The new generation of KRAS inhibitors, of which daraxonrasib is just the first to make it to approval, is genuinely a practice-changing breakthrough.
It won’t cure people with metastatic disease, but it will help a lot live better lives for longer.
Not true. The folfirinox chemo regimen was invented a couple of years ago, and it significantly increased the life expectancy and quality of life for terminal pancreatic cancer patients.[0]
True. statistically it was significant.
Progression-free survival increased from 3 months to 6 months. (overall-survival increased from 7 months to 11 months).
Both statistically significant, and at the cost of increased toxicity during those additional 3 to 4 months of “’significantly’ increased life expectancy”
Am I missing something?
From other sources:
> Median overall survival: 13.2 months versus about 6.6 to 6.7 months with chemotherapy.
First, this is a brutal drug, very hard to tolerate, and as better ways are found to manage side effects and to dial in the dosing, the treatment effects will increase.
Further, this is in metastatic pancreatic cancer, a stage of the disease where virtually all hope of a cure is gone for most. Any progress here will translate into bigger gains in earlier stages of the disease. We test new unproven therapies in metastatic cases like these because all other better options have been exhausted. And for decades, all those trials have been failures, this is the first time something has worked at all.
Additionally, this is just the first drug showing how to go after KRAS. Follow-on drugs, which some people call mere "me-too" drugs, will improve efficacy, have fewer side effects, and be more tolerable, all improving real-world results.
And as we learn to make KRAS treatments more tolerable, we'll also learn about how best to combine them with other existing chemotherapies, which requires lots of testing. Go to ASCO, and you'll see trial after trial after trial of small variations, trying to continuously improve existing treatments. It's a slow and brutal way to learn, but it's all we've got.
So if you only look at 13.2 months versus 6.7 months, yes, you are missing something! This is like discovering a whole new vein of gold ore to mine. We don't know where it will lead or how big it is, but what we can see right now is far far more than 13.2 / 6.7.
Would you say the same thing if it were doubled from five minutes to ten minutes? Who cares about doubling? It's the 7.5 months that's the point. And 7.5 months is what it is.
If all of you guys are fine with giving up those last few months, then give it to me and the others who want to soak up every last moment of life.
X is fairly large for now for me but...
Merely counting months does not do much - you'd have to count the quality of these months too.
If I had a choice between an average of 6 or 13 months, I'll take the 13.
And who knows, maybe they can refine the approach or improve screening for earlier detection or something and get it higher for future patients. This could be just the beginning.
I played with RC cars back in the 90s, 1200mAh was considered beefy before NiMh came and bumped it up slightly. Now we have lithium batteries with several times that capacity in the same form factor.
Most of that came from chipping away at it. Yes new technology got introduced but in the beginning they were not decisively superior, they had to be refined and improved little at a time.
Perhaps we'll find a miracle cure that fixes everything, maybe we have to chip away at it and get half a year here, another year there.
edit: If you asked me if I would like to live 7 more painful months for $40K/month, I'd at first think you were joking, and then I'd think you were insane. Of course not. Give 14 kids a full scholarship in my name.
We're putting a lot of effort into making sure that people leave this world completely drained of every penny.
Here you are, sacrificing your own life to make others lives better and peoiple here are like; "Nah..."
So, it could be that the new drug is much much better, but they couldn't tell at the time because they only were trying to say that it was significantly better then chemotherapy.
I don't see that as a breakthrough.
You've broken through into the unknown, and maybe there's a fresh wall of unknown composition, but maybe the tools that got you this far can get you further if tweaked.
> Treatment-related adverse events that led to treatment discontinuation occurred in 1.2% of the patients in the daraxonrasib group and in 11.2% of those in the chemotherapy group.
That's a huge reduction in side effects bad enough to stop treatment.
Anything that extends your time could allow another round of treatment, with its own over/under that gets you a little more time. So there should be a bit of compound interest here.
Cancer is a problem that can be solved, and it needs to be solved asap. Cancer too often gets either relegated to "Actually, statistically it's very unlikely" by people who don't have it or stylized to the God of Death that unlike other illness we won't solve anytime soon. There is no time to be lost tackling with AI. Dario gets it. Had cancer himself and his father passed away from it.
Unless we can reliably sign our cells like we do our binaries, we ain't "solving" cancer.
It's always impossible until it's done
I believe this is the GTA VI comment
I mean I get you but the problems you mention are specific to this Earth based middle ground we emerged into.
Others are likely so fundamentally linked to forces we cannot overpower due to the immense amount of variables in physical reality.
Consider a Riemann proof as text is a tiny amount of information relative to all the radiation and stuff floating in space or jiggle at the quantum scale.
Same with a neuralink + brain as our meat suits only want to make sense of things that are familiar.
Cosmic and quantum scale is way different than this stable middle ground we live in. We can still do a lot to stabilize here but I doubt things like space colonization work out given we cannot predict enough of the future to avoid the single catastrophic event that will end all progress and force a do over; Earth's resources can only support giving is so many mulligans. I see limited resources/time as another example of scaling laws just applied across eons.
Even though it's far more far fetched it's probably a better long term goal to focus on Stargate or The Expanse like gate technology rather than chew up Earth's resources chasing Star Trek ships.
And that BCIs have been used for people to play videogames literally since videogames were invented?
Neuralink is a refinement in the field, not a revolution.
edit: Watch the video shared in sibling comments. The big difference in neuralink from BCI SOTA in 1990 is a smaller set of hardware reading your brain (genuine improvement), configurability (minor improvement), and I think a significant improvement in fidelity. The implementation of thinking about moving a body part to give a reader a trivial muscle neural signal to read is older than I am, and has not improved as far as I can tell. That said, the user is demonstrating fine control of a game, so if those advancements were what it took to go from playing Civilization to playing GTA, kudos.
BCIs however are zero percent of the way towards anything like "scanning" your brain or "uploading" your brain, concepts that themselves might not even be physically possible. Even if you could magically copy your brain signals to a computer a la moving a VM, you would not be uploaded to the computer. A clone of your experience would. You still die in misery.
Unfortunately, the primary value Musk brought to the field was a willingness to invest money. Time and time again, his cultish supporters mistake a banal investment in engineering as groundbreaking, because they completely ignore the total lack of investment before he waved his cash around. We don't need musk to invest in things. We need a willingness to spend money towards goals. Also maybe read some magazines. BCIs were something Popular Science talked about occasionally, which is one of the ways I know about prior art in the field.
there doesn’t even need to be a commonality, all will get solved
I hear ML researchers are even picking up smoking again
For many cancers we don’t have a good animal model or cell model, let alone a simulation.
AI can and will help us identify good targets for treatment, but the learning loop will be full of slow non deterministic biomedical experiments.
I can't tell anymore.
I understand it's hard to burn through an 8-figure marketing budget but this is wild.
In math.
I’ve seen some of what’s going on with “long horizon” training and how it’s being applied to life sciences
I can extrapolate what that means and it bolsters my confidence into what I wrote
Yes, it can, but to imply it’s simply one disease is missing completely the exact reason why it’s so hard to cure.
Its actually crazy all of these threads end up being about how good AI is, when the post has 0 to do with AI.
Pancreatic cancer halted by virus injection in three patients - https://news.ycombinator.com/item?id=50563066 - Nov 2026 (157 comments)
Which is very different from being administered an adenovirus (VRT106) that recognizes a receptor often expressed on cancer cells (MXRA8), that after ‘infecting’ the cell triggers direct oncolysis and indirect immunogenic responses.
https://www.nytimes.com/2026/09/30/science/pancreatic-cancer...
Pancreatic cancer is nasty (most cancers are), but the recent breakthroughs make me hopeful that we'll one day see a world where it won't be so devastating.
You know, like we do now with measles.
If we can prove GRH with AI we really should be able to cure cancer.
The human body is not a game like chess. There is hidden information everywhere: stuff nobody ever tried to measure, stuff that somebody tried to measure and got it wrong, and the stuff people tried to measure and actually got right.
An AI can’t just deduce a prior hard physical facts about people.
And I’m pretty sure the stuff people measured correctly is a tiny fraction of all the features and data that are relevant to cancer in a human body.
You cannot simulate a human. Not even in theory. There's no magic AI can find to overcome that.
There aren't enough atoms in the universe to even contain a theoretical computer able to do it.
The universe will always be dramatically faster at "running" atoms and interactions than a computer, because a computer will always have to explicitly do the work of summing up interactions with all other atoms and particles, something the universe itself gets "for free".
Part of the possible speedup from things like quantum computers is taking advantage of the fact that the universe will "run" a set of interactions faster and with higher fidelity than you ever could hope
Otherwise I feel the parent's comment is perfectly reasonable.
Cancer is simply a fatal combination of mutations that results in a set of cells that no longer follow the regulations in place that the organism uses to control unbounded and out of place growth of that or those cells.
I see no reason AI would be unable to help with such a simple problem statement.
In many ways cancer is simpler than the Riemann hypothesis to attack.
The complexity of cancer, from what I've studied, isn't that we don't understand what's broken about the cells. It's that we don't manipulate matter as efficiently as biology to attack those broken cells.
Just this week news hit about a virus that can stop pancreatic cancer in its tracks.
Now why can the virus do this and not our medicines?
Because viruses work on the same scale of matter manipulation and process genetic information at the same physical scale that our cells do.
You're right that the ability to emit tokens, no matter how intelligent, is tangential to our ability to manipulate matter.
However DNA and RNA can be expressed as tokens.
Thus one can imagine that just as AI has helped us emit information via language and code tokens faster, it ought to be able to emit DNA sequences faster than any human typing out A's, T's, C's and G's by hand.
Thus, I can imagine a world in which cancer cell sequencing happens much faster and viral targets for those cancer cells are written by AI as easily as they write code today.
That AIs have somehow managed to wrangle all human information to advance human capability in a variety of domains is promising because individual humans likely have no hope of wrangling, in a single human mind, all of the genetic, physical, and chemical pathways in a living organism as complex as a human.
I'm not predicting when this happens but given that there are already services that will print a custom DNA or RNA strand to order, I absolutely see a world in which viruses, custom-designed to attack an individual's cancer cells bearing genetic markers matching the cancer cell type (e.g breast, lung, etc.) and mutations should be possible since fundamentally biology is built on information (RNA and DNA).
So I agree with you that we shouldn't expect a "cancer cure" in the next 6 months.
But since cancer starts in the information-holding part of the cell, building therapies that are information-sensitive with a previously unimagined rapid turnaround, and packaged in a virus which may also be heavily genetically modified to suit this purpose, does seem like exactly the type of problem these new generative manipulators of information are built to solve.
AI can now, or will soon be able to deduce:
* what additional data it needs to gather
* how to conduct measurements to gather that data
* once gathered, what the data portends in terms of the mechanism of action of cancer
* what course of treatment will kill the cancer with a minimum of interference with healthy body cells
All we would have to do is feed it enormous amounts of the right kind of information and it could come up with a treatment that will vastly ameliorate many kinds of cancers. Anthropic is already working on this; they have a wet lab where the biologists are "meat proxies" conducting experiments for Claude which is developing new drug candidates. This is already happening.
I'm realistic enough to understand that a total or near-total cure for pancreatic cancer won't happen tomorrow. But it will still be very soon compared to the time horizon typical of medical research, wherein progress happens very slowly over decades.
Right now AI can check images and give you accurate diagnosis. But that is something already there. We need to prevent, better than cure. And to prevent we need more info about the causes, we need more strict rules about environmental contamination, same as we did with smoking. We need to understand the influence of drugs, chemicals, food, stress, in the development of cancer. We need to put all knowledge in one place in order to allow specialists and geniuses to find solutions. Creating walled silos of knowledge is in benefit of none but a few, that's why open weights, open models, and open knowledge is extremely important for us all
I think the future will be better for those who prevent it, not just cure it. Accurate prevention will make us immune to cancer
I don't understand why you would believe developing a formal proof and developing a successful medical intervention share any relevant commonalities. They share neither problem structure nor solution space and require materially different forms of labor. Math is handy in that it is notably divorced from material and biological reality. Oncology notably less so.
The (somewhat reasonable) hope is that patients 1. with earlier forms of pancreatic cancer (non-metastatic) 2. that can undergo surgery 3. together with a (neo)adjuvant treatment with these kras inhibitors, will end up having a much lower recurrence rate and a chance for a long, healthy life.
I think this is genuinely good news worth celebrating!
None of these responses are terrible, these responses are no doubt true. But in conversation, those are low bars.
These segues come across, and frankly are, very dismissive of the original particular circumstances brought up.
It is best to respond to specific situations, with something thoughtful and specific to them. The specific situation was not brought up as conversation filler or general smalltalk initiation, but for its specific import, so respect that.
Its an easy bad habit to have, but once it is recognized, it is fairly easy to break. And the result is people will find conversations with you much more thoughtful, caring, and pertinent.
Need examples beyond your own? Don't respond to the news that someone lost a leg with amputation statistics, or the time you almost lost a finger. Don't respond to someone's pain with trivia or tautology. Respond with specific comfort or help, or not at all.
It depends on the nuances of the interaction and, as we all know, text is pretty bad at getting across such nuance.
I'm not saying I mind your (just a pinch gatekeepy?) comment, but you know, there's lots of people out there, with different communication styles, and I feel it's generally good to assume the best of others' intentions.
That said, I do agree with the general "pay attention to how your words might be received" gist of what you're saying. Non-sarcastic smiley face (:
Of course the fact that every curve is different is why every patient needs a consultation to decide if surgery, radiation, or chemo are your first stop. Because for people in this range of progression, surgery is less dangerous than chemo. And perhaps radiation to shrink the tumor first will make the tumor operable, which it currently isn’t because of what it’s right next to.
Or if it’s metastatic, maybe all three for different tumors. Excise this lymph node, radiation for the brain, etc.
For someone who's life expectancy after x time doubles, it also means there's more time for more treatments or innovations in treatments because of the time increase.
Unfortunately the problem is somewhat a Devil You Know. We know the consequences and the odds of old treatments, the new ones we don’t even know all of the consequences. So it’s easier to think of them as a miracle.
Of course old and frail people tend to die when they lose hope, so the placebo effect of something on the horizon, from a new treatment to your youngest’s graduation, is important.
Saying that, is it better to live longer knowing you are going to die just a little later? I would just advise people to prepare now for the idea of your death. Memento mori...
But the science, this is a RAS inhibitor which stops GTP from binding to the RAS enzymes which basically slows cell division. Many gene polymorphisms that sloe down RAS activity lead to these cells that are mutated to create higher RAS activity. (I think this is important.)
The GTP is also important before you get the cancer since low GTP will inhibit DNA repair.
https://aacrjournals.org/cancerdiscovery/article-abstract/14...
When deciding what to do today, we all know we have a limited number of days left and should not waste them. Sure 50 years is more than 6 months, but it doesn't change the fact that wasting time is a silly idea for everyone.
It is a big mystery to me why people are so bad at choosing what to do with their time compared to their stated goals. Nobody says "next weekend I plan to binge watch netflix and not get out of bed", yet many do exactly that.
Sounds like you're really good at this. Use that gift and you will go far!