage disposal breaks down, then the trash in the eye builds up, it becomes toxic, and you go blind. In really simple
terms, that's the progression. So, Robin's hypothesis is, let's find drugs that turn the garbage disposal back on high. Robin uses its agents to
suggest 30 existing safe drugs that might artificially enhance this RP fagocytosis step. In fact, it even
suggested how to perform the experiment to test these drugs. So, the human scientists ran with it. They physically conducted the experiment with the
30 drugs in real life, measured the results, and then fed all these results back into the system. So, next, the Finch agents had to analyze this data
and figure out what to do next. Here's where the magic happens. Finch launches its eight parallel
instances. It writes the Python scripts, cleans the data, runs the statistical comparisons, and forms a consensus. It confirms that out of the 30
drugs tested, a specific compound called Y27632 significantly and undeniably enhanced RP fagocytosis. The cells ate more trash. Okay, so that's a
solid win. The AI found that this drug works. But here's why Robin is bigger than it seems. Because we're
not done yet. This is an ongoing conversation. Robin can look at the conclusion from the Finch consensus and say, "Okay, we know that Y27632 makes the
cells eat more garbage." But how does it do it? What's the molecular mechanism? To find out, it's suggested to run a different, deeper experiment. It
suggested to sequence the RNA of these treated cells to see exactly which genes are turning on and
off in response to the drug. The human team listened to the instructions and extracted the RNA and ran the sequencing in real life, then plugged that
data back into the Finch agents to analyze. First, it's important to note that visualizing RNA sequencing data is really hard because it's impossibly
dense. Well, Finch created something called a volcano plot, which looks like this. It's basically a
complex scatter plot, and every single one of these thousands of dots on this graph represents a different gene in the human genome. Finch isn't just
looking for a slight bump in activity. It's looking for the dots that have literally erupted to the top extreme of the graph, showing a massive,
undeniable change in behavior. And from these volcano plots, Finch spots something incredible. It finds
that a specific gene called ABCA1 is working much harder than usual. And this is actually a huge aha moment. You see, this is totally unexpected and
not intuitive at all. The job of ABCA is to pump excess cholesterol and fats out of the cell. Why does a cholesterol pump matter for blindness? Well,
it's because we already know that ABCA directly interacts with a protein called APOE, which is
actually one of the major genetic risk factors for developing macular degeneration in the first place. So, Robin didn't just find a random drug that
happens to work. It used raw DNA data to uncover a totally unexpected, deeply hidden pathway that connects the drug directly to the genetic root cause
of the disease. It's able to connect the dots across entirely different layers of biology. And we're
not done yet. Remember, this is conversational and iterative, so we can continue digging deeper. It doesn't just settle for the first drug. Instead,
it uses this new ABCA finding to search the literature for better drugs that might hit this pathway more safely or effectively. And it comes up with
two fascinating new candidates, reposal and KL00001. Let's talk about Ripos first. Why did Robin pick
this one? Robin found that Riposutal is already an approved drug in Japan as an eye drop. It's already proven to be safe for the human eye. So the
researchers go back into the lab to test reposal. They tested this on real human cells with the disease and they took the data and fed it to Finch to
analyze again. And the results were remarkable. Reposal dramatically outperformed the original drug. It
was more potent at clearing the cellular garbage and significantly less toxic to the human cells. This showcases how incredibly powerful Robin is.
This isn't just a chatbot where you can ask one-off questions. You can keep iterating through its loop to dig deeper and find more and more medical
discoveries. Now remember that another drug Robin proposed is called KL00001. And this is just as
extraordinary and even more unexpected. You see, this is a circadian clock modulator. In other words, this is a drug designed to mess with the
internal biological clock of the cell. The thing is, no researcher has ever published anything on using a circadian clock drug to treat macular
degeneration. It sounds absurd on the surface. How could these two things be possibly related? But Robin tracing
and understanding the liyinth of gene interactions hypothesized that circadian rhythm proteins directly affect fagosytosis. Remember this is the
garbage disposal mechanism. So the AI realized that this mechanism basically operates on a schedule. If this process is failing because it's off
schedule, a drug like KL00001 might help by steadying that internal clock and therefore boost the cell's
ability to clean up waste again. So when they tested KL00001 in the lab, Robin was right. It successfully enhanced the cleaning activity in those
human eye cells, which would treat macular degeneration. Again, you can see here that it validated a completely new treatment to a major human disease
which could cause vision loss. Humans had never thought of this treatment before. What stands out about
this entire multi-round process is the beauty of the loop. It's a true dynamic conversation. It's like working alongside researchers. Robin had an
idea. The humans conducted the experiment and fed the data back to Robin. Robin then learned from the data, uncovered a new mechanism, and came back
with a fundamentally better idea, and the cycle repeats again and again. It's the actual scientific
method, fully automated and operating at a speed humans alone simply cannot match. And while we're on the topic of speed, let's talk about the
economics here because it is staggering. If we look at the time on task analysis detailed in the paper, a human scientist trying to do exactly what
Robin did, reading 551 specialized papers, generating hypotheses based on them, planning the physical
experiment, then painstakingly writing the code to analyze the raw data. It would require an estimated 400 hours of intense focused cognitive work.
That's nearly half a year of full-time work for a human scientist. But what about Robin? It synthesized those 551 papers in just around 30 minutes. It
finished the entire loop, including multiple experiments and data analysis, in less than 2 hours. And
the compute cost for everything is just $10.76. Think about how crazy this is. Imagine spending 400 hours doing some of the hardest, most complex
scientific work and compressing it down to just 2 hours and it only costs you the price of a fast food lunch. All right, so that sums up these two
papers. I hope you can see how impactful they are. AI isn't just a chatbot anymore that can answer simple
questions or help you summarize things. It's already making completely new scientific discoveries and the pace is only increasing. In just one week,
we had two separate Nature papers showing AI agents autonomously conducting research and discovering medical breakthroughs, treatments for cancer,
blindness, antimicrobial resistance, liver fibrosis, and more. We are right smack in the middle of the
acceleration, and it's only going to keep getting faster. So, you can expect an explosion in intelligence and knowledge and scientific discoveries
beyond what you think is even possible. The future is super exciting. And if there's one thing you should take away from this, it's don't die because
I think the next few years are going to be absolutely wild. I hope you enjoyed this video and let me
know in the comments below what you think of this. As always, I will be on the lookout for the top AI news and tools to share with you. So, if you
enjoyed this video, remember to like, share, subscribe, and stay tuned for more content. Also, there's just so much happening in the world of AI every
week, I can't possibly cover everything on my YouTube channel. So, to really stay uptodate with all
that's going on in AI, be sure to subscribe to my free weekly newsletter. The link to that will be in the description below. Thanks for watching and I'll see you in the next one.