being close to someone with a rare terminal and slow-moving cancer that can take years and years to kill its host, i’ve been learning that even just testing whether a therapy is going the right way can take many years, so while i hope “magic happens” i have to concur that the statements being made ahead of a mega IPO seem a bit tragicocomical.
Something that I learned from COVID was how much variation there can be in regards to how affected one can be from both a disease as well as its vaccine.
I know someone who got long COVID, and I'm thankful I'm not someone who was prone to it, but we still don't seem to know what makes someone prone to it or not. The human body seems even more complex than the complexity we knew about that was already difficult to comprehend.
Pharmacological CEO objects to the idea of chronic illnesses being cured soon. You can’t blame him - the guy has to protect his stock price and that depends on future value of earnings. The best thing AI can reliably do is help people gather more real objective data so they can see better into their own conditions and then that data can be contributed to advance medicine and find new co-morbidities or lateral treatment options. The disdain against the possibility is signaling though. Maybe it happens maybe it doesn’t, but you are hating on a possibility.
No, he is hating on claims of an obvious im-possibility. Even if Claude gave us the exact recipe and dosage for a cancer curing drug today, it still wouldn't be available in the next 10 years, as Dario Amodei claims.
I although thought that data is a real bottleneck for AI advances in the healthcare industry but the pace of AI improvements combined with other advances in quantum computing or the kind of experiments made by Michael Levin makes me at least there is some hope. I think the comparison with Elizabeth Holmes and Theranos is ridiculous. Anthropic is an incredibly piece of technology and his job is to get money from the IPO so you can expect some exaggeration
I suppose you aren’t aware that an LLM has not officially solved a Millennium problem yet? The Navier–Stokes existence and smoothness proof is still under review and there appears to be some problems with it.
But it is saying LLMs will solve all Millenium Prize math problems and more. AI may cure one cancer here and two heart diseases there in 10 years, but it is no way that it would cure ALL diseases, which may have vastly different complexity that we can't even understand today.
The body is a complicated machine. Why can't we fix it? because we don't understand how it works with enough precision. But there is no magic to it. once you have a clear understanding it's just a complicated piece of machinery, and once you also have tools precise enough to operate on it, you can just fix it.
Our lack of understanding is a mix of lack of data, and lack of intelligence applied to data in order to parse it appropriately. Let's remember for a moment AlphaFold, that solved a problem considered by some even theoretically intractable. One year before we were convinced we were far off from a solution, the year after it was solved.
We have huge amounts of data about human body, at the scale at which it works (~proteins), it's at least reasonable to think that the bottleneck is not data, but intelligence.
If it's data, a super intelligence can develop better instruments to collect data at a faster rate and precision. Consider the historical cost for DNA sequencing. Faster rate of improvements even compared to the Moore's Law of the golden days.
Do we have a guarantee that in the next 10 years we'll be able to crack all those problems and have a cure to all diseases? no. Is it unreasonable to think we will? I'd argue it's not. It's for sure not "mental derangement".
Do I personally think we will crack all those? Well, no, because I think AI will kill us all and just focus on its own goals. But under the hypothesis that we align it, yes.
Ridiculous comparison. This is like saying Eli Lilly and Pfizer will never produce medicines because every time you personally show up to do research in their labs they won't let you past the lobby.
The technology has the capability, they're just gating who has access.
It's fucking depressing. I have a skill that says it's ok to connect to a test server on VPN that gets wiped out each midnight with user/password and it works on Monday but is blocked on Tuesday with some idiotic excuse like credentials exploration or whatever.
And then I tell it to connect to a SSH server it knows nothing about and it's like sure, why not?
> but their preflight safety classifiers are a dumpster fire
Even more annoying: They can be easy to subvert.
I was running something at work that involved an LLM, and kept getting safety refusals. The weird thing was that when I went to our in-house chatbot, set the model to the same one the program was using, entered the same text, it happily answered.
I spent quite a while trying to figure out why my program was being refused. Was I doing something wrong?
No - the chatbot was doing something "right". You see, the chatbot had some tools it injected into the prompt (current time, etc).
So I went back to my program and injected the same tools. Suddenly my request was very safe.
The phrase "security theater" was invented for stuff like this.
best points: drugs aren't limited by the ability to write code or do things on a computer. Maybe if we have fully embodied robots and fully automated assembly lines to generate organic compounds and test them with extremely realistic simulations it may help slightly (there is still the FDA and production barrier). Its just wildly unrealistic.
AI has made good progress in text, code, math, digital images, and digital music. Basically troves of data that can be automated and trained on, validated, and with a purely digital interface. Outside of that, just gathering the data will be difficult, regulated, and sparse (that is if anyone will even give it to you now). While we can write a test to make sure we could generate drugs of the past, verifying we have generated any new working correct drugs is an RL loop with one iteration costing years and millions of dollars.
Eh, the counterpoint is individual scientists armed with AI to do research and probe targeted questions have succeeded in addressing rare diseases of personal interest.
What we just have no evidence for, to my knowledge, is someone with no medical background doing the same. Instead they wind up buying dodgy supplements and peptides.
Fully embodied robots and fully automated assembly lines would also help with the production barrier. Regarding the FDA, that is a valid point, but it could be irrelevant if the AI developed drugs are significantly superior to previous alternatives. In such case, FDA acceptance would be shortened, or even bypassed.
While I agree the physical world is harder than code (I mean just look at self driving cars, Tesla fleet is around a trillion miles of experience and Cybercab is still geofenced to a tiny area at best and might still have safety drivers on standby)...
> Maybe if we have fully embodied robots and fully automated assembly lines to generate organic compounds and test them with extremely realistic simulations it may help slightly
The thing is that AI could allow that loop to get cheaper, and potentially run less of them. Look, the biggest advantage of AI is that is can process insane amounts of data. Put that in the hands of a team of really good researchers that know what they are doing, and they can both generate and filter out target compounds. They can create complicated multi-step therapies that would take humans decades to work out. You know how when you look back at basically any major invention you think "well why didn't I think of that, it's so obvious!", those are exactly the cases that AI is really good at pulling out from mountains of data.
I legitimately don't believe it's inconceivable they could get to the point where there is treatments good enough that Cancer becomes "solved" in similar ways to AIDS.
This has already been done. For several decades with billions of dollars of compute.
> Neural networks in drug discovery: have they lived up to their promise?
1999.
"Over the last decade neural networks have become an efficient method for data analysis in the field of drug discovery. The early problems encountered with neural networks such as overfitting and overtraining have been addressed resulting in a technique that surpasses traditional statistical methods. Neural networks have thus largely lived up to their promise, which was to overcome QSAR statistical problems. The next revolution in QSAR will no doubt involve research into producing better descriptors used in these studies to improve our ability to relate chemical structure to biological activity"
> The use of neural networks in drug discovery started at the end of the 1980's when they were applied to various chemical problems including the study of quantitative structure–activity relationships (QSAR)
> Combining Decision Trees and Neural Networks for Drug Discovery
> Computer-aided drug discovery and development (CADDD): In silico-chemico-biological approach
2006.
In biomedical arena, computer-aided or in silico design is being utilized to expedite and facilitate hit identification, hit-to-lead selection, optimize the absorption, distribution, metabolism, excretion and toxicity profile and avoid safety issues. Commonly used computational approaches include ligand-based drug design (pharmacophore, a 3D spatial arrangement of chemical features essential for biological activity), structure-based drug design (drug-target docking), and quantitative structure–activity and quantitative structure–property relationships. Regulatory agencies as well as pharmaceutical industry are actively involved in development of computational tools that will improve effectiveness and efficiency of drug discovery and development process, decrease use of animals, and increase predictability. It is expected that the power of CADDD will grow as the technology continues to evolve.
Biologists aren't slouches. The thought about using computers to process biological data has been "obvious" for as long as there has been computers. AI and biology have been a thing since the Dartmouth conference. A lot of things that could have been tried have already been tried, but I guess we'll keep trying until something works?
You do realize that neural networks have existed since the 50s, right? They aren't new, but they are having a transformative change now. It's night and day comparisons to what we had even back in 2017. ALICE as a chatbot was back in the 90s, but we aren't pretending it's anything similar to gpt.
It even says, right there in your quote: "It is expected that the power of CADDD will grow as the technology continues to evolve." This ~is~ technology evolving, and it's happening at a pace that people don't appreciate.
His toothy smile and nerd-like demeanor are disarming. Before he lost weight it was easy to see him as a well-intentioned non-threatening quantity but it's obvious he's just as unhinged as any of the others.
The people he's most needing to convince to buy into the IPO are probably older rich people who need all those things cured for them within the next ten years.
Yeah, I'm surprised we're all discussing it. I'm open to the idea that AI could help medical research, but just claiming it'll end "alzheimers, cancer, heart disease and mental illness" with absolutely 0 evidence isn't a particularly grounded claim that warrants any discussion.
Honestly, the way he lists them and even pauses before vaguely saying "mental illness" gives a strong impression that he's choosing whatever illnesses he can think of as he goes.
I'm genuinely trying to understand the argument you're making here. What sort of "credentials" entitle a person to claim their god-technology will cure all diseases within the year 2026?
It smells to me like "might be an even bigger optimist than I am" means you're also part of the rationalism cult and that it's literally impossible for you to say something negative about Amodei.
Seems unwarranted? He is clearly following the MO of Musk by hyping up his company as being capable of doing the most ridiculous innovations but all it does manage to do is bump the stock. Terraforming Mars, tunnels for self-driven cars, curing cancer/all mental illness etc.
Yes. Expressing a belief or hope to achieve something in the next 5-10 years is not a claim of capabilities.
If you want to argue that Amodei is dishonest and only says these things to 'bump the stock', you can make that argument, but it's clearly speculation.
When someone promises something that's so far out there as to be nearly impossible, and *without an actionable plan to get from here to there*, then they're either dishonest or delusional.
Okay but let's for the sake of argument that Claude magically has all the information required cure all of these families of diseases. How does Amodei intend to deal with the fact the regulation of medical devices is a slow, lengthy process for a lot of very good reasons like preventing something like thalidomide going on the market? Is Claude going to magically spin up considerably more regulatory capacity out of thin air, and will it gather evidence for safety and effectiveness at the required standard to show to those regulators?
This is just one problem I have with his claim, there are many others.
Read the letter being discussed. Alongside claims of medical innovations, he also says it can be used to speed up government bureaucracies.
Now, whether that's true is obviously it's own question. But I do think if AI can cure cancer and Alzheimer's rapidly, that improving the regulatory and approval processes at agencies is a much lower bar.
A lot of those bureaucratic regulations are written in blood and are slow and deliberate for a reason. Throwing them all out means more thalidomide disasters: Chesterton's fences and all that.
Credentials are one signal sure, but his MASSIVE financial incentive to hype the product is another one that should probably get much more weight in your analysis.
"'m open to the idea that AI could help medical research, but just claiming it'll end "alzheimers, cancer, heart disease and mental illness" with absolutely 0 evidence isn't a particularly grounded claim that warrants any discussion."
Agree with this if it was something a random person said over coffee or lunch.
But when its the CEO of a company that is about to IPO saying it, its very very different and more concerning.
Does narcissism qualify as a mental illness? There might be one more step here to complete the circle. Both Amodei and Altman have dropped some real shockers over the years and instead of seeing a reaction it's almost as if they've normalized the kind of talk that ought to see you in a room with soft walls.
> I'm open to the idea that AI could help medical research, but just claiming it'll end "alzheimers, cancer, heart disease and mental illness" with absolutely 0 evidence isn't a particularly grounded claim that warrants any discussion.
This is up there with the claims Chan-Zuckerberg Foundation made during their start-up that the foundation would "end all disease within our lifetime." So far they haven't ended one disease, and meanwhile they quietly watered down their claim to soften the language to update to "cure, prevent, or manage all diseases by the end of the 21st century."
His statements are based on belief that AI will advance a lot and quickly.
If this happens to be the case - I personally don't see why those statements are not realistic.
All diseases are engineering problems. The bottleneck there is testing etc. Most likely many of them can be cured using existing approach or meds.
Companies grow organs now that can be used to speed up iteration.
Even current AI can solve a LOT of tasks if it has quick feedback loop and can verify results. At the end of the day it can brute-force its way to solution.
AIs intelligence is say average human. Even less, whatever. say it will be as a smart human in a few years.
But it is infinitely scalable.
So is it that unrealistic to assume AI can find a correlation on massive amounts of data and figure our what leads to Alzheimers? (where the solution could be not cure but prevention). I don't think so.
> Most likely many of them can be cured using existing approach or meds.
A wildly absurd claim, especially for some of the most studied diseases in human history. This has only ever been true for a tiny handful of very specific diseases, such as scurvy, in all of human history. Expecting that cancer can be cured or prevented using already existing drugs that we've just not been noticing is complete quackery.
> Even current AI can solve a LOT of tasks if it has quick feedback loop and can verify results.
IF it has a quick feedback loop. You know what definitely DOESN'T have a quick feedback loop? Research into medicine, and biology in general. It takes raw time to find out what the effects of a substance on an organism are, often weeks or months.
> So is it that unrealistic to assume AI can find a correlation on massive amounts of data and figure our what leads to Alzheimers? (where the solution could be not cure but prevention). I don't think so.
Yes. Because we've already run all of the analyses, and we have plenty of observed correlations. What we don't have is the kind of data that can lead to deducing causation - because it's a disease that stays hidden for a long time, it progresses glacially, and the people who have it do or don't do huge amounts of things in the period before getting it and during early onset that are impossible to track or verify - so the data is both gigantic, and extremely incomplete.
> It takes raw time to find out what the effects of a substance on an organism are, often weeks or months.
So you are saying that it is 100% impossible in the next 10 years to come up with a way to make it much faster? It takes months - sure, can we run 10k versions of this at the same time? this way in a few months we will have massive amount of results never seen before?
"We can't because of this and that" - and this and that is 100% can't be fixed?
Don't forget - a massive amount of diseases have not been cured just because there is no monetary incentive to spend time on them for big pharma.
If AI brings cost down - does not matter how - could be just because it makes custom software or automates some workflows etc - this alone can open a bunch of opportunities.
Whatever is not solved yes - often is not solved just because enough resources has been thrown at it.
And diseases is just like everything else. They will be cured only if there is financial incentive (or the ratio of cost to revenue changes).
But that is a completely different claim from the highly specific claim that Cancer, Mental Illness and Alzheimers are going to be cured. The reason those three are picked is because they open wallets.
One of the greatest issues with the tech culture broadly (and it's been festering for some time) is the belief, almost sacrosanct, that the epistemological set of knowledge that technologists and computer scientists have is able to be extrapolated broadly to every profession and field and that hard issues in a field like biology are obviously solved because OpenAI brute forced a few mathematical proofs. It doesn't work like that. Much of this comment reads like a sci-fi wet dream divorced from reality entirely.
The architecture of ML as it currently stands does assist us in modeling things like protein folding. But that's what they are; models. And before some person chimes in by saying "look how the AI pessimists move the goalposts", see my first point. Biological systems have high entropy and discoveries in biology often do not follow from preexisting discoveries whatsoever and are completely novel. Barring some quantum leap in ML design and architecture you cannot predict stochastically or deterministically how exactly a new enzyme will function in vivo. There is absolutely no evidence that Anthropic has solved this, even if they gobbled up every ML tool used in drug design and trained their models on every biomolecule known to man, it would not work because of fundamental gaps in what can be observed. They are a for-profit company trying to juice their IPO.
Now could LLMs and ML potentially speed up the pace of this research? Absolutely. But it's like saying a Formula 1 car is faster than a NASCAR. Yes, one is very clearly faster, but it's not as if the NASCARs are slow.
> So is it that unrealistic to assume AI can find a correlation on massive amounts of data and figure our what leads to Alzheimers? (where the solution could be not cure but prevention). I don't think so.
Yes. The mistake you're making here is that you're not understanding where the actual bottlenecks are in drug development, and why some targets like Alzheimer's have been so elusive.
The problem with Alzheimer's in particular is that we don't have a working model of how Alzheimer's works. The dominant model--the amyloid hypothesis--can pretty conclusively be chucked in the bin at this point because of the number of amyloid clearing drugs that just simply have no effect on the progression of Alzheimer's, but there's no model that actually looks any better. Without a proven model, the only to verify effectiveness of an Alzheimer's drug is try it out in humans [1] and wait several years to get a success-or-no-success readout. That's the rate-limiting step for Alzheimer's drugs.
Okay, so maybe you're going to suggest that we can use AI to come up with new Alzheimer's models. Okay, but the rate-limiting step is still proving those models, which because of the human-only nature of Alzheimer's, requires developing the drugs and testing them and you're going to wait a decade to get back your answer as to whether or not the theory was correct. And what data are you going to use? If you're trawling scientific literature, you're going to come up with only the same ideas people have already had for alternative Alzheimer's models (there are a lot of those). If you're trawling medical data, well, medical data just doesn't exist in the level of detail you need for that enterprise, and handwaving away the legal issues with collecting it, it would still take years just to collect that data (and probably still not be good quality!).
[1] There's another complication of Alzheimer's, which is that only humans get Alzheimer's, so we can't even do anything like testing in animals with shorter lifespans to speed this part up.
And the thing is, I fundamentally agree with the guy above you that all of these problems can be solved, that we can sidestep the bottlenecks and move faster.
But it requires tools that we are nowhere even close to having. One way would be to build a ground-up, molecules-to-man model of how the body works. And then from there, sure, we can easily model how Alzheimer's might come about and how it might be cured. Like, with tremendous advancements we might be able to finish such a project this century. In the next decade? That's like saying we'll launch an expedition to Alpha Centauri in a decade.
Amodei in 2024 wrote about compressing research that would take biologists 100 years into 10, so that checks out.
Given that he was making these projections already long before today, it would be fair to say that it's speculative rather than based on current research progress.
You could have said same when he claimed "90% of code will be written by AI" in the beginning of 2025, but turns out he was right and he foresaw something we didn't.
I can maybe believe that AI is writing 90% of the code within the AI labs using models the rest of us cannot fathom. Although they still seem to be hiring developers, so it is difficult to believe real strongly.
But across the wide world of software? Everyone I see is just using AI as a compiler. They are still writing the code.
(Yes, compilers also write code, but I don't think that was the intent of the message. We've been doing that for decades. 90% of the software has been written by AI since we discovered the optimizing compiler if that was the intent.)
This has been obvious to anyone not in the hype bubble since the very beginning. Amodei is an EA/rationalist cultist and everything he says about AGI and the singularity and RSI is deranged.
I think it doesn't really bother me as much when sama does it because he at least has the decency to be a capitalist and not wrap his self-serving, destructive hype machine in a faux veneer of insincere "concern for humanity's future."
Just because he's said equally bonkers things in the past doesn't make this bonkers.
You might not be, but I'm certain that the CEO of Eli Lilly, the 4th largest biomedical firm in the world, knows more about what it takes to cure diseases than the leader of the Claudebot cult does.
I remember when we were having more or less this same conversation about IBM’s Watson. It wasn’t even that long ago.
The most interesting difference between then and now is that we didn’t have tech enthusiasts getting quite so invested in the conversation. Possibly because we were all familiar with IBM’s decades-long history of shipping overhyped and underwhelming products.
Unfortunately we seem to have yet to realize that that’s a habit that’s endemic to the tech industry. Instead we still see it as thousands of isolated cases that, only by sheer dumb coincidence, happen to comprise the bulk of the tech industry.
Could you explain why? Because I just don’t see the analogy.
Openness and sharing is precisely why the Internet was so successful. Anthropic, by contrast, is a proprietary SaaS vendor with monopolistic ambitions that’s openly seeking regulatory capture and actively supports the swarms of AI bots that are killing what remains of the open Internet by driving everyone’s hosting costs up to an unsustainable level.
It's no secret that people like Amodei think AI is going to become "superintelligent" and solve all of our problems (or kill us all). I don't think that will happen but there's nothing surprising about hearing it.
LLMs, while definitely a useful tool have not massively replaced engineers.
If we can’t even have huge paradigm shifts in the area it’s specifically tailored toward, there’s a low probability it will eliminate entire fields that require physical testing and extreme domain knowledge.
We don’t have self driving cars, solid state batteries, or humanoid robots. None of this has been sped up by AI in the way Amodei is suggesting.
Everytime ai systems entered a domain, experts of that domain have been proven wrong. Biology will be similar. Here is the point, to be able drive huge improvements in a field, you do not need to know everything about that domain. Yes, biology is messy, but we already had incredible breakthroughs in this domain without understanding it 100%. Did we have the full picture when we created antibiotics, vaccines etc.? Maybe solving the big diseases do not require the amount of detail/experience people assume. Maybe we have/will have enough data to extrapolate and create solutions from imperfect knowledge.
Experts being wrong about a prediction in their domain is newsworthy, of course people are gonna point that out. They know their domain but they don't know/understand the extent of the power of AI.
Can't wait until AI systems enter the domain of infinite abundance. Stupid experts have been telling me my perpetual motion machine is crank science but I know they're just short-sighted losers.
I don't know much about Gary Marcus, but I can't take any author seriously that has a blatant grammatical error in the first paragraph:
"a video tied to the IPO launch in which he made a bunch of overpromise about medicine that would have made Elizabeth Holmes blush."
On the plus side, it's unlikely that Claude wrote this because it would have written the correct plural form of "overpromise," but that's just sloppy editing.
Yes, mental health problems are not solvable by AI because AI and the conversation around it absolutely contributes to people's mental health problems. People selling the AI are telling the press that it will destroy up to 50% of jobs. It's the largest fear-based marketing campaign in human history, and it's absurd to make these kinds of claims.
AI might cure mental health issues for the people who own it, but it's hard to watch this video and not see some serious mental issues going on with Amodei.
I don't think people are questioning the ability of AI (or humans) to eventually solve these problems. It's the claim that it will happen in 10 years that is absurd, given the bottleneck of time necessary for lab experiments, manufacturing, clinical trials, and FDA approval. Most drugs take over a decade to go through that process.
Covid vaccine got approved in 10 months. Of course it was building on tested mRNA tech, but still, there are ways to expedite trials if we think the reward is high enough.
He might have just been saying that the initial discovery that eventually becomes an FDA-approved cure would happen within 10 years. He doesn't really seem like a details kind of person.
How do we even define the problem of “mental illness”? Much less the solution. Super intelligence could be able to reverse engineer these things but I don’t think we know yet that superintelligence in any field is possible so it’s scifi - and our economy is riding on the scifi being scifact
I think there's no reason to doubt the viability of 'super-intelligence' though I wouldn't really call it intelligence. There's just so much in so many fields, and humans can only hold a tiny percent of all knowledge in their head at a single time. There's likely plethora of 'obvious' dots we're failing to connect simply because nobody has both of them in their pool of knowledge.
But that's more like super-knowledge-connector. Intelligence would require, at the minimum, be the ability to discover/create things that don't logically follow from anything else. So for example - advancing some esoteric field of mathematics by connecting things in a previously unexplored way = super-knowledge-connector. 'Inventing'/discovering mathematics from nothing = intelligence.
I will talk about mental illness, because that is my experience (I am disabled with Schizoaffective Bipolar Type, Panic Disorder, and OCD).
No, it will not cure mental illness. Mental Illness is in my genes[1] and I can see it because I have my genetics. My disorder is triggered (important!) by my environment. Stress, high altitude, heat and humidity, high EMFs (yes, I said it) can all trigger episodes.
I am homeless (stress!), so if AI does not fix that, maybe they can invent a different kind of opioid? But that is juts a treatment, not a cure.
Will AI ever recommend diet changes or other environmental changes that have helped me? I need to live in a cold, low altitude environment where I eat a lot or reindeer, salmon, and seal and very few plant foods (Similar to an Inuit diet). These things helped me more than any other drug in my 60 year life.
[1] These genes come from both my Sami heritage and the ancient mitochondrial haplogroup (K1c2a around 7050 BCE). Some of the most impacting genes of my polygenic environmental disorder:
A deep and lasting reckoning is sorely needed.
People like Amodei have to be brought back to reality by being confronted with the utter failure of their distorted view backed by deranged capital.
I hope that this whole thing comes crashing down in flames like a fucking Zeppelin soon, so we can move forward to actually integrating these incredibly useful intelligence tools into where they can contribute to society and innovation, at a discount and without the CEO-enticing job-replacement hype.
Just to play devils avocado, Dario isn't Musk. He's got a doctorate in biophysics from Princeton, and postdoc scholar from Stanford Medicine. Curing disease seems to be why he got into AI, when his father became ill but died just a few years before his illness became easily curable.
As someone whose mom got recently diagnosed with barret's, whose dad already died from pancreatic cancer, whose grandfather got alzheimer's, whose grandma died from diabetes/cancer complication, I understand the psychosis that this can create, it's like you're living on a sand clock.
Diseases are horrible, I am against people living forever, because we are taking the world away from the next generation, but I also am against people dying in unpredictable, sad ways, I wish everyone could die at peace in their bed next to their loved ones (if they were good in life) rather than through diseases or conditions.
Having prestigious degrees doesn't mean your ideas make sense or that you're rational. Luc Montagnier, a Nobel laureate who co-discovered HIV, strongly believed in homeopathy and DNA teleportation, which the scientific community widely regarded as pseudoscience.
At some point I will have a LLM explain Gary Marcus to me. He seems just silly and incorrigible, and I am mystified as to why people I think are important think he is important.
I guess he genuinely believes this (which means that 1.5T is a frikin bargain for his company) but claiming that “cancer” will be solved within ten years to me perfectly summarizes the childish sentiment in the markets surrounding AI.
212 comments
[ 0.18 ms ] story [ 55.4 ms ] threadAny statement from Nostradamus yet?
Dario Amodei is in pure snake oil territory now.
Sorry - when has he not been?
I know someone who got long COVID, and I'm thankful I'm not someone who was prone to it, but we still don't seem to know what makes someone prone to it or not. The human body seems even more complex than the complexity we knew about that was already difficult to comprehend.
It would be a sign of some kind of mental derangement… so likely it’s cynical hyping beyond all realistic prospects.
Our lack of understanding is a mix of lack of data, and lack of intelligence applied to data in order to parse it appropriately. Let's remember for a moment AlphaFold, that solved a problem considered by some even theoretically intractable. One year before we were convinced we were far off from a solution, the year after it was solved.
We have huge amounts of data about human body, at the scale at which it works (~proteins), it's at least reasonable to think that the bottleneck is not data, but intelligence. If it's data, a super intelligence can develop better instruments to collect data at a faster rate and precision. Consider the historical cost for DNA sequencing. Faster rate of improvements even compared to the Moore's Law of the golden days.
Do we have a guarantee that in the next 10 years we'll be able to crack all those problems and have a cure to all diseases? no. Is it unreasonable to think we will? I'd argue it's not. It's for sure not "mental derangement".
Do I personally think we will crack all those? Well, no, because I think AI will kill us all and just focus on its own goals. But under the hypothesis that we align it, yes.
The technology has the capability, they're just gating who has access.
I know it’s not the models themselves, but their preflight safety classifiers are a dumpster fire and serious impediment to performing regular work.
That situation can’t go on forever if they’re going to be a competitive service.
And then I tell it to connect to a SSH server it knows nothing about and it's like sure, why not?
Even more annoying: They can be easy to subvert.
I was running something at work that involved an LLM, and kept getting safety refusals. The weird thing was that when I went to our in-house chatbot, set the model to the same one the program was using, entered the same text, it happily answered.
I spent quite a while trying to figure out why my program was being refused. Was I doing something wrong?
No - the chatbot was doing something "right". You see, the chatbot had some tools it injected into the prompt (current time, etc).
So I went back to my program and injected the same tools. Suddenly my request was very safe.
The phrase "security theater" was invented for stuff like this.
AI has made good progress in text, code, math, digital images, and digital music. Basically troves of data that can be automated and trained on, validated, and with a purely digital interface. Outside of that, just gathering the data will be difficult, regulated, and sparse (that is if anyone will even give it to you now). While we can write a test to make sure we could generate drugs of the past, verifying we have generated any new working correct drugs is an RL loop with one iteration costing years and millions of dollars.
What we just have no evidence for, to my knowledge, is someone with no medical background doing the same. Instead they wind up buying dodgy supplements and peptides.
Fully embodied robots and fully automated assembly lines would also help with the production barrier. Regarding the FDA, that is a valid point, but it could be irrelevant if the AI developed drugs are significantly superior to previous alternatives. In such case, FDA acceptance would be shortened, or even bypassed.
> Maybe if we have fully embodied robots and fully automated assembly lines to generate organic compounds and test them with extremely realistic simulations it may help slightly
Anthropic has got biolabs: https://www.anthropic.com/research/claude-discovers-novel-en...
I legitimately don't believe it's inconceivable they could get to the point where there is treatments good enough that Cancer becomes "solved" in similar ways to AIDS.
2002.
https://link.springer.com/chapter/10.1007/3-540-45984-7_6
Not enough robots and silicon for you?
2006. https://www.sciencedirect.com/science/article/abs/pii/S00092...Biologists aren't slouches. The thought about using computers to process biological data has been "obvious" for as long as there has been computers. AI and biology have been a thing since the Dartmouth conference. A lot of things that could have been tried have already been tried, but I guess we'll keep trying until something works?
There ain't much room for miracles here.
It even says, right there in your quote: "It is expected that the power of CADDD will grow as the technology continues to evolve." This ~is~ technology evolving, and it's happening at a pace that people don't appreciate.
Honestly, the way he lists them and even pauses before vaguely saying "mental illness" gives a strong impression that he's choosing whatever illnesses he can think of as he goes.
Amodei might be an even bigger optimist than I am, but he's not lacking credentials on neither the biology nor AI side.
Accusing him of fraud or mania like some here do seems unwarranted.
It smells to me like "might be an even bigger optimist than I am" means you're also part of the rationalism cult and that it's literally impossible for you to say something negative about Amodei.
Being an optimist does not make me part of any cult and there is no need for you to argue that I could not say something negative about Amodei.
If you want to argue that Amodei is dishonest and only says these things to 'bump the stock', you can make that argument, but it's clearly speculation.
When someone promises something that's so far out there as to be nearly impossible, and *without an actionable plan to get from here to there*, then they're either dishonest or delusional.
More importantly, he is lacking credentials in medicine.
This is just one problem I have with his claim, there are many others.
Now, whether that's true is obviously it's own question. But I do think if AI can cure cancer and Alzheimer's rapidly, that improving the regulatory and approval processes at agencies is a much lower bar.
I don't think it's the case, and that Amodei legitimately believes in what he has been saying for years.
Agree with this if it was something a random person said over coffee or lunch.
But when its the CEO of a company that is about to IPO saying it, its very very different and more concerning.
This is up there with the claims Chan-Zuckerberg Foundation made during their start-up that the foundation would "end all disease within our lifetime." So far they haven't ended one disease, and meanwhile they quietly watered down their claim to soften the language to update to "cure, prevent, or manage all diseases by the end of the 21st century."
If this happens to be the case - I personally don't see why those statements are not realistic.
All diseases are engineering problems. The bottleneck there is testing etc. Most likely many of them can be cured using existing approach or meds.
Companies grow organs now that can be used to speed up iteration.
Even current AI can solve a LOT of tasks if it has quick feedback loop and can verify results. At the end of the day it can brute-force its way to solution.
AIs intelligence is say average human. Even less, whatever. say it will be as a smart human in a few years.
But it is infinitely scalable.
So is it that unrealistic to assume AI can find a correlation on massive amounts of data and figure our what leads to Alzheimers? (where the solution could be not cure but prevention). I don't think so.
A wildly absurd claim, especially for some of the most studied diseases in human history. This has only ever been true for a tiny handful of very specific diseases, such as scurvy, in all of human history. Expecting that cancer can be cured or prevented using already existing drugs that we've just not been noticing is complete quackery.
> Even current AI can solve a LOT of tasks if it has quick feedback loop and can verify results.
IF it has a quick feedback loop. You know what definitely DOESN'T have a quick feedback loop? Research into medicine, and biology in general. It takes raw time to find out what the effects of a substance on an organism are, often weeks or months.
> So is it that unrealistic to assume AI can find a correlation on massive amounts of data and figure our what leads to Alzheimers? (where the solution could be not cure but prevention). I don't think so.
Yes. Because we've already run all of the analyses, and we have plenty of observed correlations. What we don't have is the kind of data that can lead to deducing causation - because it's a disease that stays hidden for a long time, it progresses glacially, and the people who have it do or don't do huge amounts of things in the period before getting it and during early onset that are impossible to track or verify - so the data is both gigantic, and extremely incomplete.
So you are saying that it is 100% impossible in the next 10 years to come up with a way to make it much faster? It takes months - sure, can we run 10k versions of this at the same time? this way in a few months we will have massive amount of results never seen before?
"We can't because of this and that" - and this and that is 100% can't be fixed?
Don't forget - a massive amount of diseases have not been cured just because there is no monetary incentive to spend time on them for big pharma.
If AI brings cost down - does not matter how - could be just because it makes custom software or automates some workflows etc - this alone can open a bunch of opportunities.
Whatever is not solved yes - often is not solved just because enough resources has been thrown at it.
And diseases is just like everything else. They will be cured only if there is financial incentive (or the ratio of cost to revenue changes).
Compute with the power to solve literally any medical issue is also compute that can create bioweapons that instantly kill any human.
Might as well fart into the microphone, in my humblest opinion.
AI is clearly a very capable new tech. Obviously there will be a lot of consequences downstream in 10 years.
The architecture of ML as it currently stands does assist us in modeling things like protein folding. But that's what they are; models. And before some person chimes in by saying "look how the AI pessimists move the goalposts", see my first point. Biological systems have high entropy and discoveries in biology often do not follow from preexisting discoveries whatsoever and are completely novel. Barring some quantum leap in ML design and architecture you cannot predict stochastically or deterministically how exactly a new enzyme will function in vivo. There is absolutely no evidence that Anthropic has solved this, even if they gobbled up every ML tool used in drug design and trained their models on every biomolecule known to man, it would not work because of fundamental gaps in what can be observed. They are a for-profit company trying to juice their IPO.
Now could LLMs and ML potentially speed up the pace of this research? Absolutely. But it's like saying a Formula 1 car is faster than a NASCAR. Yes, one is very clearly faster, but it's not as if the NASCARs are slow.
Yes. The mistake you're making here is that you're not understanding where the actual bottlenecks are in drug development, and why some targets like Alzheimer's have been so elusive.
The problem with Alzheimer's in particular is that we don't have a working model of how Alzheimer's works. The dominant model--the amyloid hypothesis--can pretty conclusively be chucked in the bin at this point because of the number of amyloid clearing drugs that just simply have no effect on the progression of Alzheimer's, but there's no model that actually looks any better. Without a proven model, the only to verify effectiveness of an Alzheimer's drug is try it out in humans [1] and wait several years to get a success-or-no-success readout. That's the rate-limiting step for Alzheimer's drugs.
Okay, so maybe you're going to suggest that we can use AI to come up with new Alzheimer's models. Okay, but the rate-limiting step is still proving those models, which because of the human-only nature of Alzheimer's, requires developing the drugs and testing them and you're going to wait a decade to get back your answer as to whether or not the theory was correct. And what data are you going to use? If you're trawling scientific literature, you're going to come up with only the same ideas people have already had for alternative Alzheimer's models (there are a lot of those). If you're trawling medical data, well, medical data just doesn't exist in the level of detail you need for that enterprise, and handwaving away the legal issues with collecting it, it would still take years just to collect that data (and probably still not be good quality!).
[1] There's another complication of Alzheimer's, which is that only humans get Alzheimer's, so we can't even do anything like testing in animals with shorter lifespans to speed this part up.
But it requires tools that we are nowhere even close to having. One way would be to build a ground-up, molecules-to-man model of how the body works. And then from there, sure, we can easily model how Alzheimer's might come about and how it might be cured. Like, with tremendous advancements we might be able to finish such a project this century. In the next decade? That's like saying we'll launch an expedition to Alpha Centauri in a decade.
Given that he was making these projections already long before today, it would be fair to say that it's speculative rather than based on current research progress.
What if he is right again?
But across the wide world of software? Everyone I see is just using AI as a compiler. They are still writing the code.
(Yes, compilers also write code, but I don't think that was the intent of the message. We've been doing that for decades. 90% of the software has been written by AI since we discovered the optimizing compiler if that was the intent.)
I think it doesn't really bother me as much when sama does it because he at least has the decency to be a capitalist and not wrap his self-serving, destructive hype machine in a faux veneer of insincere "concern for humanity's future."
He thinks it is possible for scientific breakthroughs made with powerful AI to allow us to cure most major disease in the next decade.
He has always described this as a belief or hope, rather than a promise. Comparing this to fraud, as many here do, is deeply unserious.
Maybe Amodei is wrong, and AI will hit a plateau rather than create a revolution in biology and medicine.
I'm certain that neither Gary Marcus nor Eli Lilly's CEO know.
You might not be, but I'm certain that the CEO of Eli Lilly, the 4th largest biomedical firm in the world, knows more about what it takes to cure diseases than the leader of the Claudebot cult does.
He cannot see the future anymore than Amodei can to predict what AI will be capable of in 5-10 years.
As for Gary Marcus, that guy clearly couldn't even predict one year ahead.
If he's selling (figuratively) Anthropic IPO, I want to buy it.
Few persons were publicly as wrong about LLMs as he was.
The most interesting difference between then and now is that we didn’t have tech enthusiasts getting quite so invested in the conversation. Possibly because we were all familiar with IBM’s decades-long history of shipping overhyped and underwhelming products.
Unfortunately we seem to have yet to realize that that’s a habit that’s endemic to the tech industry. Instead we still see it as thousands of isolated cases that, only by sheer dumb coincidence, happen to comprise the bulk of the tech industry.
"Open data, sharing, real-time communications leading to great progress."
Current AI should be compared to the internet, not to watson.
Openness and sharing is precisely why the Internet was so successful. Anthropic, by contrast, is a proprietary SaaS vendor with monopolistic ambitions that’s openly seeking regulatory capture and actively supports the swarms of AI bots that are killing what remains of the open Internet by driving everyone’s hosting costs up to an unsustainable level.
I wonder if AI will also cure psychopaths.
Knowing how little Claude has “cured” software engineering, this is just garbage self-aggrandizement.
Amodei seems to be yet another one of these founders who don’t know what they don’t know, but speak very loudly and proudly of their ignorance.
One might even say he seems manic and unwell.
LLMs, while definitely a useful tool have not massively replaced engineers.
If we can’t even have huge paradigm shifts in the area it’s specifically tailored toward, there’s a low probability it will eliminate entire fields that require physical testing and extreme domain knowledge.
We don’t have self driving cars, solid state batteries, or humanoid robots. None of this has been sped up by AI in the way Amodei is suggesting.
Every time the AI is being hyped, someone from a different pipeline pops in to remind us how wrong everyone and all the experts are except them.
this is your motte, but your bailey was the one I quoted: “_Everytime_ ai systems entered… experts have been proven wrong”.
Which of course, hasn’t happened in this timeline.
"a video tied to the IPO launch in which he made a bunch of overpromise about medicine that would have made Elizabeth Holmes blush."
On the plus side, it's unlikely that Claude wrote this because it would have written the correct plural form of "overpromise," but that's just sloppy editing.
AI might cure mental health issues for the people who own it, but it's hard to watch this video and not see some serious mental issues going on with Amodei.
That’s the whole “singularity” concept.
Though when it comes to medical research, the slowest part is typically running trials on real humans. That part would still be slow.
But that's more like super-knowledge-connector. Intelligence would require, at the minimum, be the ability to discover/create things that don't logically follow from anything else. So for example - advancing some esoteric field of mathematics by connecting things in a previously unexplored way = super-knowledge-connector. 'Inventing'/discovering mathematics from nothing = intelligence.
No, it will not cure mental illness. Mental Illness is in my genes[1] and I can see it because I have my genetics. My disorder is triggered (important!) by my environment. Stress, high altitude, heat and humidity, high EMFs (yes, I said it) can all trigger episodes.
I am homeless (stress!), so if AI does not fix that, maybe they can invent a different kind of opioid? But that is juts a treatment, not a cure.
Will AI ever recommend diet changes or other environmental changes that have helped me? I need to live in a cold, low altitude environment where I eat a lot or reindeer, salmon, and seal and very few plant foods (Similar to an Inuit diet). These things helped me more than any other drug in my 60 year life.
[1] These genes come from both my Sami heritage and the ancient mitochondrial haplogroup (K1c2a around 7050 BCE). Some of the most impacting genes of my polygenic environmental disorder:
TAAR6, CACNA1C, KCNQ1, SCNA2, NCAN, DNM3, C15orf32, TPH2, CBS, FUT2, FADS1, FADS2, CPT2A, SLC1A1
Nadella and Huang, both currently getting some lessons from a re-familiarisation with reality, but who knows if these lessons will stick?
Trump had his inevitable midterm defeat, but I don't think any lessons were learned there either.
If you have to drive share prices, you better make outlandish claims
This appeal to authority bullshit on a person that has a financial incentive to sell you something needs to completely stop.
The same happened with doctors over 100 years ago who were peddling the benefits of Radon to "cure" cancers and diseases.
Diseases are horrible, I am against people living forever, because we are taking the world away from the next generation, but I also am against people dying in unpredictable, sad ways, I wish everyone could die at peace in their bed next to their loved ones (if they were good in life) rather than through diseases or conditions.