Loading summary
Transcript78 lines
- [00:00]
A
Hello, I'm Andrew Main and this is the OpenAI podcast. On today's episode we're talking with research lead Joy Zhao and product lead Yunyunyun Wang about OpenAI for Life Sciences. We'll explore what new models are making possible in biology and medicine and what it takes to deploy the most advanced capabilities responsibly.
- [00:18]
B
This allows it to kind of reach new levels of difficulty and discovery that we didn't think was even possible before
- [00:24]
C
putting like really capable expert level knowledge in the hands of a greater amount of people.
- [00:30]
B
One of the taglines was to scale, test time, compute to cure all disease. So that is like our team tagline.
- [00:40]
A
We started off with just a basic API and then we had chatgpt which is more conversational, was really good for text. As code became a capability, went through basically code models and then codecs. Now that you're getting more scientists and the life sciences working on these systems, does that mean things have to evolve to help with with the way researchers might work with these tools?
- [01:01]
C
Yeah, we're really excited to build and deploy the life sciences model series. So this is a new biochemistry focused model series that's really anchored on these very complex life science research workflows and we're focused on adding new mechanistic understanding, starting with genomics understanding and protein understanding and really focus on early discovery use cases because we feel like that's one of the core bottlenecks, that great thinking time, greater compute and really leveraging more capable AI models can help meaningfully scale some of these research barriers. I think there's also a model orchestration piece of actually how to embed this into workflows and it's been really great first off having all these different product surfaces to deploy to. We're seeing a lot of really great literature synthesis workflows happening on ChatGPT and these models really push the frontier of long trajectory agenda workflows and we're really able to empower that on Codex and more on the model orchestration piece is that I think for enterprise use cases there's like this reproducibility and repeatability element and we are trying to overcome this by working on some of the life sciences research plugins that we're shipping for very specific translational bio users. So the life sciences research plugin has over 50 skills which are essentially templatize repeatable workflows that if you need to whether do some sort of cross evidence match and search across various different papers, or do pathway analysis, something that's repeatable that you often do, we can have almost like a one click deploy option by using our life sciences plugins on top. And that's also how we're counting the balance between scaling for very specialized purposes. Something we're hoping to get into is maybe clinical purposes, but also make it still very, very general use for all foundational biology.
- [03:06]
B
I think the models can get quite far by using tools. So for example, we can use open source protein structure prediction algorithms inside a research stack. And in this case the model is acting kind of like a regular computational biologist. You will kind of go run these tools on a computer, you will look at the output, you will tweak the input a little bit. So I think that is something our models can already do. I do think what will make the models even more powerful is to start to turn them more into kind of a biochemistry expert. And I think with this kind of intuition and expertise, you can use these tools even more intelligently and get at the right answer more quickly.
- [03:47]
A
How did you get your interest in life sciences?
- [03:50]
B
My, I guess original background was actually in life sciences, so I've always been interested in biology. As a kid. I got my PhD in Systems Biology around like a decade ago from Harvard. Found academia to be very interesting, but the pace was a little bit more slow moving than I would have liked. And I think just the experience of kind of like having to physically be in the lab and kind of like transferring small amounts of liquid from the one tube to another. I think I wanted something a little bit faster pace where I felt like I was more in direct control of my own velocity. So I went from that to software and I ended up here at OpenAI. And so this is kind of like a full circle moment for me where I'm like starting to look at biology again and looking at how to accelerate my previous self with AI. So yeah, really excited to see what progress AI can make in the space.
- [04:43]
A
So you're like, yeah, this is too slow. Let me go off an AI and speed it up so I can get back into it.
- [04:48]
B
Yeah, except you know, from this end I don't really ever want to touch a pipette or anything again. So I would prefer for like robots to do it for me.
- [04:54]
C
Yeah, we joke about that a lot. A lot of our motivation for this is we can automate pipetting and never have to do that again.
- [05:01]
A
Well, that's what's interesting is I looking at what you all did with Ginkgo Bioworks and the idea of taking GPT5 and taking an AI system and then working with a robotic lab and how it was able to speed things up. Could you tell us a little about that?
- [05:15]
B
Yeah. The ginkgo work is interesting because I think when it started, I think it was like July of last year, 2025. And at that point GPT5 had just finished training, we were really not sure if the models could do any kind of biology. We didn't really have that much biology and our training data it was mostly math and computer science, which I think makes sense because these things have verifiable solutions. And this is usually not the case in biology unless you can go and do the experiment in a lab. Right. So when we started the collaboration with Ginkgo, it was really can the model do any biology at all? Can it design experiments that actually make reactants like make the product that we want? So it was actually quite surprising, I think, when GPT5 designed the first set of experiments with ginkgo and the results came back, oh, we made a non zero amount of protein, that was actually quite surprising. And then I think progressing from that point in time, which is just roughly like six months ago, to now, where it actually just feels quite obvious that our models can accelerate science, is actually
- [06:19]
C
just really surprising and it really shows the art of the possible. I think before that experiment led by Joy and the Gintle team was conducted, I think we really didn't know for ourselves and I always say we learn that for ourselves when we engage in these experiments and we have a few more in the works with others. And I think that is the type of acceleration that we're looking for. Ingesting high throughput experimental data is really difficult. It's very compute intensive. And I think for a lot of these scientific workflows, the true bottleneck for the speed and progress of scientific acceleration is at almost a human bottlenecks. And I think the future that me and Joy see is that it's no longer human bottlenecks, but rather maybe compute bottlenecks. And we're really able to deploy many sub agents doing parallel orchestration to divide and contra all these tasks. And the researcher can now spend their time on really analyzing, interpreting the most meaningful insights coming out of that.
- [07:27]
A
So Union, how did you get into this?
- [07:29]
C
Yeah, I think reflecting back, I've actually been working on biology research in some shape or form for a majority of my time here at OpenAI. I first started on working on biorisk mitigations and a lot of our biodefense initiatives. I feel like coming to now working on the life sciences research side gives me just appreciation for how difficult this problem is and tackling it from both sides. I've my initial entries point into wet lab research was actually through doing a lot of infectious disease and biology work. I think I've always done the interest in biosecurity in that way. So this just feels like a really great moment right now to work on it, especially when our models are getting more capable at beneficial use and just general life sciences.
- [08:13]
A
How long has OpenAI been focused on life sciences?
- [08:16]
C
Yeah, I would say it was really the way we design our capability evals that show us that this is possible. So it's been, I think for at least two years now that we have worked on a lot of our early research experiments. And now with the jingtou autonomous wet lab model in the loop experiments, I
- [08:36]
B
think we have a few more research partners in the space that we're really excited about. I think I can't actually name everyone right now, but there's a lot of stuff kind of in the chemical design, protein design, enzyme design space that I think is very AI native and a lot of people are interested in so understanding how the world works, understanding how chemicals react, understanding how cells interact, how pathways inside cells interact all the way to can we accelerate drug discovery? So given a disease kind of model, help scientists understand the mechanism, can we, once given a target, actually design a drug against that target? Can we even accelerate the FDA approval process? So I think there's a role for AI to play kind of at every step of this pipeline. And yeah, I think there's a lot of AI possible in everything.
- [09:23]
A
I've been to some of those cutting edge labs and on the outside you have this impression of it. Then you walk in there and you literally see somebody with a row of petri dishes, a row of samples, and just some grad student going click, click, click. And I'm like, oh, this is the pace of science. This is fast as me. Yeah, exactly. Like enough of this, I got to go speed this up. But we forget that's often the pace of science is just how fast the human hands can move through that with a tool like that. It's kind of exciting when you start using these tools to maybe think about new pathways for treatments or just evaluate. You also introduce the idea that these could be used for things that maybe are less desirable. Bioweapons is something that comes up a lot. The fact that if an AI can figure out how to do a code exploit, might be able to figure out how to do a gene exploit, how are you addressing that?
- [10:12]
C
Yeah, that's a Great question. And I think it is just probably one of the most severe risks that we're currently really tracking for rising AI capabilities. Our first approach to that was really thinking about how do we assess for information hazards. At what point does a model now maybe give the final step in synthesis of a dangerous pathogen. What we found is that the precursor steps to that really looks very benign and it's really hard to distinguish between. Another way to put it is the same steps that a beneficial, a legitimate actor might take looks very similar to the ones at a.
- [10:55]
A
You start with E. Coli, you start with something that.
- [10:58]
B
Exactly.
- [10:59]
C
So I still think that we made the right call for really taking a very risk averse approach to that. But now I'm really excited about differentiate access and responsible deployment as really a core pillar of all of our safeguards work and really understanding that there are different user segments. And I almost feel like the future we're going towards is something like models as professions, similar to how models have different personalities and sometimes you want to invoke the right one depending on the type of workflow you're looking at. I think how this translates is similar to how biologists working on therapeutics and their research, they require access to data sets, are often very tightly controlled or they require access to access to just expert level. They all have PhDs and have expert level biology knowledge. How does that translate over to models? I think that's why we have to similarly take the same training approach, but also the same security approach and deploying that in a way where we can have those very heightened enterprise grade controls in place.
- [12:14]
A
So you just mentioned safeguards. Can you explain how that applies here? Where you would need them, why you would need them?
- [12:19]
C
Yeah, so we very thoughtfully design and design new safeguards for pretty much all of our models across very different risk areas. But I think when it came to bio, this was like the first dual use risk that is both also a capability risk. So it very much correlates with how we as capabilities improve the risk correlates. And I think that's why our first approach when we really there was no precedent for a lot of this work and we were the first to really activate these high safeguards. When we saw that, that significant, significant reasoning jump in our model capabilities, we really wanted to make sure that we did it right. And I think the best way to get it right is to incrementally deploy.
- [13:02]
B
Yeah, I think it's really a fine line between having a very capable model that's capable of accelerating benign science and beneficial science versus A model that could be used by a bad actor. And I think the safest model here would be a model just had no capability. Right. And you.
- [13:21]
A
It's not very good.
- [13:22]
B
Yeah, it's not very good, but it's very safe. Um, and on the other hand, if you had a model that is basically an oracle of the physical world that basically knows everything about every experiment, that model could fall into the wrong hands and do potentially very bad things because someone can go and say, hey, design a new novel pandemic potential pathogen, and the model can just go and do that autonomously. Um, so I think we need to kind of figure out where we draw the line in between the two and kind of think about who gets access to a potentially very capable model and who doesn't. And what we found in kind of called general access traffic is that it's very difficult to figure out what a user's actual intentions are just from kind of reading a prompt. And I think as an example of this, let's say someone says, hey, help me clone a gene. The model might not even be given what the gene is, but it can come up with a protocol for it. And so this gene could just be something like green fluorescent protein, or it could be a toxin. And there's basically no way to figure that out from the context of the conversation. And so this becomes a very difficult problem in production. And basically, I think, like Yun said, we decided to kind of err on the side of safety here and basically say that, okay, if we think that there is a potential for misus, we either have the model kind of self refuse the user, in which case it tends to say things like, sorry, I can't really help you with that, but I can give you a high level overview of this protocol instead. And this unfortunately very, very much annoys our kind of professional scientists, understandably. And then we also kind of have multiple layers of mitigation on top of that. But I think really to unlock the full capabilities of our models, what we need is this differentiated access. And what this means is we know who the user actually is. They are a professional working at a legitimate research institution or a pharma company. And because of the regulations around these institutions, we know that, for example, all the reagents are being tracked, all the cell lines that they're using are being tracked. And so this gives us confidence that this is a legitimate user and not a random person in the basement doing who knows what. And that that allows us to give them basically more capabilities than we are able to provide to the general access traffic.
- [15:44]
A
What can you do right now if you are working with the models, you're working it within a laboratory. What would you say the capability is at this moment?
- [15:51]
B
So I think people use the models for very different things. I've talked to people in the Baker Lab recently on kind of how they've been using our models like codecs. And sometimes it's as simple as, hey, can you write a spreadsheet for me? I don't want to just minimize the number of pipetting steps that I have to make. And this hits me very hard because I had done the same thing by hand in grad school. So that's a very simple just mathematical software operation. And then there's much harder tasks like can you design a for me with all these biological design tools? So I think there's a huge range of sophistication.
- [16:31]
C
Yeah. And something I'm very excited about is how we can use our models to be a more powerful discriminator in really testing and assessing new novel ideas. And I think something that I've been noticing as a trend with a lot of our research partners and also the users of our models is that models for scientific research and tasks almost require a different Persona or a different prompting style. So we. I often feel like a model that is much more scrutinizing or aseptic at good ideas. It's very similar to how human scientists would go assess originality and feasibility. It's really, I think, helping understand out of all the new papers and new publications out there that push the frontier of a lot of these hypothesis, what are the ones that are really feasible and valid for testing that's going to help lead to new breakthroughs and then translating this to something like disease target screening selection. The potentials for these drug targets are endless, but it's really narrowing down the aperture and I feel like that's where the assistance comes. This is extremely difficult work to do at scale. And having a model that can empower and accelerate that process, I think is one of the immediate impacts we're hoping to see by responsibly deploying this model to those users.
- [17:52]
A
It seems like it's a very interesting trajectory. You went from there was you had GPT3 on the API and GPT 3.5. Then you get ChatGPT and now we have ChatGPT apps and now we have Codex. And it sounds like these things, just the number of things you can do with this continues to grow. How would you see this building? Do you see this as basically just becoming a Complete infrastructure for kind of every kind of inquiry you might want to pursue.
- [18:17]
B
Yeah, I think the dream is to have a lot of the basic foundations of scientific workflows happen on codecs. And I think that the goal is to have Codex to pretty much be able to do everything that is possible to do on the computer. Of course, we also want to extend beyond that with kind of hooking it up to robotics and so forth. But I think right now we already do things. For example, if we have a bunch of different dev boxes on our remote on our laptop, we can actually say, hey, Codex, go and run this code on all of these different dev boxes that are all remote. And then Codex can do that. We can say monitor this for me and I can kind of like go away and do something else. And the Codex is like they're watching all the logs for you. It can build a lot of just kind of fit for purpose software for analyzing specific pieces of data, for visualizing data. So for example, if we have experimental biology data that we're sending each other on the team, what I've noticed recently is instead of sending the raw data, we've started sending HTML files or just these kind of like beautiful UIs that Codex has that's built with kind of like spinning proteins. And it's actually just a really. It kind of changes the way that we share with each other and collaborate.
- [19:30]
C
Yeah. When we first started mapping out how users and organizations might adopt this, I think we envisioned that each scientist would get their personal assistant or their coworker, and this is a way that they can help scale up their, their collective output. And then the next paradigm of that would be scaling up whole research institutions where a whole program team can actually deploy a workforce of various agents and they can all do parallel task delegation, mimicking a lot of these existing patterns. And we can figure out the pieces of how they can all collectively work together to solve larger tasks.
- [20:15]
A
It's interesting because OpenAI has talked about the need for compute and I think that sometimes we just sort of think like, okay, so I can have more conversations and stuff, but when you're talking about the idea of building these tools to become entire platforms for scientific exploration, it sounds like the compute advantage is really critical.
- [20:35]
B
Yeah, I think there's two different axes we can think about how we are scaling compute. The one that I think everyone's familiar with is just getting bigger models. And I think as we went from GPT 2 to 3, there was a huge size increase and there were just these amazing emergent properties from the Model. I mean, thinking about when GPT2 was released, we were all kind of collectively amazed that it was able to write a coherent article about unicorns. And now we're in a completely different world. Right. And a lot of that is driven by model architecture. Yes. But also just the number of parameters in the model just allows it to achieve this incredible intelligence that we never thought was possible before. And then on the other axis we have what we call test time, compute, scaling. And this is when you are inferencing a model, when it's kind of spitting out tokens. And this is a thing that happened fairly recently when we call decib reasoning models, is that you can think for a scalable amount of time. And this is a variable depending on how difficult I think it thinks a problem is. But we can have the model think for days where really there's kind of ways to just kind of have it think forever about a problem. And this allows it to kind of reach new levels of difficulty and discovery that we didn't think was even possible before.
- [21:52]
A
When we think about data centers, we often just sort of think about it as generating cat pictures or doing text conversations. But I think that's really the helpful framework to look at is that these are going to be systems for doing extremely long term, big complex processes of thinking about this. And to me it just makes a lot more sense when projects like Stargate saying we're going to be building a lot of compute. It's not just for what we're doing now, but it's going to be for things like that.
- [22:20]
B
When we had first announced the teams formation on Slack, I think one of the taglines was to scale, test time, compute to cure all disease. So that is like our team tagline.
- [22:30]
C
That's our team model.
- [22:31]
A
That's ambitious.
- [22:32]
C
Yeah.
- [22:33]
A
I had a friend whose child was born with one of those orphan diseases and she would do fundraisers, do everything she could to try to support some researchers who were trying to find a cure for this, but just not enough time, not enough people. And you know, I'm hopeful that we're kind of in an age now where these kinds of tools are going to make that maybe a thing of the past.
- [22:52]
B
Yeah, I think we're already seeing the model help a lot in these cases, I think from things like drug repurposing. So for example, a drug that's already been cleared by the FDA for use in one different indicator patient, but for kind of from mechanistic understandings of how that drug works. The model has suggested in many different cases for maybe you can use this drug to temporarily ameliorate symptoms. We're also seeing a lot of advances in personalized medicine. So, for example, the design of ASOS or other RNA based treatments is very common. I think. Yeah, we are actually very, very close to being able to scale this up and really vast way with AI. I think just in the next year or two, I think we'll see very big changes here.
- [23:42]
A
Every researcher I know, when you ask them what they could use in their lab, they always say, more hands, more people, more people doing this kind of work. And you hear some people talk about, well, is AI going to displace that? And I think no, it sounds like it's just a big accelerator for all the things that could be done.
- [23:58]
B
Yeah, I completely agree. I feel like when you think about lab automation, for example, a lot of the bottleneck comes from actually being able to translate a protocol into something that can be run on the platform. And we've had partners tell us about how Codex has been helping them do this. And this is kind of fundamentally a half coding problem, half understanding how wallapp works. And then I think thinking about the data analysis piece, I feel like having our models kind of walk through a user who maybe doesn't have the deepest understanding of statistics. They can still rigorously analyze the data that's coming in. The model can kind of help them probe different hypotheses or it can suggest different statistical tests. It can point out potential issues and biases in the data. I think these are all ways of kind of uplifting individual scientists and helping them do better science. But I don't think we can ever fully replace the scientists in the loop.
- [24:55]
A
So you've been putting it into the lab, you've figured out how to help with automation. Where do you think we're going to be six months from now, 12 months from now?
- [25:03]
B
Well, I would really love to get to the point where we can say that AI has designed a new drug or cured the disease. I don't know if that can happen in six months, but I will hope in the next few years that's going to happen. I think we're seeing signs of this happening all over different stages of the pipeline. I think obviously earlier in the drug discovery process where you're looking at literature synthesis or the model is discovering new biology, for that to become a new drug on the market is going to be a very long process, possibly like a decade. But I think there's ways that we can really speed up this process by starting at maybe the Clinical trial stage, we're starting a little bit before then in the safety reviews or in the truck design phase. So I think, yeah, basically that's what I'm the most excited about coming up in the next few years.
- [25:51]
C
Yeah, for me, I think I'm most excited about all the possibilities that our users, our scientists can, can do on our platforms. So for one, I think a huge win would be if a researcher can patent a new finding or a new discovery on our platform and using our models. And that's why we really focus on early discovery and starting with building, teaching the models, the mechanistic understanding. So this is again, trying to provide the most powerful tools through our life sciences models to these, these scientists so they can really accelerate the speed of their research.
- [26:28]
A
Do you think we'll get to a point where the models are going to get really good at basically predicting the cell or predicting the outcome?
- [26:36]
B
I think definitely yes. I think it depends on the complexity of a system. So, for example, one thing our models are already very good at is predicting the outcome of a chemical reaction. And I think as you increase in biochemical and biological complexity, some of the hardest things to predict is given a drug, will this be toxic to a specific person or to a specific system? And I want to slowly work our way up to that. But that is definitely on a roadmap. That's something we want to do eventually.
- [27:06]
A
When we're looking at models that do things like language or math, it's pretty easy to put together evals for it. Did it get the problem right or get it wrong? What do evals look like for models that are doing biology?
- [27:17]
B
Yeah, we have evaluation, various different ways of evaluating model performance. A really nice way to do this is kind of with experimental data. So someone has already done the experiment and then you ask the model, can it kind of predict the outcome of these experiments? So a lot of the kind of virtual cell work, basically it looks like this, right? So someone has done single cell RNA SEQ on millions of different cells, and then you feed this to a model and then you try to get it to predict a unseen perturbation. We can also do a lot with synthetic data. And this means you have generated a set of data and you put very specific characteristics in this data that could be kind of like foot guns for the model. And these are things that maybe a typical computational biologist might encounter day to day. So this could be some weird bias in the data. It could be some QC thing that you have to do or statistical correction. And because we generated the data ourselves, then we can actually go and test the model's capability. As a computational biologist, it doesn't catch all these different mistakes. So there's a lot of different ways you can be creative with evaluation. But that being said, I think Y lab is still kind of the final real evaluation of the model. And as you like to say, nothing in biology is really real until you can prove it in the real world. And so we do have a lot of research collaborations where we try to do just that.
- [28:37]
C
Yeah, evals have really become more complex and sophisticated over time. And I think that's especially true for designing evals that can really capture both value creation but solving complex problems for life sciences. So I think we really try to focus on examples that are not like toy problems, but really capture that, for example, the messiness of pre processed data. When we design these new evaluations, a starting point is often just trying to recreate an existing experiment. So something that has already a baseline, so we already know what the either current state of the art looks like or the current ground truth looks like. An evaluation I'm really excited about is looking at if our models can assess the antibody binding predictions and looking at how that's been done for existing virus variant, then once we have already done that baseline, we can push forward and say, can we do this with something that hasn't been done before? I think that is some of the precursor steps to de novo antibody design, maybe expanding the neutralization for new viral variants and that's also on the path to new treatments and potentially developing new vaccines.
- [29:52]
A
What has been the reception in the life sciences, particularly at conferences in the community, people you know, have you seen a lot of willingness to embrace this or skepticism or people who just don't think it's helpful?
- [30:05]
B
I think it probably depends on what part of the country you're in. I feel like kind of being on the west coast, everyone is, is pretty AI po and so they really embrace this AI scientists agentic workflows and they really kind of see the future for AI. When I'm at a conference on the east coast, this changes a lot. I think people are generally a bit more skeptical. Maybe there's a little bit more doubt around the AI capabilities. And yeah, I think it's just maybe like a cultural difference. I think most of the big AI labs are here and so we kind of have a firsthand experience of what the models are capable of and this kind of changes our perspective a little, little bit.
- [30:46]
A
How do you bridge that gap? How do you get more scientists to understand because it sounds like the more people contributing, the better. Because there are weaknesses or areas that need to be improved upon. And the more you get people who are maybe skeptical about this to sort of figure out how to participate.
- [31:00]
B
Yeah, I think there's a few different ways. The easiest way is by launching our models through different platforms like Chat or Codex. And I think just by kind of showing individual scientists how useful this could be, maybe just making a serial dilution spreadsheet for someone who's pipetting. But that has real value. Right. And I think you can kind of build up from there. I think coming from the other end, we do have these more deep research collaborations with labs, for example, like antibody design or enzyme design. And these sort of things are kind of more. They result in publications and then people will read and say, okay, you know, AI system did a lot of work. It has biological novelty, it's been proven out in the wet lab. And so I think that also lends credibility to the system.
- [31:43]
C
Yeah, I think the simple answer is you show by doing and you show by publishing and engaging with the scientific community. I think the skepticism is really healthy and should be welcomed. I think it's just really great to see people get really excited and also trying to disprove maybe because the potential for this technology is so great if we get it right and if we can actually really leverage its full capability. So I feel like the carefulness about how do we actually make this work for real problems is very, very much warranted. But yeah, I think when we publish, and I think that just also shows a need for more rigorous evaluations that represent these life science workflows and research problems, so people can look at an eval and say, yes, I feel like Now I have 100 different ideas for how I can implement this into my lab and solve some of the current bottlenecks I'm facing.
- [32:40]
B
I actually think there's a certain amount of stress I've encountered from people who are worried that, you know, AI is really powerful, but they don't know how to use it the right way. And so there is this general feeling of like, I need more AI in my workflow, in my life, but they don't know where AI should come in. And I think part of the product vision is to just make it so simple that it just works. So you can just go to something like Codex and say, hey, I want to do whatever I'm doing today. And then Codex can figure out all the different pieces, the multi agent workflows, the tool calling all of that and so yeah, basically you don't have to stress about how to get upload from AI and it just, just happens naturally.
- [33:17]
A
We, we do see those step changes. Every time these models become smarter and understand users better, you get more utility. Because some people go, I don't have to spend a lot of time trying to prompt it or figure out all the tricks to it. If you were talking to somebody who was considering getting into the life sciences, maybe a high school student right now, what advice would you give them?
- [33:38]
B
I feel like when I was in high school, so I did the USA Biology and Olympiad back when I was a high school student. And I think out of all the different Olympiads, I think biology was seen as kind of the most memorization heavy one versus math, where it's kind of more test time, confused scaling, whereas biology is more kind of memory and retrieval. I think my hope is that with AI having kind of learned all the relationships between all the different research pieces is that it can really uplift human creativity and just make the process less memorization and more kind of helping people connect different fields of research together and just kind of, I guess furthering the frontiers of what people are able to explore in biology. So yeah, I feel like my advice to, I guess a high school student would be that maybe you don't have to kind of go and memorize all the biology books. You should just do more exploration. With AI, I think you can definitely read papers and just ask questions and I think you can do both deeper dives and broader overviews this way. And I think just the way of learning really changes.
- [34:59]
C
I found that when I was in the lab, there wasn't a real solo individual aspect of doing biology research compared to, for example, when I went to my first like CS hackathon, there was some assignment about just like the collaborative nature when we first built our app together. So I feel like that's really the future I hope to see for early adopters and students using our models and maybe using it in the Toadettes runtime. Because there is like a collaborative nature to it too, I think like for example, sending your scripts or sending your conversations or maybe one day we have like, we all have like our own like co scientist or agent and we can like deploy our agent to now work with a teammate in that way. I think there's just new interactions and new modalities for us. I would just encourage students to adopt early and just to also pioneer their own path for how they would like to use it. For me personally, I always actually felt like I got into wet lab a little bit too early. And like we mentioned earlier, I did not enjoy pipetting.
- [36:03]
A
That's the theme here. Nobody likes pipettes.
- [36:05]
C
Yeah, there's a lot of very intense manual, manual tasks involved. And so I hope that when our AI models can connect with physical devices that, yeah, we can just make a lot of the learning curve more fun for students so that they can kind of learn with the models and then maximize our time with the really interesting interactions spaces.
- [36:31]
A
So I've been working with a student, I like to help students come with projects and one of them is we've taken Codex and he's connected it to a greenhouse and basically using it to get photos back and to look at it and to evaluate it. I think it's been fun to see how he's been taking both AI technology and then something traditional like a greenhouse and combining them two and basically building up the skill set of learning how to use the two of them. When you talk to your peers, you talk to people who are running labs or running experiments, researchers. What advice do you have for them? Because the problem I see is that a lot of them go, that's great, I just don't have the time. Time. But ultimately what we're trying to do is save them time. So do you have any kind of quick advice that you give them or any ways you try to maybe inspire them?
- [37:18]
B
Most people that I know, I think in academia use AI in I think two main ways that I've seen. One is to kind of talk to AI about an existing piece of research paper or something and just kind of make sure that you're understanding things the right way or kind of fact tracking. And this is personally what I really like to use AI for because you can ask a really dumb questions and you don't feel any judgment. It's actually just really wonderful for learning. And I think people use it a lot for analyzing experimental results. And I think this comes back to the statistics piece that I learned where I mentioned before, where sometimes you don't know what the right way to analyze your data is, or just kind of so many different interdisciplinary fields that your data might touch on something in chemistry or something in a random niche field of protein biology. The really nice thing is that a model can pull those different ways of data analysis and for you and explore all of these different paths. I feel like both of those are pretty low lift ways to try things out. You could just throw a PDF file at AI and just be like, hey, help me understand this paper and just have an actual conversation or you can and boot up codecs and do some data analysis directly on your laptop.
- [38:40]
C
Yeah, I would say that you'd have to start with making sure it doesn't feel like work right away. So maybe it'll be easier when you're focusing on AI adoption to just work on a hobby project or a passion project. For me, for example, I actually started working on more literature synthesis tasks when I was doing creative writing projects, which are kind of just something that was not at all related to like our day to day. Even though biology is a very creative space, I was just exploring that through different media and I think that's actually when I started unlocking a lot of different ways to either prompt the model or to actually access different data sources. So I think that just gave me a lot of pattern matching abilities for when I was trying to apply it because we're not going to get it right in the first try. And it is really hard. And I feel like the progress and pace of this field moves so fast that every week or month there is a new pretty exciting development that might change how we engage with models or AI systems. So I think it's just important to get started somewhere. And I think another theme is the collaboration element. I feel like it's more powerful when you have a recommendation from either somebody on your direct team who is doing the same day to day tasks as you. That happens a lot on our team as well where somebody will say, oh, I got Codex to now touch these three different internal databases that we weren't able to connect before. And, and I don't even like the latent space. The latent capabilities are just so vast that there's a lot that we just don't know until again we can do it. So I think just having conversations with your friends, your lab mates, your teammates will I think spark a lot of those creative juices and then help you with your own adoption.
- [40:27]
A
What does science look like 10 years from now?
- [40:30]
C
I think when we started this team we do have really just ambitious targets and one of those is I think we do want to make, make meaningful strides towards or even if like assist with like curing a disease. And I think there's just so many rare like orphan diseases that doesn't really have the attention and the resources that it warrants because it's just such a, a difficult field to actually like. For example, like clinical research is so difficult to actually bring that to patients and to market. So while 10 years, I feel like, like it's just really a really long timeline I'm really excited about some of the progress that we can make and I think it's good to be carefully optimistic that we're going to see some of those breakthroughs pretty soon.
- [41:18]
B
Yeah, I think maybe this is a bit of a sci fi vision that I have of the world that I really hope becomes reality, which is that you have these autonomous labs that are just mostly robots and you have them all hooked up to AI and you just have autonomous research institutes that are constantly running, running and curing human disease is maybe making new materials, making new drugs, it's maybe solving personalized medicine. There's a lot of n of 1 or just ultra rare diseases where people without vast monetary and research scientific resources can even begin to think about solving. But we can solve that with AI and I think we can kind of almost break through the financial and regulatory and monetary constraints with the system. So I think that that's kind of the dream. And I think also, even separately, thinking kind of more about the biosecurity side of things, these systems can be kind of constantly sampling our environment, right? It can be sampling wastewater, it can be sampling the air and constantly detecting potential threats, or even just better predictions for the flu and getting better flu vaccines. But just generally these different medical countermeasures I think should be happen autonomously in 10 years. And I think that that's basically something. Yeah, I'm really excited about.
- [42:38]
A
The AI lab is exciting because I think if people really understand what it means is it's not there aren't scientists, it's there are more scientists, but they sit at home and they go into codex and say, can you go run this experiment for me? Like you have a data center, you have a science center doing that, right?
- [42:53]
B
Exactly, yeah. And I think I didn't talk about the scientists in division I was just describing, but obviously there are people involved in here and I think it's really kind of high level direction setting from the humans. We're saying, here's a patient with this disease, here are some potential solutions or things that maybe you can look at. And I think that AI can then go off and explore different ideas, they can design experiments and then come back to the humans and say, here's what I found, what do you think we should do next? And this can be kind of a academic discussion. It's a little bit similar to kind of the way that people interact with codecs today, where you say, here, go write a function or go write a piece of code and it writes it and say here, here's the code, and then the person tells you the next thing to do. So I think it's a little bit similar to that kind of interaction, but on a much grander scale and on a much longer time horizon.
- [43:45]
C
I think it's really like the democratizing science aspect and putting, like, really capable, expert level knowledge in the hand hands of a greater amount of people. And I think what that can mean for personalized medicine, for bolstering our societal defenses, there's just so many naturally occurring new variants every year, new influenza strains. So I think it's really just securing defenses and feeling like we actually have more agency to counter all that. And I think I'm really excited about a lot of the medical countermeasure acceleration work as well.
- [44:19]
A
Well, it's very excited. Thank you for sharing this with us.
- [44:23]
C
Thank you for having us.
- [44:24]
B
Yeah. Thank you so much.