I think I am your target audience in disposition (you are the first fellow slatestarcodex-reading bio person I've encountered, hello), though probably not aptitude. I was driven to go into experimental biology because I thought my enjoyment of math/cs problem-solving would translate to experimental methods development. The NGS, optogenetics, and cas9 waves were all huge when I was in college, and it seemed like that momentum would continue.
I worked at a metabolic engineering company for 3 years after college to get broader exposure, which was super eye-opening. There were ~50 PhDs there from all kinds of different areas, which was a huge contrast from the more focused nature of many academic labs.
While I enjoyed talking to the people in experimental biology, I didn't enjoy the experimental work itself. I found it pretty stressful and repetitive -- long hours, a single moment of inattention crushing a week of work, hard to translate one's skills to different roles... I wasn't super money-driven, but seeing random FAANG friends working flexibly and making 250k, conspired to get me out of experimental work. I didn't want to be 30 and not have employable computational skills.
I think the repetition of the work deters a lot of people. My swe friends would say that their work too, was repetitive, and complain about "protobuf plumbing." I felt they _fundamentally didn't understand_ how repetitive and time-consuming (and delicate!) doing westerns was just to collect a little bit of information.
I was further convinced not to pursue methods development when I saw a PhD-level job talk from a guy at Big Name Lab at our company whose PhD research had like... not fully worked out. Methods is high-risk high-reward. Interviewers had a positive impression of him and gave him an offer, but were skeptical of how his skills would translate to the company/role, and he also didn't have computational skills.
As you said, I think the lack of agency is another hard part. Since experiments are expensive, it's hard to work on your own ideas. I know lots of experimental grad students and post-docs who have only ever worked on PI's projects... another aspect of this is that since people are spending like ~50-60 hours a week *just doing the same lab protocols*, this leaves a lot less time for actually thinking. I think this probably colors the incentive structure of experimental vs computational fields.
Many people are repeating maybe a dozen protocols hundreds of times, not interacting a lot with people from people from other environments. I remember asking a protein engineering PhD how she thought some reaction worked, and she asked "well what does the active site look like?" It felt like a question out of left-field for me, because Actually Thinking About Structure Details felt like a totally different level of abstraction compared to what I was used to.
Tying into this is that many academic labs are pretty siloed. (I think the biology groups I've worked in have been some of the most intellectually vibrant / fun environments ever though, so not trying to knock everything.) But many people are spending 5-7+ years in a single lab with like <8 people, and conflate that with some kind of Broad View of Science and Society (not claiming to have this, lol). This also means that labs turnover more slowly than companies do, since people might churn in <2-3 years. So I think a company might give one more exposure to ideas, but the constraint of "this has to literally-actually make money in the immediate future" is quite different.
In industry there's a lot of people at the bench who have 20+ yoe. In academia... your PI doesn't pipette, and the most senior person in the lab might be like ~3rd year post-doc. I think this means there's like an entire arc of IC skill development that young academic scientists don't see.
At my old job, I would talk to people with ~20 yoe... every 10-15 minutes throughout the day, and we have a shared context and mission of what's going on. In an academic lab, you talk to your PI once a week-ish, and your labmates sort of know what you're doing, but not really. I felt like I learned _a ton_ from just off-hand technical convos throughout the day with people at the company, and found the academic model sort of isolating.
Another thing from working at a company you pick up is learning to sniff out when somebody with 20+ more years of experience than you... can still be wrong. If there's 50+ people with PhDs, you're going to have a lot of contrasting views, and you have a lot of experiences where Bob, who has a Genetics PhD and 20 years more experience than you, is trying to lecture you about something, but it turns out the other senior people disagree and your hunch of calling bs was right. I think this is relevant in grad school, because your PI is the only senior person you interact with, and while they know more than you, they're also a person with their own idiosyncrasies.
I think this is all super true, but at least from what I'm seeing a fair bit of this is getting overhauled thanks to automation, where the idea is you can actually 10x your output and not worry about "a single moment of inattention crushing a week of work". It opens a lot of other potential problems and costs but the trend is for sure away from pipetting things manually.
I'm sure automation will happen eventually, but people have been framing automation as the catch-all obvious solution for 10-15 years now, and haven't seen a big change in the landscape yet. There was a lot of automation/robots at the company I worked at, hasn't made it to academia in any significant amount yet. Not sure what the line is.
Hey, man. Really liked your post and thought I'd reach out. I found especially interesting that you make the point that 10x engineers are born out of freely tinkering their own ideas around, learning what works n all until they have enough of a system understanding that they become cracked at it. And how you then mention that this 'attempt-result Cycle' doesn't really work for bio stuff, but you're anyway at the "top 1% in 'amount of fucking around and trying stuff'". I myself will be starting university for a major in physics later this year, and hopefully eventually work with some biophysics stuff or neuroscience. In any case, I wanted to ask you if you had any tips, or some important things you learned along the way (and the challenges involved in them!) when it comes to just experimenting with stuff and ideas that you'd want to share for someone that'll be just starting this path to excelence in the field. Would be very much appreciated!
Man I have no idea what that looks like in physics, it's really not my area. Biophysics best guess is to find some stuff in the simulation space that you can do on a computer, try to be playing with the latest toys and models.
As best you can try to hang around professors who seem to encourage playing around and trying long shot ideas, which might not necessarily be the ones with the best publication record.
From the outside, think you're probably on target here. I knew a kid who I think had the genius necessary to be a world class software guy, but switched lanes to biotech in college. Last I heard he was off in a lab somewhere in the Sacramento area messing with neuro-something-or-other. No big news coming out of there in intervening 10 years. Pretty confident he'd have been A Big Deal if he stuck to software, but in biotech he's just another faceless lab monkey. And it's fine, it's what he wanted to do - but - there is something about the field that seems to inhibit radical developments.
the 10x or 10^n x engineer simply doesn't exist. take that example: being in the situation where you get to create Google Maps from scratch. it's absurd in the context of modern product design.
You need to sync with so many people with so many different needs and synthesize that into a product with all the various testing (A/B, integration, unit, etc) that goes with it. This is a lot of work, most of which isn't all that technically complex in nature. Because fundamentally, software engineering's primary work is in understanding human needs; not the ability to write lots of code quickly or create crazy innovative solutions to NP-Hard problems. Writing the code is something you do after months of deliberation.
Sorry but lone geniuses aren't really that important, unless you're in a research context, and even then it's vastly overstated.
What's interesting is that a lot of the article is about the environment of the engineer more than the engineer itself.
While 10x engineers may not exist, there are situations that call for speed, efficiency and crazy creativity. It's ultimately up to the environment to either relax the constraints, or have none in the first place.
I feel like the idea of the 10x engineer is kind of crazy. But of course there are outliers in productivity or creativity!
But the real question underneath is whether your environment supports the flourishing of creativity and innovation or not.
I have thought about this before - why are there no FAANG style biotech that started in someone’s garage.
Biology is way more complex and just keeps getting more so with every new layer we uncover. But maybe AI can change that. Sadly I have more hope for AI breaking the mold than anything in academia shifting away from power grubbing advisors squashing the life right out of you.
I think there are academics whose labs are making big progress in biotech and churning out solutions to big problems frequently: George Church's lab keeps making foundational molecular tools that work making a lot more stuff possible. Jay Keasling's lab keeps making valuable new molecules biologically. Markus Covert's lab has made 'whole cell' computer models close enough to reality they've worked to discover numerous real, fundamental biological facts. Now, those guys aren't even really quite engineers, they're closer to CEOs managing a huge operation from a distance. However, unlike software, it takes big teams of people with diverse expertise working together to do this stuff. All 3 of the labs I mentioned above have somewhat hidden, nearly Carmack level software accomplishments that are enabling the wet lab results. Also, until maybe *right now* biotech was held back by 3 big obstacles: lack of predictive modeling/understanding, expense of DNA synthesis, and lack of flexible laboratory automation. Pretty much all 3 of these converged on low cost solutions very very recently, and I am seeing firsthand, as an academic in this field, tons of people finally making stuff work after decades of no progress.
I think there are lots of people in biology with aptitude, experience, and instincts as good as the hypothetical 'cracked 10x engineer' but, as you say, their productivity is hindered by time of execution and high complexity:
Biology vastly more complex and indirectly understood than most fields. Someone with elite instincts may see their hunches come up correct ~25% of the time vs 2.5% for others. If you can only fit 3 experiments into a week, it's still really hard to turn a 25% hit rate into huge productivity.
Theory driven biotech is cracked. There is a theory tying the double helix found in DNA to the double helix found in the 24-cell both with 48 roots for chromosomes. Because the 24-cell has another 24-cell geometry sitting just inside it the entire geometry can turn itself inside out computationally. It is platonic in nature, meaning the distance between roots or chromosomes are always equal to each other, like a multidimensional number line, even when turned inside out. A 4D computational structure ably visualized in the movie Inception with the Penrose stairs. As the characters traverse the 4D staircase they could stop anytime to help pick up papers, which is equivalent to DNA repair and impossible in 3D. Lactose as energy sets things in motion because of the self trapping energy nature of the 24-cell. DNA mechanics and computation are based on 4D loops.
More of this theory predicted brain structure, in particular the thalamus and neocortex. The biological Davydov soliton becomes a robust qubit with the right cavity dynamics, which can be found at the base of a tetrahedron, or as you biotech guys might call pyramidal cells. The three choices vs. two of a transistor plus the interconnection between the different layers of pyramidal cells indicate we all have a quantum processor in our brain already.
The math behind this explains why Michael Levin can turn things inside out and get something new. Matrix mechanics. Flipping around the trace is turning the geometry inside out, is that part of collective intelligence? Type theory explains genetics as well as everything else. Homotopic type emergent discrete geometry expression, or HOT EDGE. Voevodsky type proofs in geometric form, generated by a single univalence description.
We get particles and elements and neurons the same way.
Artificial brains coming soon! Instead of virtual reality, real virtuality. Like real time surgery visualization. Things are going to get cracked soon enough.
I have actually met several of those. Like you mentioned, most developments take way more time than software because biology is inherently slow and chaotic than software development. There are no debuggers, no stackoverflow etc.
Cool article. I think you can’t really be cracked at biotech because so much of biotech is just research driven. Research is hard, there’s the dead ends but there’s also just the difficulty of the protocols. Lab automation helps, but to your point the system are very complex and slow.
I think we’ll see some cracked biotechnologies once’s we can engineer cells from scratch and aren’t just monkeying around the edges. But that’s science fiction for now. We’re decades or more away from deterministic whole cell engineering where the genetic transformations and customs printed organelle architecture are placed with atomic precision. I fear we were born in the wrong century.
I think I am your target audience in disposition (you are the first fellow slatestarcodex-reading bio person I've encountered, hello), though probably not aptitude. I was driven to go into experimental biology because I thought my enjoyment of math/cs problem-solving would translate to experimental methods development. The NGS, optogenetics, and cas9 waves were all huge when I was in college, and it seemed like that momentum would continue.
I worked at a metabolic engineering company for 3 years after college to get broader exposure, which was super eye-opening. There were ~50 PhDs there from all kinds of different areas, which was a huge contrast from the more focused nature of many academic labs.
While I enjoyed talking to the people in experimental biology, I didn't enjoy the experimental work itself. I found it pretty stressful and repetitive -- long hours, a single moment of inattention crushing a week of work, hard to translate one's skills to different roles... I wasn't super money-driven, but seeing random FAANG friends working flexibly and making 250k, conspired to get me out of experimental work. I didn't want to be 30 and not have employable computational skills.
I think the repetition of the work deters a lot of people. My swe friends would say that their work too, was repetitive, and complain about "protobuf plumbing." I felt they _fundamentally didn't understand_ how repetitive and time-consuming (and delicate!) doing westerns was just to collect a little bit of information.
I was further convinced not to pursue methods development when I saw a PhD-level job talk from a guy at Big Name Lab at our company whose PhD research had like... not fully worked out. Methods is high-risk high-reward. Interviewers had a positive impression of him and gave him an offer, but were skeptical of how his skills would translate to the company/role, and he also didn't have computational skills.
As you said, I think the lack of agency is another hard part. Since experiments are expensive, it's hard to work on your own ideas. I know lots of experimental grad students and post-docs who have only ever worked on PI's projects... another aspect of this is that since people are spending like ~50-60 hours a week *just doing the same lab protocols*, this leaves a lot less time for actually thinking. I think this probably colors the incentive structure of experimental vs computational fields.
Many people are repeating maybe a dozen protocols hundreds of times, not interacting a lot with people from people from other environments. I remember asking a protein engineering PhD how she thought some reaction worked, and she asked "well what does the active site look like?" It felt like a question out of left-field for me, because Actually Thinking About Structure Details felt like a totally different level of abstraction compared to what I was used to.
Tying into this is that many academic labs are pretty siloed. (I think the biology groups I've worked in have been some of the most intellectually vibrant / fun environments ever though, so not trying to knock everything.) But many people are spending 5-7+ years in a single lab with like <8 people, and conflate that with some kind of Broad View of Science and Society (not claiming to have this, lol). This also means that labs turnover more slowly than companies do, since people might churn in <2-3 years. So I think a company might give one more exposure to ideas, but the constraint of "this has to literally-actually make money in the immediate future" is quite different.
In industry there's a lot of people at the bench who have 20+ yoe. In academia... your PI doesn't pipette, and the most senior person in the lab might be like ~3rd year post-doc. I think this means there's like an entire arc of IC skill development that young academic scientists don't see.
At my old job, I would talk to people with ~20 yoe... every 10-15 minutes throughout the day, and we have a shared context and mission of what's going on. In an academic lab, you talk to your PI once a week-ish, and your labmates sort of know what you're doing, but not really. I felt like I learned _a ton_ from just off-hand technical convos throughout the day with people at the company, and found the academic model sort of isolating.
Another thing from working at a company you pick up is learning to sniff out when somebody with 20+ more years of experience than you... can still be wrong. If there's 50+ people with PhDs, you're going to have a lot of contrasting views, and you have a lot of experiences where Bob, who has a Genetics PhD and 20 years more experience than you, is trying to lecture you about something, but it turns out the other senior people disagree and your hunch of calling bs was right. I think this is relevant in grad school, because your PI is the only senior person you interact with, and while they know more than you, they're also a person with their own idiosyncrasies.
I think this is all super true, but at least from what I'm seeing a fair bit of this is getting overhauled thanks to automation, where the idea is you can actually 10x your output and not worry about "a single moment of inattention crushing a week of work". It opens a lot of other potential problems and costs but the trend is for sure away from pipetting things manually.
I'm sure automation will happen eventually, but people have been framing automation as the catch-all obvious solution for 10-15 years now, and haven't seen a big change in the landscape yet. There was a lot of automation/robots at the company I worked at, hasn't made it to academia in any significant amount yet. Not sure what the line is.
damn well said
Very, VERY interesting. Thank you for sharing.
Hey, man. Really liked your post and thought I'd reach out. I found especially interesting that you make the point that 10x engineers are born out of freely tinkering their own ideas around, learning what works n all until they have enough of a system understanding that they become cracked at it. And how you then mention that this 'attempt-result Cycle' doesn't really work for bio stuff, but you're anyway at the "top 1% in 'amount of fucking around and trying stuff'". I myself will be starting university for a major in physics later this year, and hopefully eventually work with some biophysics stuff or neuroscience. In any case, I wanted to ask you if you had any tips, or some important things you learned along the way (and the challenges involved in them!) when it comes to just experimenting with stuff and ideas that you'd want to share for someone that'll be just starting this path to excelence in the field. Would be very much appreciated!
Man I have no idea what that looks like in physics, it's really not my area. Biophysics best guess is to find some stuff in the simulation space that you can do on a computer, try to be playing with the latest toys and models.
As best you can try to hang around professors who seem to encourage playing around and trying long shot ideas, which might not necessarily be the ones with the best publication record.
From the outside, think you're probably on target here. I knew a kid who I think had the genius necessary to be a world class software guy, but switched lanes to biotech in college. Last I heard he was off in a lab somewhere in the Sacramento area messing with neuro-something-or-other. No big news coming out of there in intervening 10 years. Pretty confident he'd have been A Big Deal if he stuck to software, but in biotech he's just another faceless lab monkey. And it's fine, it's what he wanted to do - but - there is something about the field that seems to inhibit radical developments.
the 10x or 10^n x engineer simply doesn't exist. take that example: being in the situation where you get to create Google Maps from scratch. it's absurd in the context of modern product design.
You need to sync with so many people with so many different needs and synthesize that into a product with all the various testing (A/B, integration, unit, etc) that goes with it. This is a lot of work, most of which isn't all that technically complex in nature. Because fundamentally, software engineering's primary work is in understanding human needs; not the ability to write lots of code quickly or create crazy innovative solutions to NP-Hard problems. Writing the code is something you do after months of deliberation.
Sorry but lone geniuses aren't really that important, unless you're in a research context, and even then it's vastly overstated.
What's interesting is that a lot of the article is about the environment of the engineer more than the engineer itself.
While 10x engineers may not exist, there are situations that call for speed, efficiency and crazy creativity. It's ultimately up to the environment to either relax the constraints, or have none in the first place.
I feel like the idea of the 10x engineer is kind of crazy. But of course there are outliers in productivity or creativity!
But the real question underneath is whether your environment supports the flourishing of creativity and innovation or not.
I have thought about this before - why are there no FAANG style biotech that started in someone’s garage.
Biology is way more complex and just keeps getting more so with every new layer we uncover. But maybe AI can change that. Sadly I have more hope for AI breaking the mold than anything in academia shifting away from power grubbing advisors squashing the life right out of you.
I think there are academics whose labs are making big progress in biotech and churning out solutions to big problems frequently: George Church's lab keeps making foundational molecular tools that work making a lot more stuff possible. Jay Keasling's lab keeps making valuable new molecules biologically. Markus Covert's lab has made 'whole cell' computer models close enough to reality they've worked to discover numerous real, fundamental biological facts. Now, those guys aren't even really quite engineers, they're closer to CEOs managing a huge operation from a distance. However, unlike software, it takes big teams of people with diverse expertise working together to do this stuff. All 3 of the labs I mentioned above have somewhat hidden, nearly Carmack level software accomplishments that are enabling the wet lab results. Also, until maybe *right now* biotech was held back by 3 big obstacles: lack of predictive modeling/understanding, expense of DNA synthesis, and lack of flexible laboratory automation. Pretty much all 3 of these converged on low cost solutions very very recently, and I am seeing firsthand, as an academic in this field, tons of people finally making stuff work after decades of no progress.
I'm pretty cracked. You'll find people like that in the general direction of the Midwestern Doctor substack.
Re: DNA I think you don't see too much DNA stuff because 1) It doesn't work as presented I think and 2) mRNA has dominated the field financially.
what if the answer here that there were cracked biotechnologists, and now there aren’t?
And the fact that there aren’t has more to do with the complexity of the science and questions at hand.
There were singular Medical Doctors making leaping advances between 1860s and 1960.
It seems more likely that the facts of the science changed than that biotechnologist got worse at their jobs.
I’m guessing similar dynamics will be unfolding this century for CS and AI
I think there are lots of people in biology with aptitude, experience, and instincts as good as the hypothetical 'cracked 10x engineer' but, as you say, their productivity is hindered by time of execution and high complexity:
Biology vastly more complex and indirectly understood than most fields. Someone with elite instincts may see their hunches come up correct ~25% of the time vs 2.5% for others. If you can only fit 3 experiments into a week, it's still really hard to turn a 25% hit rate into huge productivity.
Theory driven biotech is cracked. There is a theory tying the double helix found in DNA to the double helix found in the 24-cell both with 48 roots for chromosomes. Because the 24-cell has another 24-cell geometry sitting just inside it the entire geometry can turn itself inside out computationally. It is platonic in nature, meaning the distance between roots or chromosomes are always equal to each other, like a multidimensional number line, even when turned inside out. A 4D computational structure ably visualized in the movie Inception with the Penrose stairs. As the characters traverse the 4D staircase they could stop anytime to help pick up papers, which is equivalent to DNA repair and impossible in 3D. Lactose as energy sets things in motion because of the self trapping energy nature of the 24-cell. DNA mechanics and computation are based on 4D loops.
More of this theory predicted brain structure, in particular the thalamus and neocortex. The biological Davydov soliton becomes a robust qubit with the right cavity dynamics, which can be found at the base of a tetrahedron, or as you biotech guys might call pyramidal cells. The three choices vs. two of a transistor plus the interconnection between the different layers of pyramidal cells indicate we all have a quantum processor in our brain already.
The math behind this explains why Michael Levin can turn things inside out and get something new. Matrix mechanics. Flipping around the trace is turning the geometry inside out, is that part of collective intelligence? Type theory explains genetics as well as everything else. Homotopic type emergent discrete geometry expression, or HOT EDGE. Voevodsky type proofs in geometric form, generated by a single univalence description.
We get particles and elements and neurons the same way.
Artificial brains coming soon! Instead of virtual reality, real virtuality. Like real time surgery visualization. Things are going to get cracked soon enough.
Don’t give up on theory yet! Decades my ass.
I have actually met several of those. Like you mentioned, most developments take way more time than software because biology is inherently slow and chaotic than software development. There are no debuggers, no stackoverflow etc.
I think "cracked" biotechnologists just don't make the news as much...George Church, James Craig Venter, and Jennifer Doudna all come to mind.
I could do with a few cracked neuroscientists, quantum biologists and maybe a cracked philosopher. On a quest to build something.
Cool article. I think you can’t really be cracked at biotech because so much of biotech is just research driven. Research is hard, there’s the dead ends but there’s also just the difficulty of the protocols. Lab automation helps, but to your point the system are very complex and slow.
I think we’ll see some cracked biotechnologies once’s we can engineer cells from scratch and aren’t just monkeying around the edges. But that’s science fiction for now. We’re decades or more away from deterministic whole cell engineering where the genetic transformations and customs printed organelle architecture are placed with atomic precision. I fear we were born in the wrong century.
Agreed, and all I need is a pretty small seed round to make damn sure it happens. https://youtu.be/XmHbVW93Jck?si=-Fcdq92qcz06rBLa
Terrific read. Looking forward to your future posts!