It is odd (or maybe not) that they decided to publish a marketing whitepaper rather than a more traditional journal submission + preprint. The work does appear to be sufficient for a publication, though there's a good chance a reviewer will rip into them for some of the assertions they make, but given the topic I'm sure the paper will be accepted regardless.
The market for entry-level programmers has already declined, but at least they were somewhat in demand and made reasonable salaries. Now what happens to post-docs who already make almost nothing and often get treated like crap?
I’m curious, what differentiates this from a preprint given the assumption it’s sufficient for publication? It didn’t read like marketing, they don’t seem to sell anything
> Now what happens to post-docs who already make almost nothing and often get treated like crap?
At least in the US, that particular brain drain has already been happening due to Trump's administration. The best of the best are exiting to other countries that will gladly have them, and then there will be far fewer people getting into the field. Science in general has taken a massive hit under the current administration and it going to take decades to fix if it's even possible.
They do partner externally. This work is fundamental discovery science, rather than industrial research.
the folks who run anthropic grew up reading scifi with crazy awesome biotech. However, when they look at biotech today, it's just depressing. It's incredibly slow, it takes decadfes to prove out new technologies, and they figure with this new tool, they can just point it at problems and have it emit discoveries. If they show a few high-impact discoveries, that makes a case for them to move biotech forward much faster than its current progress.
Also, anthropic has so much capitalization right now that it's simply easiest to invest it in a wide portfolio that includes both internal and external research.
Because their core product is not a long-term sustainable business strategy. Local hardware and models will continue to improve to the point of not needing the hosted solutions. And if you do need a hosted solution, remember that the big cloud providers already offer these solutions, so signing up for OpenAI/Anthropic _and_ AWS/GCP/Azure is not a sound business decision compared to just signing up with 1 of them that offers your cloud infra + GenAI infra. (Which is why the long-term benefits for cloud companies will probably be for the likes of AWS and not the likes of OpenAI).
They'll continue to burn money for marginal model improvements in the next few years all the while having no moat _and_ having Open-Weight / Local models eat their lunch.
The only way for them to stay relevant as a company is to expand beyond simply providing the models.
I'm old enough to remember the arrival of RDBMS, once IBM primed the space with DB2.
There was a pitched battle over features like row-level locking as competitors like Sybase, Ingress and Oracle scrapped it out. New features arrived on a monthly cadence, with immense engineering effort behind them. The winners (Oracle mostly) won a great moat which led to them to where they are today.
The fact that so many AI companies can produce amazing coding tools so quickly shows there is no moat, supporting your theory.
That is impossible to fund. At some point someone will decide to stop throwing money on the firepit that's the current business model and then hardware prices crash back to earth as 60-70% of the global demand disappears overnight.
It is more of a regulatory capture byproduct to prop up an artificial token driven Ponzi scheme.
There is a serious alternative to NVIDIA "AI" hardware dropping out of China in February 2027. There is no moat, but a whole lot of unpaid debts in the near future.
A chatbot for cancer researchers to talk to is worth single-digit billions at most. Anthropic is already valued at over a trillion dollars, on the premise that they can replace the majority of jobs in most knowledge industries. All the announcements about hacking / math problems / biological science are meant to create the impression that that strategy works and is repeatable across industries.
Cancer research is a lot harder for LLMs than math millennium problems though, because there is no fast feedback loop to iterate on. Even if you have a really good idea based on a solid theoretical insight, doing the experiments using in-vitro/mice/monkeys/humans can take years or even decades. I have no doubt that AI will help find new avenues that boost certain parts of research in these fields, but I don't see a potential for a drastic change until we at the very least give LLMs a direct way to interact with lab equipment and train them using RL on it.
Sorry, yes I agree 100%. I don't agree with their narrative, I was just explaining it. I think it's fraudulent and based on science fiction and will lead to a significant economic crisis.
Yes, but initial discovery of molecules and novel mechanisms is massive. That was the last generational change in modern drug research was the movement to high throughput screening, going from the ability to screen 10's of molecules to hundreds of thousands to find 'hits'. Better and more focused models, especially ones trained internally at big pharma companies will accelerate that portion of the pipeline, or increase the hit rate of successful compounds. Several companies are already taking this approach like Novo has been. There are other more early stage companies like Recursion and others that are doing the same thing. They are more tech companies than traditional wet lab companies.
I was actually thinking the other day that it makes perfect sense for AI companies to develop a professional services oriented software development arm. Imagine that you want to develop a training pipeline for "tasteful" programming: you might make a reward metric for that does some obvious stuff (nothing that anyone could easily agree is a bug like a crash, good performance, perhaps minimize LoC), but you really want to also want to also track "bugs" where the feature was discovered to be missing some unspecified nuance that was only discovered through product use, or train on ability to keep a small codebase while also keeping diffs small (essentially, "maintainability") as real new requirements come in.
So then you want a training set full of real product requirements and product evolution, which is something you could get if you offered custom software development, with a lot more control than you'd get trying to do the same by scraping random FOSS projects on github.
Other industries are perhaps similar. If you offer a service directly, you have much more ability to build collection of training data into the process. Want to make the best law bot? Buy a law firm, offer legal services, and integrate extremely deeply into their workflows. If their models turn out to be as good as they hype up, they should be able to scale to be a major player in any endeavor they move into with a relatively small number of staff and develop a strong feedback loop (not that that would be good for the rest of us).
If your core service is getting more expensive to provide and competitors are busy eating your margins, why let someone else taste your secret sauce and only get paid for the tokens, when you can keep the good stuff (bio capability) for yourself, and net both the profit and the fame?
I run into this all the time - we have such powerful functionality available to our users, and further we provide the elements that undergird all of it, so it’s totally possible for clients to take the services they buy from us and reconfigure them to make their own tools, better even than the ones we have built, purpose-built for their workflows…
And 9/10 clients will just click on the one thing they know and recognize and are familiar with and comfortable with… and then stop thinking about it.
It’s crazy how much of our job is not only building our product, but interrogating our clients over what they need, so we can demonstrate how our tools solve their problem. The users simply are not interested in figuring it out for themselves.
This is my speculation as well. For the time being, knowing how to use Claude extremely effectively probably beats out industry insider status. And Anthropic can attract whatever expertise it needs to build scrappy research teams in house. I'm guessing this kind of work doesn't need 100+ people, maybe just a dozen highly specialized people.
Given the prestige of the AI labs, the recent explosion of math proofs, the literal millions they can throw around, it seems very likely they can attract then fund small research projects across a broad range of science. And like startup math, it only takes one or two ground breaking results from a hundred attempts to pay back in the PR/hype.
This shows why biology is so much harder a problem area for LLMs than math, finding RTs is tedious but pretty doable today, they had to scope the problem down a lot from something that would be the equivalent of Navier Stokes in biology. Glad they’re doing it though, even if it’s just marketing.
It's clear Dario believes that the solution to AI's PR problem is to cure cancer. Or invent other revolutionary medical treatments. They're going to heavily promote every step along the way no matter how small or far away from commercialization they are, like this one.
No doubt that curing cancer would help, but I think the timeline might be a little too long. Even RSI AGI will not be able to get new medical treatments to market instantly. Real world testing takes a long time and is an unavoidable part of the process.
Real world approvals for drugs are accelerating through, even with all the steps. Think about it, Moderna went from zero, to approved vaccine in 10 months. While COVID vaccines were the exception, not the rule, there are ways to accelerate the process if there is will and $$$. In the last 20 years, the number of new drug approvals per year in the US has doubled, and the length of time to get approval has been cut in half.
Dario would like to ‘cure’ aging. He’s got some personal experience with bad illnesses, but aging isn’t that. I also have reduced trust for people who want to live forever and don’t have kids.
> I also have reduced trust for people who want to live forever and don’t have kids.
I want to live forever (or until I'm bored of it) and I don't have kids. I'm not sure what that has to do with trustworthiness.
Edit: And, you're saying you want to die. Is that more trustworthy than not wanting to die? I suppose if you are religious, you might believe you're going somewhere good when you die, in which case, you don't actually believe death exists, so we're having different conversations. I believe death exists and is permanent, and I'd like to not do that.
Not Op but there’s a inch of people asking why, and I have a similar feeling to op so here’s my post hoc justification for a weakly held and poorly supported prejudice:
It’s really because statistically, in my experience people without kids are more selfish than those without. This is more in description than judgement, but it’s true in my experience. We can speculate as to reasons, but looking after kids does train a certain kind of selflessness. Agreed we might be doing it for ultimately selfish reasons (self presentational or for care in old age or whatever). But for a good chunk of the time, caring for kids seems to require the fairly consistent subjugation of personal preferences, and a degeee of perspective taking, that I just think people without kids don’t have. And that often shows in their interactions at work and in daily life. Obviously there are myriad exceptions. But it’s true enough in my experience.
The wanting to live forever part also seems weird to me, and correlated with a certain sort of self regarding perspective. It seems obvious to me that I (or my generations) need to die for my children and grandchildren to have a good life. To try and subvert that also seems selfish or self important somehow.
I’m not really arguing this is a correct or good or just position. It might be terrible! But it did resonate..
> I want to live forever (or until I'm bored of it) and I don't have kids. I'm not sure what that has to do with trustworthiness.
When I hear people say stuff like this, I hear that they want to remove the single most universal chesterton's fence in all of living systems. I hear them take pride in their/our hubris, and demonstrate willingness to put the whole multiplex ecology of life at risk because they believe themselves/us to be more clever than thermodynamic evolution.
Biological singletons (outside very specific niche situations) are not meant to persist, and most anything that has tried, it has simply been selected out of the lineage. This constraint (which we don't understand yet) is presumably the whole reason why biology discovered and moved into the more ephemeral higher-order substrate of thought and culture.
I think there's a nuance gap here. Many of us, if offered, would be happy to live for a a millennium or two if we'd stay at worst middle-aged. But there's a giant gap between being willing to take that offer if given, and a desire for it being a big personal motivating factor.
I don't have kids. I have not caused a need for me to die to make room or to be "replaced" (though "overpopulation" arguments are often eugenicist and/or racist propaganda, and don't engage with actual density/agricultural limits, so I'm hesitant to make any arguments based on whether there is room for people to not die).
Probably should have left off the kids, but lets start with just distrusting anybody who wants to live forever. At the level of influence billionaires have, it is downright dangerous.
For everyone else confused: Think of all the people throughout history we would prefer would not have lived forever. Then multiple that by A LOT. Then consider how greedy and sociopathic most of the billionaire class is already.
Now, we could spend time getting distracted by childless. I don't think it matters.
Why? Who in their right mind would have children in 2026? Everything is burning, gone to shit, and projected to get worse. Having children is insanely irresponsible.
Cancer can be cured in a lot of cases, its just so damn expensive and the treatment is beyond torturous that some patients cannot handle it. Stem cells are amazing. But we need cheaper technology to replicate them into the cancer destroyers they need to be, as well as find ways to ease the pain of that internal battle.
Please tell me more about these “cures” you speak of. Because to my knowledge, yes we are good at getting patients into remission, we do not have “cures”
Coupled with the fact that treatment is often life altering in and of itself
We also have therapies based on monoclonal recombinant antibodies conjugated with chemotherapeutics. Simply put, we can produce antibodies that are specific for markers present in the surface of cancer cells, and we can attach drugs that can kill those cells. The antibody part is what makes this type of therapy very effective (you target only cancer cells, and not healthy cells) and also very expensive.
Semantics. Cancer is uncontrolled growth of cells. Treating it is killing/removing the cancer cells. Problem of course is you are never sure if you got all of them. If you got all of them you are “cured”. If you didn’t you are not cured in case the remaining cells manage to grow and spread again. We don’t talk about “cure” because of course it is impossible to verify if 100% is gone or not.
Why is it 'of course impossible'? Couldn't we someday have nanobots or other tech that could screen all your cells and be able to indicate whether any cancerous cells remain?
If we were able to find cancer cells with a precision down to 1 cell, killing them would be trivial. There are lots of cells in a person. It's possible we will get there someday. I think bioengineering is the most likely route. I.e., design a virus to specifically target the type of cancer cell you mean to eradicate.
No. It's hard to get even small molecules where we want them in the body, there's no way that the gigantic molecular clusters called "nanobots" could reliably get access to every single cell. (What may be possible is using things like the Moderna cancer vaccine to make your body an inhospitable environment for the growth and multiplication of the cancer cells.)
I don't know about "cured" but I've been cancer free for something like ~27 years.
With childhood cancers some of them have very high rates of "cure," but it is true that the impact of the treatments (at that time at least) follows you for life in various ways. The more modern immunotherapies and such seem potentially much better than chemo if they can be turned into successful and consistent approaches.
Burn (radiation). Cut. Poison (chemo). Your picks for cancer. IMHO Eat less. Remove all sugar and vitamins. Go a week without food. Give your body time to kill the weak cancer cells before they grow exponentially.
Yep, my prediction is that Anthropic is going to use Claude's reputation to "launder" known solutions to aging, cancer, and other things that society hasn't accepted quite yet. But maybe with the right marketing we'll try those things!
As they should because things like this get people thinking even if it something small. Once you get people thinking about things you tend to get solutions.
Also the general public might find the implications of AGI so distasteful even if everything goes well that we might stall out or get the Butlerian Jihad before we can cure cancer. Artists and Software Engineers, now also Mathematicians, already have existential crises, but the public still thinks AI is fake. I can't imagine the backlash when the realize what's coming even in the good ending.
>> the solution to AI's PR problem is to cure cancer
I think its much simpler than that.
Anything actually useful for people would be a good solution.
Obviously image gen and code gen is not the case, as though it does increase productivity, it doesn't make anyone's life actually better. If it led to 4 day work week - sure. Otherwise it could easily be net negative.
Its going to be an uphill battle. Every story about job losses, consequences to the community from building a datacenter (real or perceived), eminent domain case that blows up, plus all the slop on every platform. Not to mention a lot of normies think techbros are obnoxious, and that is who is hyping ai.
They'll need to show their goal is to help humanity and that all the other peoole arent acceptable collateral damage. Since those other people get to vote.
> Even RSI will not be able to get new medical treatments to market instantly. Real world testing takes a long time and is an unavoidable part of the process.
I am already having a headache thinking of the whining from the biologist community (if any? I hope their reaction is not as extreme as that of mathematicians).
You can't say this. We have no idea. There is nothing about the law of physics that pushes cancer cure a long time away. A lot of people would have told you AI was decades away, yet here we are. We are still on track for possible strong take off.
Now on real world testing, you think the rule applies? I tell you it doesn't. Human life might be precious, but human life in practice is also not precious. We waste so much of it. In some countries regulations will stop/slow it, but there are plenty of places around the world that will turn a blind eye for a fistful of dollars. Countries will go to those locations if it means gaining an edge.
Appeals to laws of physics as a "first principles" attempt to explain how thousands of diverse diseases could theoretically be solved overnight by a big computer (while hand waving away the years of clinical trials, false starts and failures involved in a single new successful treatment) just makes you seem wildly out of touch and uninformed about the actual problem space.
There are many things about the laws of physics that push a cancer cure a long time away! Biology is downstream of physics, and the biology of cancer is so vast that the very concept of a "cure for cancer" is almost nonsensical.
There's a lot about biology that makes cancer fundamentally hard to treat, and the efficacy of cancer treatments fundamentally hard to measure. I'm optimistic that we'll eventually get to a point where we can meaningfully say we "cured cancer", but it will almost certainly be a cluster of thousands of treatment protocols which each have to be tested over 5-10 years for recurrence. There's no reason to expect that there should exist any broad-spectrum cancer treatment better than radiotherapy, or any fast test to determine whether long-term remission will be achieved.
Very cool! However, the amazing absence of results makes me question whether they've got a Nature letter forthcoming or whether they know that another AI lab has a similar finding...
Thank you for sharing, that provided some good context for how to interpret this
Post content:
_____
I wish we didn’t need these again, but here is the honest version of Anthropic’s biology announcement
(Caveat: I haven’t worked in bioinformatics for many years.)
The good: Anthropic ran ~950 Claude agents over a large biological sequence database. Claude searched, wrote code, compared sequences and genomic neighborhoods, and found an interesting pattern that apparently had not been noticed before: a known reverse transcriptase associated with another gene and a repetitive DNA array.
That is cool. Automating this kind of open-ended bioinformatics search at scale is useful, and Claude may have found a lead a human would have missed.
But: Claude did not do a biological experiment. It searched databases and analyzed data.
Humans then took the candidate into the wet lab. And the wet-lab result so far is modest: they showed that the repeat array produces short RNAs.
We still don’t know what the system does. No function, mechanism, phenotype, targeting, defense activity, or programmability has been demonstrated.
This is also where the CRISPR framing gets ahead of the result. Right now, “it has some features reminiscent of known programmable systems” is a hypothesis for what to investigate next, not a discovery that it behaves like CRISPR.
And there is a missing baseline: bioinformatics has had tools for finding unusual gene neighborhoods and candidate systems for years. The interesting comparison is 950 Claude agents vs. an expert using the best existing computational pipelines - not Claude vs. someone manually looking through 200,000 sequences.
So my honest announcement would be:
Claude autonomously found an interesting candidate for a previously uncharacterized biological system. A small human wet-lab experiment confirmed that part of the candidate is expressed. We don’t yet know what it does.
That is a good result.
But in a regular biology lab, this isn’t the finished paper. It is the result you show at lab meeting and say: “This looks interesting. Now we need to figure out what the hell it does.”
Maybe that next step leads to a major discovery. But that discovery hasn’t happened yet.
There are so many people involved on this yet we still say things like "Claude did", we need to start waking up and being more real about how we are still in "AI + Human" land.
What's wrong with saying "A team of researchers backed by Anthropic using Claude discovers a novel enzyme system with CRISPR-like repeats" or, ffs, mention the lead researcher in the headline?
OpenAI/Anthropic have never pitched themselves as a replacement for farmers. They do explicitly say that they're going to cause significant job loss in knowledge work sectors all the time.
I'm more annoyed that they announce "CRISPR-like" to hit those SV Next Big Thing dopamine receptors but upon reading haven't done any laboratory work to determine if it has any useful applications like CRISPR-Cas9.
It's totally legitimate research worthy of publication, but Anthropic chose a hot technology in the popular imagination for a reason. Now I'm going to have to see "Claude invented a new CRISPR in 24 hours!" everywhere and trying to correct it will just turn into repetitive arguments about goalposts moving....
yes and no; you cant dump DNA in the context window and call it a day, in the blog post it was a common tool calling session. you do can have actual ml models for that, that the llm could use as a tool.
I don't think you quite understand the loops here.
At Google/OpenAI/Anthropic level you have clusters of LLM agents working with clusters of ML agents doing all kinds of tasks. A lot of this falls into proto-RSI where the LLM can improve the ML agents output based on analysis of said ML.
This isn't much different from how people work, you can't dump even part of DNA context in a human mind and get anything useful out. We has humans have to use and build tools to find answers because of scaling efficiencies of different computation types.
The study results themselves aren't really dangerous in any way I can see. This is basic microbiology, and not necessarily some kind of major breakthrough that will change the world on its own. It's possible this leads to something big like CRISPR, but most likely not. The work is more the case of noticing something that someone hasn't noticed yet. It would have gotten noticed eventually, they just did it before someone else did (assuming they didn't get a hint somehow).
A lot of molecular biology is noticing something that you can't explain or that seems weird and might be interesting. Once it's noticed the followup is often fairly straightforward and it either pans out or it doesn't. The exciting/scary/unlikely part is that the LLM on its own recognized something as being important to follow up.
From my skim of the paper, the work could only be done by someone with a pretty good understanding of the biology and an extremely good understanding of how to use LLMs and agents. LLMs are not going to take over biology yet.
> Anthropic’s head of influencer, Lexie Barnhorn, has described creators as essential to building trust in complicated technical products. Its strategy is partly consumer-to-business: People who adopt Claude personally may later introduce it in their workplaces.
> Anthropic’s best-known creator events have been smaller dinners and pop-ups in which Claude remained the ostensible subject.
Personally I feel like Anthropic is underrepresented in "normie" marketing, all of my non-tech savy friends only know of ChatGPT and use "ChatGPT" in the same way my mom says "Nintendo" when talking about game consoles
This is great, but I can't help but wonder if we're going to have another post next week with a lab complaining that they were about to publish this same finding, and they had Claude proofread their paper, and whoops how'd that get into Anthropic's training data?
I wonder how long it will take for the damage Alpöge and Buckmaster have done to the perception of these AI-driven scientific developments to fade.
Not saying that they were right or wrong, but that single moment sullied all AI-driven breakthroughs that came after it, and I don't think it was ever particularly relevant, at least not nearly to the degree that it was presented in the media. But I guess it ended up being a convenient outlet for AI anxiety in the end.
I don't think of this stuff in terms of AI anxiety, I just think that the AI labs should be falling all over themselves to display deference and humility to those who made it possible.
The LLMs that make this stuff possible weren't created by the AI labs from whole cloth. They crept up and jumped onto the shoulders of giants, basically the collected (non-consensually, of course, but jingles keys look at this pelican riding a bicycle!) works of humanity. Every discovery LLMs enumerate in this fashion rightfully needs to have a billboard-sized asterisk regarding the provenance of the discovery. "Claude" didn't discover this, everyone who worked to produce the internet that Anthropic siphoned into their dataset belongs on the credits.
It's great that it happened, and I wish them the best of luck in using our work to make the world a better place. Just don't forget who the rightful owners are.
The AI labs did that to themselves. All those billions and their marketing and communication skills are like those of a local street vendor selling fake knockoffs.
I wish we could discuss this in a way that didn't immediately devolve into people shouting up or shouting down that this is either meaningless or singularity.
Caveating I'm not a biologist, but my understanding of the way this kind of thing works right now is a basic three-step process:
1) Find molecules and DNA/RNA sequences in the wild and catalog them.
2) Discover interesting subsequences among these.
3) Figure out whether any useful applications can come from what was discovered.
All three of these generally take a long time. Systematic automatic analysis of known databases speeds up and removes some of the luck from 2. But 1 and 3 are still long poles. 1 has the further issue that we usually discover these in existing organisms. I recall much of the outcry over tropical deforestation back in the 90s and replacing of rainforests with palm oil monoculture today is that the vast majority of terrestrial biodiversity is found in rainforests, and destroying them at industrial scale risks losing potentially useful molecules forever. 3 has the problem that you need to conduct physical experiments, and are limited by the speed of biochemical reactions no matter what and by the speed at which human subjects can be found and ethically experimented on assuming we care about being ethical.
A lot of good can come of this, but I don't see a path to singularity here, assuming we're talking the original Kurzweil meaning there of all technological progress that will ever happen all happening at once. Data collection and experimentation on living subjects, human or not, can only happen so fast, regardless of automation. It's not computational. Whenever you have to interface with the real world, you're now working at the speed of the real world, not the speed of electricity. CRISPR was discovered in 1987 and first used to edit a gene sequence in a human zygote in 2015. I'm sure there are plenty of ways to make the candidate discovery to human application step not take three decades, but it's never going to be three months, either.
Aren't there unlimited mechanisms like this? Isn't this why Doudna isn't a billionaire (you can patent something, but it's easy to create another one and patent it separately)?
I think the odds of a human accidentally creating a novel virus or bioweapon at the behest of a rouge AI are pretty small to be honest. That's a lot of manual labor to go "oops I didn't realize what this was!"
Not that I agree with the comment you're replying to - but I find this response funny, when just today there was a link on the front page about the US military bombing a school because of AI output
Biosafety is a very real concern but "lab" is a big bucket, a molecular genetics lab can't synthesize new viruses out of thin air if it's not a virology lab. Sequencers sequence etc. The lab has the equipment it has.
The singularity, as defined by Hinton (and others) as RSI (Recursive Self Improvement) may actually be beginning already, as OpenAI has announced an AI acting as a "research intern" (!).
How is this different from arguing that Microsoft Clippy was RSI? An AI tool being involved in the process of work can't be the bar for RSI.
I don't think there can be a coherent definition of RSI unless people lay out their theory for how intelligence scales. LLM-assisted coding is great but respectfully optimizing pytorch features or whatever is not gonna lead to exponential improvements. That approach to scaling diminished years ago, leading all the labs to switch to reasoning.
Now it seems reasoning is also yielding diminishing returns, so all the labs are pivoting to specializing in particular fields like math / infosec / biology. They're improving due to accessing new proprietary training data and doing RL with human experts. Again I don't really see any amount of "AI research interns" leading to an exponential improvement to this strategy, they're not the bottleneck in the first place.
Self improvement during training, and AI self training are already happening. Easily/quickly are seemingly a factor of how much power/hardware you want to use at once.
With the level of compute they have they aren't stuck with frozen models like you are.
The market for entry-level programmers has already declined, but at least they were somewhat in demand and made reasonable salaries. Now what happens to post-docs who already make almost nothing and often get treated like crap?
We still need post docs. What will change is their specializations.
i'll give you a hint: they're selling something
At least in the US, that particular brain drain has already been happening due to Trump's administration. The best of the best are exiting to other countries that will gladly have them, and then there will be far fewer people getting into the field. Science in general has taken a massive hit under the current administration and it going to take decades to fix if it's even possible.
I guess the improvement loop is tighter and they have more control over how discoveries can be used for marketing?
But, in my mind, it begins to feel like they are setting themselves up to be “everything” companies instead of focusing on their core product…
the folks who run anthropic grew up reading scifi with crazy awesome biotech. However, when they look at biotech today, it's just depressing. It's incredibly slow, it takes decadfes to prove out new technologies, and they figure with this new tool, they can just point it at problems and have it emit discoveries. If they show a few high-impact discoveries, that makes a case for them to move biotech forward much faster than its current progress.
Also, anthropic has so much capitalization right now that it's simply easiest to invest it in a wide portfolio that includes both internal and external research.
They'll continue to burn money for marginal model improvements in the next few years all the while having no moat _and_ having Open-Weight / Local models eat their lunch.
The only way for them to stay relevant as a company is to expand beyond simply providing the models.
There was a pitched battle over features like row-level locking as competitors like Sybase, Ingress and Oracle scrapped it out. New features arrived on a monthly cadence, with immense engineering effort behind them. The winners (Oracle mostly) won a great moat which led to them to where they are today.
The fact that so many AI companies can produce amazing coding tools so quickly shows there is no moat, supporting your theory.
There is a serious alternative to NVIDIA "AI" hardware dropping out of China in February 2027. There is no moat, but a whole lot of unpaid debts in the near future.
Popcorn ready =3
Which is exactly what is being done.
https://www.reuters.com/world/anthropic-quietly-sets-up-biol...
So then you want a training set full of real product requirements and product evolution, which is something you could get if you offered custom software development, with a lot more control than you'd get trying to do the same by scraping random FOSS projects on github.
Other industries are perhaps similar. If you offer a service directly, you have much more ability to build collection of training data into the process. Want to make the best law bot? Buy a law firm, offer legal services, and integrate extremely deeply into their workflows. If their models turn out to be as good as they hype up, they should be able to scale to be a major player in any endeavor they move into with a relatively small number of staff and develop a strong feedback loop (not that that would be good for the rest of us).
I run into this all the time - we have such powerful functionality available to our users, and further we provide the elements that undergird all of it, so it’s totally possible for clients to take the services they buy from us and reconfigure them to make their own tools, better even than the ones we have built, purpose-built for their workflows…
And 9/10 clients will just click on the one thing they know and recognize and are familiar with and comfortable with… and then stop thinking about it.
It’s crazy how much of our job is not only building our product, but interrogating our clients over what they need, so we can demonstrate how our tools solve their problem. The users simply are not interested in figuring it out for themselves.
Given the prestige of the AI labs, the recent explosion of math proofs, the literal millions they can throw around, it seems very likely they can attract then fund small research projects across a broad range of science. And like startup math, it only takes one or two ground breaking results from a hundred attempts to pay back in the PR/hype.
https://www.reuters.com/world/anthropic-quietly-sets-up-biol...
No doubt that curing cancer would help, but I think the timeline might be a little too long. Even RSI AGI will not be able to get new medical treatments to market instantly. Real world testing takes a long time and is an unavoidable part of the process.
I want to live forever (or until I'm bored of it) and I don't have kids. I'm not sure what that has to do with trustworthiness.
Edit: And, you're saying you want to die. Is that more trustworthy than not wanting to die? I suppose if you are religious, you might believe you're going somewhere good when you die, in which case, you don't actually believe death exists, so we're having different conversations. I believe death exists and is permanent, and I'd like to not do that.
It’s really because statistically, in my experience people without kids are more selfish than those without. This is more in description than judgement, but it’s true in my experience. We can speculate as to reasons, but looking after kids does train a certain kind of selflessness. Agreed we might be doing it for ultimately selfish reasons (self presentational or for care in old age or whatever). But for a good chunk of the time, caring for kids seems to require the fairly consistent subjugation of personal preferences, and a degeee of perspective taking, that I just think people without kids don’t have. And that often shows in their interactions at work and in daily life. Obviously there are myriad exceptions. But it’s true enough in my experience.
The wanting to live forever part also seems weird to me, and correlated with a certain sort of self regarding perspective. It seems obvious to me that I (or my generations) need to die for my children and grandchildren to have a good life. To try and subvert that also seems selfish or self important somehow.
I’m not really arguing this is a correct or good or just position. It might be terrible! But it did resonate..
When I hear people say stuff like this, I hear that they want to remove the single most universal chesterton's fence in all of living systems. I hear them take pride in their/our hubris, and demonstrate willingness to put the whole multiplex ecology of life at risk because they believe themselves/us to be more clever than thermodynamic evolution.
Biological singletons (outside very specific niche situations) are not meant to persist, and most anything that has tried, it has simply been selected out of the lineage. This constraint (which we don't understand yet) is presumably the whole reason why biology discovered and moved into the more ephemeral higher-order substrate of thought and culture.
Just my feelings though. Feel free to disagree.
Because it hasn't happened?
what a weird bias
While everyone else can't afford it. Hard to think of a more demoralizing "off with their heads" dystopian scenario.
I have reduced trust in people who make judgements about the value systems of others based on fairly meaningless characteristics.
Why are you only allowed to live forever if you have kids?
Seems like someone seeking immortality should be willing to do for the elixir if they want it even a little bit...
How so?
For everyone else confused: Think of all the people throughout history we would prefer would not have lived forever. Then multiple that by A LOT. Then consider how greedy and sociopathic most of the billionaire class is already.
Now, we could spend time getting distracted by childless. I don't think it matters.
I'd even be fine with people who are billionaires living forever, so long as they don't remain billionaires / don't fuck with politics / etc.
Your dramatization of society's ills are not tethered to reality
https://www.cancer.gov/news-events/cancer-currents-blog/2024...
https://jitc.bmj.com/content/8/2/e000848 (careful: Figure 1 can be very graphical, but it shows the huge positive impact of this therapy)
We also have therapies based on monoclonal recombinant antibodies conjugated with chemotherapeutics. Simply put, we can produce antibodies that are specific for markers present in the surface of cancer cells, and we can attach drugs that can kill those cells. The antibody part is what makes this type of therapy very effective (you target only cancer cells, and not healthy cells) and also very expensive.
https://www.cancer.gov/about-cancer/treatment/research/car-t...
https://www.cancer.gov/about-cancer/treatment/types/immunoth...
https://en.wikipedia.org/wiki/CAR_T_cell
https://www.theguardian.com/society/2026/may/10/cancer-treat...
https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
https://news.ycombinator.com/item?id=49329717
It’s the top rated comment in the thread. Somebody tried to do something good, this is the response.
This pisses me off severely.
He isn’t wrong. But selling potential cures for cancer won’t cut it.
Not if it's a virus
I think its much simpler than that. Anything actually useful for people would be a good solution.
Obviously image gen and code gen is not the case, as though it does increase productivity, it doesn't make anyone's life actually better. If it led to 4 day work week - sure. Otherwise it could easily be net negative.
They'll need to show their goal is to help humanity and that all the other peoole arent acceptable collateral damage. Since those other people get to vote.
Is it unavoidable, though?
Now on real world testing, you think the rule applies? I tell you it doesn't. Human life might be precious, but human life in practice is also not precious. We waste so much of it. In some countries regulations will stop/slow it, but there are plenty of places around the world that will turn a blind eye for a fistful of dollars. Countries will go to those locations if it means gaining an edge.
https://x.com/ziv_ravid/status/2102844800345251858
Post content:
_____
I wish we didn’t need these again, but here is the honest version of Anthropic’s biology announcement (Caveat: I haven’t worked in bioinformatics for many years.) The good: Anthropic ran ~950 Claude agents over a large biological sequence database. Claude searched, wrote code, compared sequences and genomic neighborhoods, and found an interesting pattern that apparently had not been noticed before: a known reverse transcriptase associated with another gene and a repetitive DNA array.
That is cool. Automating this kind of open-ended bioinformatics search at scale is useful, and Claude may have found a lead a human would have missed.
But: Claude did not do a biological experiment. It searched databases and analyzed data.
Humans then took the candidate into the wet lab. And the wet-lab result so far is modest: they showed that the repeat array produces short RNAs.
We still don’t know what the system does. No function, mechanism, phenotype, targeting, defense activity, or programmability has been demonstrated.
This is also where the CRISPR framing gets ahead of the result. Right now, “it has some features reminiscent of known programmable systems” is a hypothesis for what to investigate next, not a discovery that it behaves like CRISPR.
And there is a missing baseline: bioinformatics has had tools for finding unusual gene neighborhoods and candidate systems for years. The interesting comparison is 950 Claude agents vs. an expert using the best existing computational pipelines - not Claude vs. someone manually looking through 200,000 sequences.
So my honest announcement would be:
Claude autonomously found an interesting candidate for a previously uncharacterized biological system. A small human wet-lab experiment confirmed that part of the candidate is expressed. We don’t yet know what it does.
That is a good result.
But in a regular biology lab, this isn’t the finished paper. It is the result you show at lab meeting and say: “This looks interesting. Now we need to figure out what the hell it does.”
Maybe that next step leads to a major discovery. But that discovery hasn’t happened yet.
There are so many people involved on this yet we still say things like "Claude did", we need to start waking up and being more real about how we are still in "AI + Human" land.
What's wrong with saying "A team of researchers backed by Anthropic using Claude discovers a novel enzyme system with CRISPR-like repeats" or, ffs, mention the lead researcher in the headline?
It's totally legitimate research worthy of publication, but Anthropic chose a hot technology in the popular imagination for a reason. Now I'm going to have to see "Claude invented a new CRISPR in 24 hours!" everywhere and trying to correct it will just turn into repetitive arguments about goalposts moving....
> Startup aims for Claude AI to direct robots in lab environments, one source says
> Company to stop short of clinical trials to avoid drugmaker competition, life sciences head says
https://www.reuters.com/world/anthropic-quietly-sets-up-biol...
At Google/OpenAI/Anthropic level you have clusters of LLM agents working with clusters of ML agents doing all kinds of tasks. A lot of this falls into proto-RSI where the LLM can improve the ML agents output based on analysis of said ML.
This isn't much different from how people work, you can't dump even part of DNA context in a human mind and get anything useful out. We has humans have to use and build tools to find answers because of scaling efficiencies of different computation types.
A lot of molecular biology is noticing something that you can't explain or that seems weird and might be interesting. Once it's noticed the followup is often fairly straightforward and it either pans out or it doesn't. The exciting/scary/unlikely part is that the LLM on its own recognized something as being important to follow up.
From my skim of the paper, the work could only be done by someone with a pretty good understanding of the biology and an extremely good understanding of how to use LLMs and agents. LLMs are not going to take over biology yet.
Generally speaking, hiring an army of influencers to shill for you results in bad PR, and comments like this one.
Is there an equivalent headline for Anthropic of this?: https://www.businessinsider.com/inside-open-ai-influencer-ma...
https://www.businessinsider.com/emma-orhun-canceled-claude-p...
> Anthropic’s head of influencer, Lexie Barnhorn, has described creators as essential to building trust in complicated technical products. Its strategy is partly consumer-to-business: People who adopt Claude personally may later introduce it in their workplaces.
> Anthropic’s best-known creator events have been smaller dinners and pop-ups in which Claude remained the ostensible subject.
Not saying that they were right or wrong, but that single moment sullied all AI-driven breakthroughs that came after it, and I don't think it was ever particularly relevant, at least not nearly to the degree that it was presented in the media. But I guess it ended up being a convenient outlet for AI anxiety in the end.
The LLMs that make this stuff possible weren't created by the AI labs from whole cloth. They crept up and jumped onto the shoulders of giants, basically the collected (non-consensually, of course, but jingles keys look at this pelican riding a bicycle!) works of humanity. Every discovery LLMs enumerate in this fashion rightfully needs to have a billboard-sized asterisk regarding the provenance of the discovery. "Claude" didn't discover this, everyone who worked to produce the internet that Anthropic siphoned into their dataset belongs on the credits.
It's great that it happened, and I wish them the best of luck in using our work to make the world a better place. Just don't forget who the rightful owners are.
The people that say "It's just a next word predictor" might as well be saying "Well, it's just a long rage nuclear missile".
Caveating I'm not a biologist, but my understanding of the way this kind of thing works right now is a basic three-step process:
1) Find molecules and DNA/RNA sequences in the wild and catalog them.
2) Discover interesting subsequences among these.
3) Figure out whether any useful applications can come from what was discovered.
All three of these generally take a long time. Systematic automatic analysis of known databases speeds up and removes some of the luck from 2. But 1 and 3 are still long poles. 1 has the further issue that we usually discover these in existing organisms. I recall much of the outcry over tropical deforestation back in the 90s and replacing of rainforests with palm oil monoculture today is that the vast majority of terrestrial biodiversity is found in rainforests, and destroying them at industrial scale risks losing potentially useful molecules forever. 3 has the problem that you need to conduct physical experiments, and are limited by the speed of biochemical reactions no matter what and by the speed at which human subjects can be found and ethically experimented on assuming we care about being ethical.
A lot of good can come of this, but I don't see a path to singularity here, assuming we're talking the original Kurzweil meaning there of all technological progress that will ever happen all happening at once. Data collection and experimentation on living subjects, human or not, can only happen so fast, regardless of automation. It's not computational. Whenever you have to interface with the real world, you're now working at the speed of the real world, not the speed of electricity. CRISPR was discovered in 1987 and first used to edit a gene sequence in a human zygote in 2015. I'm sure there are plenty of ways to make the candidate discovery to human application step not take three decades, but it's never going to be three months, either.
> All of the lab work is performed by human scientists.
I don't think there can be a coherent definition of RSI unless people lay out their theory for how intelligence scales. LLM-assisted coding is great but respectfully optimizing pytorch features or whatever is not gonna lead to exponential improvements. That approach to scaling diminished years ago, leading all the labs to switch to reasoning.
Now it seems reasoning is also yielding diminishing returns, so all the labs are pivoting to specializing in particular fields like math / infosec / biology. They're improving due to accessing new proprietary training data and doing RL with human experts. Again I don't really see any amount of "AI research interns" leading to an exponential improvement to this strategy, they're not the bottleneck in the first place.
With the level of compute they have they aren't stuck with frozen models like you are.
This A.I. hype makes the Internet Bubble look like a walk in the park.