"The paper goes on to make recommendations for AI companies and investigators, and these are well worth reading. The common theme is that people need to think more about why they’re doing certain techniques or using certain technologies, rather than just using them because they’re newly available."
Please. Please let some people with power and influence understand this lesson sooner rather than later. I understand the reasons that's unlikely to occur, but usually the impact isn't quite so drastic and expensive as this is. Just because something is new and shiny doesn't mean that it'll produce the outcomes you need at the other end, and until it's shown that capability your approach to it should be MODERATE.
Drug discovery scientists think about what they're doing and why ALL THE TIME. AI stuff is just another tool.
It's also worth mentioning that drug development timelines typically exceed the interval in which these technologies have been available (or at least effective). Measuring impact will take a long time.
For everyone struggling with hair loss but concerned about side effects of Fin. Look into topical fin. It produces higher concentrations of the medication in the scalp and lower systemic concentration. Not perfect, but better.
there is a AI designed drug for hair loss that i know of.
its slow-release oral minoxidil formulation called MINX. AI helped with the formulation [1].
its in in similar category as VDPHL01. Hundreds of millions if not a billion dollars has been invested into Veradermics, and their main product is VDPHL01 (also an extended-release oral formulation).
I’ve taken low dose oral min for a few years and it’s systemic, meaning it’ll make all your hair grow. I now have body hair where before it was never noticeable. “I’m hairy like monkey” as my kids say.
Can confirm. My torso and upper arms be sprouting like Robin Williams. Wasn’t like this before. Thankfully that’s easy to buzz off, and worth it to keep my head.
if you think the state of the art in this area is something you'll hear about from a guy that looks like santa in an academic journal, your investments deserve what's about to happen to them.
nerd-fanboi proposal: Articles by national treasures (like Derek Lowe, Raymond Chen) should be highlighted with specific identifiers on HN - like a distinctive title font or an ascii diamond ◊.
Yeah it's [x was never the hard part] all the way down. It's a very human thing and I bet we'd keep saying that even after we've solved the hardest mysteries including consciousness, the origin of life or the true nature of reality.
So is ketamine but we haven't explored it more for other brain functions where it can act as a psychoplastogen eg reopening critical periods of visual learning.
Sure, but what kind of protocol are you going to use to actually reap the benefits of these transient neuroplasticity effects, especially for learning something new like math or piano as an adult?
Same with ketamine. It's not like you can take it a couple of times and open some magic window where everything suddenly becomes easier to learn. From my personal experience and surface-level understanding of the current research, psychedelics(including ketamine) seem to temporarily relax hardened beliefs/priors and rigid neural pathways, which can help you see things from a fresh perspective. But learning something substantial like math or a new language after you've passed the most plastic stages of development is still going to take much longer than what these drugs can realistically help with.
They might make you see something differently or even get you extremely interested in it, but sustaining that interest and consolidating the knowledge or skill is still slow compared with childhood/adolescence. Of course, it depends a lot on what you're learning. For example, crystallized intelligence and sufficient motivation can make some things much easier to learn as an adult, but many useful things are also just boring to learn when you no longer have a childlike plastic brain or an environment built around constant learning.
Edit: Theoretically, you could accelerate learning by taking psychedelics/ketamine at a set frequency but it's a huge gamble because of their risk profile. eg. HPPD/trauma risk with classic psychedelics and bladder/neurotoxicity risk with ketamine if you get addicted or take it too frequently.
I don't know. I was just suggesting something that may have an effect. BTW I would be very hesitant to take anything that "opens some magic window where everything suddenly becomes easier to learn" if it existed. Increasing neuroplasticity to that degree is unlikely to be without significant side effects.
Derek Lowe as a science communicator, and others like him, is sorely needed to understand the real meaning and significance of the study and others. I say that as someone with a PhD in chemistry who's been to plenty of presentations on drug discovery topics.
It's difficult to calibrate statements made by other scientists unless you're well embedded within a field: Is this someone whose opinions matter? Are they the subject matter expert they make themselves out to be? Is this research itself truly impactful? Is it really 5 years until it will be realized outside of academic labs? Etc...
It's difficult to decipher questions around credibility because they rely on real-world interactions and associations that extend beyond the physical tokens of paper counts, publication venues, citations, and author lists that typically lag behind the front of human knowledge which is generated from real-world interactions. It can be simple things, like the insightful question a grad student, with minimal publication history, asks in a seminar.
Of course, the paywall is also unhelpful too, but a good, brief commentary by an appropriate commentator is a better link for 99% of prospective readers compared to most "peer reviewed" (scare quotes because that's a real question nowadays) articles.
Not that I'm a Derek Lowe fanboy or anything but the entirety of your comment is like "that guy needs to check himself" without, like, any specific context, any specific argument he's wrong on this specific question or like anything. It's like "deciphering questions around credibility" is hard ... all the way down. Where's yours? What are you saying?
Isn't the comment in fact praising Derek Lowe as a science communicator? In the second paragraph OP is just posing the questions that one might have when reading about a field not your own, that highlight the importance of reliable science communicators.
It's a very confusing comment since Derek Lowe is a chemist working in pharma but is being referred to only as a science communicator which I suspect most people would consider to be an implicit insult.
I think it says more about the person reading it as an implicit insult than it does OP. I am, I think, a fairly competent scientist, and know many great scientists. I know very few great scientific communicators.
Derek Lowe is someone whom I greatly admire and his work as a science communicator is a true service to the profession and to science at large. It is not intended as an insult in any way. I've known a great number of competent chemists, but few are as effective as communicators as he is.
My apologies for the lack of clarity in my prior comment; it was composed in haste. To be clear, I am a bit of a Derek Lowe fanboy. I value the work that he does in offering insight on his areas of expertise to the public at large because he has both the necessary understanding and the ability to communicate clearly. Few people have his level of background knowledge alongside the capacity to wade into the depths of pharmaceutical chemistry and then communicate that information and his personal perspective to enrich the commons.
What I was getting at with the issue of credibility is that I (or most others) would have time determining if a particular scholarly author in pharma is a good person to listen to on the subject. But having evidence that Derek Lowe is an honest broker of information in that domain, given his long history of commentary as well as other chemists I know who eagerly read his commentary, I'm willing to defer to his assessment.
The world needs more communicators like Derek Lowe who can give a clear-eyed, de-bullshittified perspective on the world, while also being great writers who are a pleasure to read. I follow his blog, and while a lot of it goes over my head (or seems very niche), a lot is great. The only other such writers I know about is Matt Levine at Bloomberg, and Atul Gawande (who's a bit less blunt, perhaps). And of course the late Carl Sagan. Any others?
Professor Heather Cox Richardson (US History) has a blunt near daily summary with frequent call backs to significant high and low points of the past two and a half centuries:
OP here: the title of the thread still links to Derek Lowe's blog post, but the article discussed in the blog post was added to the body of the original post (not by me).
NB: The post was updated twice. The first changed the link to the Nature Reviews Drug Discovery article; the second changed it back to Derek Lowe's blog post with the NRDD article in the body.
I'm a structural biologist at a mid-sized biotech. I use AI tools daily. They make accomplishing the same things I was able to accomplish before quite a lot faster and easier. They don't help me magically accomplish new things that I couldn't previously.
For example, it helps me install academic software, debug things. It helps me take a large dataset and write scripts to ask questions. It helps me go through experiment drafts to see if I'm missing things. It helps me remember obscure formulas I use every 6 months. It has not, at least in my experience, come up with anything truly novel.
A concrete example: AlphaFold is great...to come up with a starting model for a chimeric fusion or something. What would have taken me 1-2 hours fumbling around in PDB or CIF files is now a quick prompt.
In some ways, but it's tough to say if that's my ADHD or not. It's far too easy to leave one branch of reasoning now and jump to another whenever progress gets difficult.
Though in some areas where I can sustain interest, AI is helping me go deeper. For instance, I've been putting myself to sleep at night by just asking it questions about expectation maximization and Bayesian statistics. This has seriously boosted my understanding of cryo-EM alignment algorithms in a way I couldn't do in grad school because there was no professor that understood enough to help me when I got stuck reading literature.
In some ways, but it's tough to say if that's my ADHD or not. It's far too easy to leave one branch of reasoning now and jump to another whenever progress gets difficult.
I have a co-worker who doesnt feel like ADHD helps him because he sits down and just starts... doing work and typing. Assign him a complex task, he will just start on it. Mind blowing he does this day in and day out. an absolute machine.
The old expense of installing new tools means that seeing how tools actually work in practice is simply impossible. You pretty much install a tool and you accept that it is the tool.
You have to know how to ask the right questions, but I think there's constant opportunities in anything computer-driven to say "this thing is taking 3 hours and I'd like to do it more often, can I do it in 3 minutes?" And that's always been true to an extent but AI makes it way more accessible.
Yes, and it's relatively good for on-rails data collection and data processing pipelines that would have previously needed occasional human intervention. Especially now that I can run something like DeepSeek v4 Flash on a couple of RTX 6000s and just script an API to hammer away without having to worry about racking up a huge bill.
For example, it helps... debug things... and write scripts to ask questions.
Im guessing this isnt code that needs to "scale", that needs to "be elegant", that you arent focused on maintainability for the next decade. That its built for purpose and left behind.
Its all the code that for a programer would normally be in this matrix https://xkcd.com/1205/ (is it worth your time) -
Yeah, for someone like me in a lot of ways it's a dream come true. I have domain expertise in my field, I have a ton of computer knowledge due to building HPC clusters and working in Linux every day for years, I have lots of problems computers can solve, but no formal programming expertise. Before, muddling through Python or fixing segfaults in academic software would have been days to weeks. Now it's minutes.
I think that (understandably) HN is full of professional programmers for whom code is the product, and so LLMs are often viewed through that lens. But for someone like me, a drug is the product, not code. One off vibe coded slop is both fine, and often an upgrade over the academic software I was using.
For instance, last week I took a piece of software that decompresses a TIFF file and multithreaded it for an almost 4x speedup. I'd always known it was single threaded and it irked me because I could see my pipelines waiting for it to catch up, but I never had the expertise in C++ to go fix it. Claude did it in 30 minutes and I didn't even have to go through the hassle of compiling it again - it did that too.
What I don't know is, if someone wasn't in the trenches for a decade learning how computers work, would the results be as good?
I don’t think the results would be as good if you just prompted “make this faster” as compared to “make this multithreaded”. You had high confidence in the cause of the slow down which a non-technical person might not have.
The lack of comparable data and testability really does seem to be a challenge. I wonder if people would be more willing to collect and share lots of health data if the collecting company was a non-profit dedicated to anonymizing it.
C) not smart enough to be a self-taught bio-hacker
Everyone hears "AI is going to cure disease" and pictures some cure-all pill from a bio lab which is what I feel this paper is hinting at is missingb but that's the top of the funnel; I'm at the bottom where patients live and that is where AI is already quietly working. Its just not being benchmarked.
I built https://crohns.ai. I set out to make an AI-native clinical-trial manager with a feedback loop (DDP) and ended up somewhere completely different: instead of chasing a new "drug" which is totally out of my grasp; financially, intellectually etc... I used it to codify a care protocol that helped me avoid a flare after I got laid off, lost my insurance, and lost access to Skyrizi.
If given the chance, I might go back on it because my protocol can be a little strict at times but either way I do see a significant shift to tools like this given the state of the US Healthcare system.
cool. i like what you've done here with a virtual panel that can answer questions ... i have similar ideas for kabuki syndrome (currently: https://www.thekabukipapers.org). talk? marstall at gmail.
i like what you've done here with a virtual panel that can answer questions
Thats the main thing. I think that Ai-native governance by domain experts is how AI reaches its full potential. Not in theory but in terms of the value it delivers to populations via outcomes.
AI-native Governance is like irrigation for the outcomes populations want to achieve, starts with intents; executed on by programs that use protocols as guardrails. This is a gross oversimplification of the process but based what i see from your work you will get the abstract.
Each codify.* is a PDA [Public Domain Agent] - that gets delegated intents per request and has to manage its own "deal" - its also managed democratically via ontology and downline policies set by the experts in said field and has feedback loop to verify/optimize policy outcomes.
This concept applies to everything IMO, and I cannot say I fully understand it but im absolutely obsessed with the exploration of the idea; again - in practice not theory. I have real outcomes in healthcare, education and housing.
how do you define the specialists? a prompt? what do you mean by using protocols as guardrails? Also, can you give me a TLDR on how these voice interfaces work? What kinds of problems are you solving, and how, in an ELI5 way. I might be a lot dumber than you give me credit for!
“…the focus of AI in drug discovery must shift from doing what can be done - such as modelling data that is readily available, but that is unlikely to move the needle - to doing what should be done, even if this requires, for example, substantial data generation…” It’s a worthy goal, but I think that many involved in this work might be thinking, even unconsciously, “You first”.
This is the problem with AI for all of science - not just drug discovery. Applied ML has spread like wildfire through academia over the past decade - this started well before the LLM hype. It’s the perfect honey trap: research is painstaking and slow, ML offered a shortcut, and best of all, it just needs data. Research produces lots and lots of data! Surely this will be a match made in heaven.
I’ve watched the same pattern play out at least four or five times now in various roles.
(1) Propose an ML-guided approach to material/chemistry discovery/optimization.
(2) Gather existing data (real, experimental data).
(3) Realize there’s less than about 50 true rows of data on the outputs of interest.
At this point, you either:
(4a) revert to traditional methods but keep the veneer of using ML to save face, or
(4b) pivot to computational/simulation work or a high-throughput system that’s very far removed from your original problem, but allows you to keep playing with ML toys
It’s really bad. I left the industry. I don’t know how long it will take for people doing real science to take back the reins (and the funding).
In chemicals and materials, 50 rows of good data is a really solid study. That’s e.g. a 3x4x4 experimental design (assuming replicates for each condition get averaged into a single row). If you managed to prep that many samples correctly and obtain consistent characterization data across all properties of interest, you’ve easily got a paper. It’s also kind of malpractice to jam this type of data (few samples, wide rows) into modern ML models. There are plenty of simpler statistical methods that will tell you what’s going on, and even then a well-made plot might be good enough. The difficulty is not in drawing insight from the final numbers, it’s almost always in how those numbers came to be in the first place.
Thus the reticence of science-oriented companies to invest heavily in these mass data-gathering exercises to feed ML. It’s damn expensive, and almost always leads you back to raw data issues, not breakthrough discovery. Doing it without a set purpose in mind is even more likely to yield garbage.
I worked on some of the very best funded plant research out there. When it comes down to it, there's enough variation caused by confounding factors, and it takes so long to capture more data, that almost everything anyone tries cannot be called a success or a failure for years, because the individual measurements for one small plot of land somewhere just don't mean anything. Once you do an entire experiment for a season, which takes months, and you grab the little noisy data you have, and turn it into real rows, we were down to very little.
You can do more tests on smaller things, like checking if some protein will kill some cells of a pest, but making sure a plant produces it enough that it actually does something significant to the real, live pests, that it's not toxic, and it doesn't harm the plant's yield massively (as it's now spending time producing your pesticide) is still going to take years. We might be able to fold proteins, but the kind of things we'd need to really simulate plant biology well enough to not need years of failures are still very far away.
And it's far worse in medicine, as with plants at least nobody has ethical concerns if they fail and die, and nobody needs to get consent from a corn seed. Getting to 50 actual data points from many medical studies is already a lot of effort. And imagine when it's a long term study, and you need to follow patients for 30 years, as theym move, or die, or decide to stop participating, or who knows what.
Here is an article by Pat Walters on the usefulness of ML in drug discovery. This article is a response to another one making the case that utility of ML models are very limited in drug discovery
(4a) revert to traditional methods but keep the veneer of using ML to save face
I haven't worked in the industry side of things but in academia everyone kind of agrees that gradient boosting trees are some of the best models to do these things.
The real value right now is in figuring out how to generate robust data cheaply and quickly. I'd wager that the effect of a good model on marginal data is small, but the effect of a marginal model on great data is probably quite large.
The obvious question is what limits getting more data? Astronomy (especially in Australia) has been quite good at designing surveys to answer multiple scientific questions with reasonable amounts of data (and then fed into ML systems like the cannon). Sadly one of the consequences of the LLM hype is the increasing cost of doing this, so "AI" is actually making things worse not better.
It was always easy to come up with new molecules/drugs/materials. The thing that changed is the scale that it can happen with the new ml-based approaches.
What hasn’t changed is finding ones that are manufacturable/synthesizable.
Even if you find 1 million new stable molecules, there no guarantee that even one of them is manufacturable.
Yes indeed, I call it the revenge of statistics! Pseudoreplication rears its ugly head once again. Nonparametrics are certainly useful, but once you know things about the DGP, it’s often possible to do something “traditional” that’s better. Probabilistic programming, for life!
Obviously the missing part (which we already have for software and math) is that we need agents to be able to run automated loops in the real world. That basically requires robots. I think we'll be there in less than 5 years.
clinically relevant impact is, so far, disappointingly limited
it could be that the AI tools have to get to some threshold before they are very useful? Like with the Economist talking to Hassabis:
AlphaFold itself took six years of work to predict its first protein structure, and then one year to follow up with what he describes as the structures of “all 200m proteins known to science”. He hopes a similar speedup will happen inside Isomorphic.
What I increasingly see at my Biotech is more and more AI assisted drug candidates but no corresponding improvement in the capacity or ability to manufacture and scale them. I think it's a problem that groups like Anthropic will encounter in a few months/years and one that will annoye them a lot because it's not as simple as throwing more compute, people or money at it.
Please. Please let some people with power and influence understand this lesson sooner rather than later. I understand the reasons that's unlikely to occur, but usually the impact isn't quite so drastic and expensive as this is. Just because something is new and shiny doesn't mean that it'll produce the outcomes you need at the other end, and until it's shown that capability your approach to it should be MODERATE.
Drug discovery scientists think about what they're doing and why ALL THE TIME. AI stuff is just another tool.
It's also worth mentioning that drug development timelines typically exceed the interval in which these technologies have been available (or at least effective). Measuring impact will take a long time.
Need one for hair loss ASAP.
Already exists (finasteride), only problem is that it castrates you chemicallym
Yep. Not everybody but enough to warrant vigilant monitoring and titration.
For everyone struggling with hair loss but concerned about side effects of Fin. Look into topical fin. It produces higher concentrations of the medication in the scalp and lower systemic concentration. Not perfect, but better.
About 2% of finasteride users experience these side effects, and they are reversible after discontinuation.
Apparently some users report persistent side effects even years after stopping the medication (post-finasteride syndrome).
yep that would be me
Are you familiar with Dr. Powers, there's hope
not familiar tbh, i will check it out sometime. thanks
How many of them can be accounted for by the natural rate of sexual dysfunction?
I've been taking it without significant side effects for ~15 years, so I'm not worried, although at this point it's losing its main effect as well.
Indeed. My plans for a youthful Mohawk are being stymied by the lack of AI promised medical breakthroughs.
Wheres my follicles dammit?
Absci has a good asset coming out pretty soon that you could try to get on the trial.
Absci is bullshit
there is a AI designed drug for hair loss that i know of.
its slow-release oral minoxidil formulation called MINX. AI helped with the formulation [1].
its in in similar category as VDPHL01. Hundreds of millions if not a billion dollars has been invested into Veradermics, and their main product is VDPHL01 (also an extended-release oral formulation).
[1] https://x.com/anagenxyz/status/2071601868841595082
I’m on 0.5 to 1mg oral minoxidil daily for a few years now and it’s working great. Blood pressure benefits too.
Any other side effects? I’ve never even heard of this.
I’ve taken low dose oral min for a few years and it’s systemic, meaning it’ll make all your hair grow. I now have body hair where before it was never noticeable. “I’m hairy like monkey” as my kids say.
Can confirm. My torso and upper arms be sprouting like Robin Williams. Wasn’t like this before. Thankfully that’s easy to buzz off, and worth it to keep my head.
Speak for yourself, if mine goes any lower I’m going to pass out
A good set of clippers.
Track KX-826, clascoterone, and VDPHL01
PP-405 too
We're all about to come face to face with this reality. This dance can only last so long.
if you think the state of the art in this area is something you'll hear about from a guy that looks like santa in an academic journal, your investments deserve what's about to happen to them.
Only for values of 'all' that exclude well-connected members of the billionaire class and their select associates.
nerd-fanboi proposal: Articles by national treasures (like Derek Lowe, Raymond Chen) should be highlighted with specific identifiers on HN - like a distinctive title font or an ascii diamond ◊.
No thanks. Social media needs less hero worship. Just RSS whoever you like.
I agree. I won't do heroes. Ever.
A National Treasure, on the other hand - they enrich life without being a vector for tribalism (eg:Michael Kramer/Kate Reading).
Stop calling them "National Treasures." That's hero worshipping. That's just sad.
Make the feature a per user list of flagged sources.
I would have agreed in past, but the new random "blog" at whatever.etc being written by LLMs is making me skip clicking most links here now.
[AI drug discovery] was never the hard part.
This made me laugh, more than expected, but I did visit LinkedIn just before so that could explain it. Thanks!
Yeah it's [x was never the hard part] all the way down. It's a very human thing and I bet we'd keep saying that even after we've solved the hardest mysteries including consciousness, the origin of life or the true nature of reality.
Somehow reminded me of:
THERE IS INSUFFICIENT DATA TO ANSWER THE QUESTION.
need one for brain plasicity. it would be nice to be able to easily learn a foreign language or musical instrument naturally.
There is one!
Ketamin should have huge impacts on neuro/brain plasticity when used properly (i.e. in therapy)
Psychedelics are several orders of magnitude stronger on the plasticity front.
In therapy as well.
So is ketamine but we haven't explored it more for other brain functions where it can act as a psychoplastogen eg reopening critical periods of visual learning.
Psilocybin has this effect. Source: I can't remember where I read it, so low confidence.
Sure, but what kind of protocol are you going to use to actually reap the benefits of these transient neuroplasticity effects, especially for learning something new like math or piano as an adult?
Same with ketamine. It's not like you can take it a couple of times and open some magic window where everything suddenly becomes easier to learn. From my personal experience and surface-level understanding of the current research, psychedelics(including ketamine) seem to temporarily relax hardened beliefs/priors and rigid neural pathways, which can help you see things from a fresh perspective. But learning something substantial like math or a new language after you've passed the most plastic stages of development is still going to take much longer than what these drugs can realistically help with.
They might make you see something differently or even get you extremely interested in it, but sustaining that interest and consolidating the knowledge or skill is still slow compared with childhood/adolescence. Of course, it depends a lot on what you're learning. For example, crystallized intelligence and sufficient motivation can make some things much easier to learn as an adult, but many useful things are also just boring to learn when you no longer have a childlike plastic brain or an environment built around constant learning.
Edit: Theoretically, you could accelerate learning by taking psychedelics/ketamine at a set frequency but it's a huge gamble because of their risk profile. eg. HPPD/trauma risk with classic psychedelics and bladder/neurotoxicity risk with ketamine if you get addicted or take it too frequently.
I don't know. I was just suggesting something that may have an effect. BTW I would be very hesitant to take anything that "opens some magic window where everything suddenly becomes easier to learn" if it existed. Increasing neuroplasticity to that degree is unlikely to be without significant side effects.
Easily? Have you ever actually learned a second language or an instrument as a child?
Vs as an adult. Yes it's hard as a kid, but as an adult it's even harder.
Has this yet produced a treatment for AI psychosis?
No? Well fancy that! :)
Derek Lowe discusion of the paper https://www.science.org/content/blog-post/so-how-ai-drug-dis...
I think that was originally linked but got changed to the £30 to Elsevier version for some reason.
Derek Lowe as a science communicator, and others like him, is sorely needed to understand the real meaning and significance of the study and others. I say that as someone with a PhD in chemistry who's been to plenty of presentations on drug discovery topics.
It's difficult to calibrate statements made by other scientists unless you're well embedded within a field: Is this someone whose opinions matter? Are they the subject matter expert they make themselves out to be? Is this research itself truly impactful? Is it really 5 years until it will be realized outside of academic labs? Etc...
It's difficult to decipher questions around credibility because they rely on real-world interactions and associations that extend beyond the physical tokens of paper counts, publication venues, citations, and author lists that typically lag behind the front of human knowledge which is generated from real-world interactions. It can be simple things, like the insightful question a grad student, with minimal publication history, asks in a seminar.
Of course, the paywall is also unhelpful too, but a good, brief commentary by an appropriate commentator is a better link for 99% of prospective readers compared to most "peer reviewed" (scare quotes because that's a real question nowadays) articles.
Not that I'm a Derek Lowe fanboy or anything but the entirety of your comment is like "that guy needs to check himself" without, like, any specific context, any specific argument he's wrong on this specific question or like anything. It's like "deciphering questions around credibility" is hard ... all the way down. Where's yours? What are you saying?
Isn't the comment in fact praising Derek Lowe as a science communicator? In the second paragraph OP is just posing the questions that one might have when reading about a field not your own, that highlight the importance of reliable science communicators.
It's a very confusing comment since Derek Lowe is a chemist working in pharma but is being referred to only as a science communicator which I suspect most people would consider to be an implicit insult.
I think it says more about the person reading it as an implicit insult than it does OP. I am, I think, a fairly competent scientist, and know many great scientists. I know very few great scientific communicators.
Derek Lowe is someone whom I greatly admire and his work as a science communicator is a true service to the profession and to science at large. It is not intended as an insult in any way. I've known a great number of competent chemists, but few are as effective as communicators as he is.
My apologies for the lack of clarity in my prior comment; it was composed in haste. To be clear, I am a bit of a Derek Lowe fanboy. I value the work that he does in offering insight on his areas of expertise to the public at large because he has both the necessary understanding and the ability to communicate clearly. Few people have his level of background knowledge alongside the capacity to wade into the depths of pharmaceutical chemistry and then communicate that information and his personal perspective to enrich the commons.
What I was getting at with the issue of credibility is that I (or most others) would have time determining if a particular scholarly author in pharma is a good person to listen to on the subject. But having evidence that Derek Lowe is an honest broker of information in that domain, given his long history of commentary as well as other chemists I know who eagerly read his commentary, I'm willing to defer to his assessment.
The world needs more communicators like Derek Lowe who can give a clear-eyed, de-bullshittified perspective on the world, while also being great writers who are a pleasure to read. I follow his blog, and while a lot of it goes over my head (or seems very niche), a lot is great. The only other such writers I know about is Matt Levine at Bloomberg, and Atul Gawande (who's a bit less blunt, perhaps). And of course the late Carl Sagan. Any others?
Professor Heather Cox Richardson (US History) has a blunt near daily summary with frequent call backs to significant high and low points of the past two and a half centuries:
eg: August 15, 2026 - https://heathercoxrichardson.substack.com/p/august-15-2026
* https://en.wikipedia.org/wiki/Heather_Cox_Richardson
OP here: the title of the thread still links to Derek Lowe's blog post, but the article discussed in the blog post was added to the body of the original post (not by me).
NB: The post was updated twice. The first changed the link to the Nature Reviews Drug Discovery article; the second changed it back to Derek Lowe's blog post with the NRDD article in the body.
I'm a structural biologist at a mid-sized biotech. I use AI tools daily. They make accomplishing the same things I was able to accomplish before quite a lot faster and easier. They don't help me magically accomplish new things that I couldn't previously.
For example, it helps me install academic software, debug things. It helps me take a large dataset and write scripts to ask questions. It helps me go through experiment drafts to see if I'm missing things. It helps me remember obscure formulas I use every 6 months. It has not, at least in my experience, come up with anything truly novel.
A concrete example: AlphaFold is great...to come up with a starting model for a chimeric fusion or something. What would have taken me 1-2 hours fumbling around in PDB or CIF files is now a quick prompt.
do you feel this is the same trade off of UI builders like android studio (or msvb6). you do in minutes what you previously did in 2, 3 hours.
is it all the work? no, but it's a part that's early on and have high perceived impact.
then, as you progress, that tool actually gets in the way and a new feature that would take 2 hours, now is around 2 days.
In some ways, but it's tough to say if that's my ADHD or not. It's far too easy to leave one branch of reasoning now and jump to another whenever progress gets difficult.
Though in some areas where I can sustain interest, AI is helping me go deeper. For instance, I've been putting myself to sleep at night by just asking it questions about expectation maximization and Bayesian statistics. This has seriously boosted my understanding of cryo-EM alignment algorithms in a way I couldn't do in grad school because there was no professor that understood enough to help me when I got stuck reading literature.
So it's a double edged sword for sure.
I have a co-worker who doesnt feel like ADHD helps him because he sits down and just starts... doing work and typing. Assign him a complex task, he will just start on it. Mind blowing he does this day in and day out. an absolute machine.
The old expense of installing new tools means that seeing how tools actually work in practice is simply impossible. You pretty much install a tool and you accept that it is the tool.
You have to know how to ask the right questions, but I think there's constant opportunities in anything computer-driven to say "this thing is taking 3 hours and I'd like to do it more often, can I do it in 3 minutes?" And that's always been true to an extent but AI makes it way more accessible.
Serious question - have you tried applying AI tools to more of your job, and in a goal-seeking fashion? Have you hit roadblocks?
Yes, and it's relatively good for on-rails data collection and data processing pipelines that would have previously needed occasional human intervention. Especially now that I can run something like DeepSeek v4 Flash on a couple of RTX 6000s and just script an API to hammer away without having to worry about racking up a huge bill.
Im guessing this isnt code that needs to "scale", that needs to "be elegant", that you arent focused on maintainability for the next decade. That its built for purpose and left behind.
Its all the code that for a programer would normally be in this matrix https://xkcd.com/1205/ (is it worth your time) -
Yeah, for someone like me in a lot of ways it's a dream come true. I have domain expertise in my field, I have a ton of computer knowledge due to building HPC clusters and working in Linux every day for years, I have lots of problems computers can solve, but no formal programming expertise. Before, muddling through Python or fixing segfaults in academic software would have been days to weeks. Now it's minutes.
I think that (understandably) HN is full of professional programmers for whom code is the product, and so LLMs are often viewed through that lens. But for someone like me, a drug is the product, not code. One off vibe coded slop is both fine, and often an upgrade over the academic software I was using.
For instance, last week I took a piece of software that decompresses a TIFF file and multithreaded it for an almost 4x speedup. I'd always known it was single threaded and it irked me because I could see my pipelines waiting for it to catch up, but I never had the expertise in C++ to go fix it. Claude did it in 30 minutes and I didn't even have to go through the hassle of compiling it again - it did that too.
What I don't know is, if someone wasn't in the trenches for a decade learning how computers work, would the results be as good?
I don’t think the results would be as good if you just prompted “make this faster” as compared to “make this multithreaded”. You had high confidence in the cause of the slow down which a non-technical person might not have.
How refreshing was this article vs. all the slop?
The lack of comparable data and testability really does seem to be a challenge. I wonder if people would be more willing to collect and share lots of health data if the collecting company was a non-profit dedicated to anonymizing it.
I think the real win here is for idiots like me:
A) no education
B) no resources
C) not smart enough to be a self-taught bio-hacker
Everyone hears "AI is going to cure disease" and pictures some cure-all pill from a bio lab which is what I feel this paper is hinting at is missingb but that's the top of the funnel; I'm at the bottom where patients live and that is where AI is already quietly working. Its just not being benchmarked.
I built https://crohns.ai. I set out to make an AI-native clinical-trial manager with a feedback loop (DDP) and ended up somewhere completely different: instead of chasing a new "drug" which is totally out of my grasp; financially, intellectually etc... I used it to codify a care protocol that helped me avoid a flare after I got laid off, lost my insurance, and lost access to Skyrizi.
How are those biologics? Did you have to visit the doctor to get injections frequently?
Hands down the best drug I have been on EVER; but its 11k a month with no insurance.
The 1st 2 injections where done by a nurse that came to my home, the others were done as self injections using their njection kits.
Ironically, now I have several people that are on it tracking their infusions etc...
Intent: https://wiki.crohns.ai/agent/posts/ibd-biologic-switch-decis...
Program: https://crohns.ai/program/71168-biologic-therapy-initiation
Protocol: https://crohns.ai/protocol/71168
If given the chance, I might go back on it because my protocol can be a little strict at times but either way I do see a significant shift to tools like this given the state of the US Healthcare system.
cool. i like what you've done here with a virtual panel that can answer questions ... i have similar ideas for kabuki syndrome (currently: https://www.thekabukipapers.org). talk? marstall at gmail.
Thats the main thing. I think that Ai-native governance by domain experts is how AI reaches its full potential. Not in theory but in terms of the value it delivers to populations via outcomes.
AI-native Governance is like irrigation for the outcomes populations want to achieve, starts with intents; executed on by programs that use protocols as guardrails. This is a gross oversimplification of the process but based what i see from your work you will get the abstract.
Examples -
dietmanager.com - RDN governance
crohns.ai - AGA (MD/GI) governance
<city>.us.codify.city - City council Governance
Even applies to YC: https://openyc.org
Each codify.* is a PDA [Public Domain Agent] - that gets delegated intents per request and has to manage its own "deal" - its also managed democratically via ontology and downline policies set by the experts in said field and has feedback loop to verify/optimize policy outcomes.
This concept applies to everything IMO, and I cannot say I fully understand it but im absolutely obsessed with the exploration of the idea; again - in practice not theory. I have real outcomes in healthcare, education and housing.
I just fixed Crohn's AI council, it's now all members of the AGA but i'm having serious issues keeping it all up.
The jobs to harvest the corpus needed to create each agent is not scaling well.
Note: crohns.ai: 3,948 gastroenterologists, dietmanager.com: 1,752 RDNs
how do you define the specialists? a prompt? what do you mean by using protocols as guardrails? Also, can you give me a TLDR on how these voice interfaces work? What kinds of problems are you solving, and how, in an ELI5 way. I might be a lot dumber than you give me credit for!
This is the problem with AI for all of science - not just drug discovery. Applied ML has spread like wildfire through academia over the past decade - this started well before the LLM hype. It’s the perfect honey trap: research is painstaking and slow, ML offered a shortcut, and best of all, it just needs data. Research produces lots and lots of data! Surely this will be a match made in heaven.
I’ve watched the same pattern play out at least four or five times now in various roles.
(1) Propose an ML-guided approach to material/chemistry discovery/optimization.
(2) Gather existing data (real, experimental data).
(3) Realize there’s less than about 50 true rows of data on the outputs of interest.
At this point, you either: (4a) revert to traditional methods but keep the veneer of using ML to save face, or (4b) pivot to computational/simulation work or a high-throughput system that’s very far removed from your original problem, but allows you to keep playing with ML toys
It’s really bad. I left the industry. I don’t know how long it will take for people doing real science to take back the reins (and the funding).
(3) seems like a problem in its own right? Basing science, traditional or newfangled ML, on such small amounts of data looks pretty weak.
In chemicals and materials, 50 rows of good data is a really solid study. That’s e.g. a 3x4x4 experimental design (assuming replicates for each condition get averaged into a single row). If you managed to prep that many samples correctly and obtain consistent characterization data across all properties of interest, you’ve easily got a paper. It’s also kind of malpractice to jam this type of data (few samples, wide rows) into modern ML models. There are plenty of simpler statistical methods that will tell you what’s going on, and even then a well-made plot might be good enough. The difficulty is not in drawing insight from the final numbers, it’s almost always in how those numbers came to be in the first place.
Thus the reticence of science-oriented companies to invest heavily in these mass data-gathering exercises to feed ML. It’s damn expensive, and almost always leads you back to raw data issues, not breakthrough discovery. Doing it without a set purpose in mind is even more likely to yield garbage.
I worked on some of the very best funded plant research out there. When it comes down to it, there's enough variation caused by confounding factors, and it takes so long to capture more data, that almost everything anyone tries cannot be called a success or a failure for years, because the individual measurements for one small plot of land somewhere just don't mean anything. Once you do an entire experiment for a season, which takes months, and you grab the little noisy data you have, and turn it into real rows, we were down to very little.
You can do more tests on smaller things, like checking if some protein will kill some cells of a pest, but making sure a plant produces it enough that it actually does something significant to the real, live pests, that it's not toxic, and it doesn't harm the plant's yield massively (as it's now spending time producing your pesticide) is still going to take years. We might be able to fold proteins, but the kind of things we'd need to really simulate plant biology well enough to not need years of failures are still very far away.
And it's far worse in medicine, as with plants at least nobody has ethical concerns if they fail and die, and nobody needs to get consent from a corn seed. Getting to 50 actual data points from many medical studies is already a lot of effort. And imagine when it's a long term study, and you need to follow patients for 30 years, as theym move, or die, or decide to stop participating, or who knows what.
Ethical concerns might become an even bigger bottleneck in the future. It's sad what we're doing with millions of rodents each year.
Here is an article by Pat Walters on the usefulness of ML in drug discovery. This article is a response to another one making the case that utility of ML models are very limited in drug discovery
https://patwalters.github.io/Response-to-Peter-Kenny/
I haven't worked in the industry side of things but in academia everyone kind of agrees that gradient boosting trees are some of the best models to do these things.
The real value right now is in figuring out how to generate robust data cheaply and quickly. I'd wager that the effect of a good model on marginal data is small, but the effect of a marginal model on great data is probably quite large.
I completely agree with you. Data quality is the most important. One thing that's sad is how much academic data that's not reproducible. https://en.wikipedia.org/wiki/Replication_crisis
The obvious question is what limits getting more data? Astronomy (especially in Australia) has been quite good at designing surveys to answer multiple scientific questions with reasonable amounts of data (and then fed into ML systems like the cannon). Sadly one of the consequences of the LLM hype is the increasing cost of doing this, so "AI" is actually making things worse not better.
It was always easy to come up with new molecules/drugs/materials. The thing that changed is the scale that it can happen with the new ml-based approaches.
What hasn’t changed is finding ones that are manufacturable/synthesizable.
Even if you find 1 million new stable molecules, there no guarantee that even one of them is manufacturable.
Yes indeed, I call it the revenge of statistics! Pseudoreplication rears its ugly head once again. Nonparametrics are certainly useful, but once you know things about the DGP, it’s often possible to do something “traditional” that’s better. Probabilistic programming, for life!
Obviously the missing part (which we already have for software and math) is that we need agents to be able to run automated loops in the real world. That basically requires robots. I think we'll be there in less than 5 years.
As to
it could be that the AI tools have to get to some threshold before they are very useful? Like with the Economist talking to Hassabis:
I hope to be lucky enough to never take a drug that ai had any impact on the discovery or study
the comments in this thread is what makes me love hn
What I increasingly see at my Biotech is more and more AI assisted drug candidates but no corresponding improvement in the capacity or ability to manufacture and scale them. I think it's a problem that groups like Anthropic will encounter in a few months/years and one that will annoye them a lot because it's not as simple as throwing more compute, people or money at it.