Transcript [MUSIC]      [BOOK PASSAGE]   PETER LEE: “In ‘The Little Black Bag,’ a classic science fiction story, a high-tech doctor’s kit of the future is accidentally transported back to the 1950s, into the shaky hands of a washed-up, alcoholic doctor. The ultimate medical tool, it redeems the doctor wielding it, allowing him to practice gratifyingly heroic medicine. … The tale ends badly for the doctor and his treacherous assistant, but it offered a picture of how advanced technology could transform medicine—powerful when it was written nearly 75 years ago and still so today. What would be the Al equivalent of that little black bag? At this moment when new capabilities are emerging, how do we imagine them into medicine?”   [END OF BOOK PASSAGE]     [THEME MUSIC]     This is The AI Revolution in Medicine, Revisited. I’m your host, Peter Lee.    Shortly after OpenAI’s GPT-4 was publicly released, Carey Goldberg, Dr. Zak Kohane, and I published The AI Revolution in Medicine to help educate the world of healthcare and medical research about the transformative impact this new generative AI technology could have. But because we wrote the book when GPT-4 was still a secret, we had to speculate. Now, two years later, what did we get right, and what did we get wrong?     In this series, we’ll talk to clinicians, patients, hospital administrators, and others to understand the reality of AI in the field and where we go from here.   [THEME MUSIC FADES] The book passage I read at the top is from “Chapter 10: The Big Black Bag.”  In imagining AI in medicine, Carey, Zak, and I included in our book two fictional accounts. In the first, a medical resident consults GPT-4 on her personal phone as the patient in front of her crashes. Within seconds, it offers an alternate response based on recent literature. In the second account, a 90-year-old woman with several chronic conditions is living independently and receiving near-constant medical support from an AI aide.    In our conversations with the guests we’ve spoken to so far, we’ve caught a glimpse of these predicted futures, seeing how clinicians and patients are actually using AI today and how developers are leveraging the technology in the healthcare products and services they’re creating. In fact, that first fictional account isn’t so fictional after all, as most of the doctors in the real world actually appear to be using AI at least occasionally—and sometimes much more than occasionally—to help in their daily clinical work. And as for the second fictional account, which is more of a science fiction account, it seems we are indeed on the verge of a new way of delivering and receiving healthcare, though the future is still very much open.  As we continue to examine the current state of AI in healthcare and its potential to transform the field, I’m pleased to welcome Bill Gates and Sébastien Bubeck.   Bill may be best known as the co-founder of Microsoft, having created the company with his childhood friend Paul Allen in 1975. He’s now the founder of Breakthrough Energy, which aims to advance clean energy innovation, and TerraPower, a company developing groundbreaking nuclear energy and science technologies. He also chairs the world’s largest philanthropic organization, the Gates Foundation, and focuses on solving a variety of health challenges around the globe and here at home.  Sébastien is a research lead at OpenAI. He was previously a distinguished scientist, vice president of AI, and a colleague of mine here at Microsoft, where his work included spearheading the development of the family of small language models known as Phi. While at Microsoft, he also coauthored the discussion-provoking 2023 paper “Sparks of Artificial General Intelligence,” which presented the results of early experiments with GPT-4 conducted by a small team from Microsoft Research.    [TRANSITION MUSIC]   Here’s my conversation with Bill Gates and Sébastien Bubeck.  LEE: Bill, welcome.  BILL GATES: Thank you.  LEE: Seb …  SÉBASTIEN BUBECK: Yeah. Hi, hi, Peter. Nice to be here.  LEE: You know, one of the things that I’ve been doing just to get the conversation warmed up is to talk about origin stories, and what I mean about origin stories is, you know, what was the first contact that you had with large language models or the concept of generative AI that convinced you or made you think that something really important was happening?  And so, Bill, I think I’ve heard the story about, you know, the time when the OpenAI folks—Sam Altman, Greg Brockman, and others—showed you something, but could we hear from you what those early encounters were like and what was going through your mind?   GATES: Well, I’d been visiting OpenAI soon after it was created to see things like GPT-2 and to see the little arm they had that was trying to match human manipulation and, you know, looking at their games like Dota that they were trying to get as good as human play. And honestly, I didn’t think the language model stuff they were doing, even when they got to GPT-3, would show the ability to learn, you know, in the same sense that a human reads a biology book and is able to take that knowledge and access it not only to pass a test but also to create new medicines.  And so my challenge to them was that if their LLM could get a five on the advanced placement biology test, then I would say, OK, it took biologic knowledge and encoded it in an accessible way and that I didn’t expect them to do that very quickly but it would be profound.   And it was only about six months after I challenged them to do that, that an early version of GPT-4 they brought up to a dinner at my house, and in fact, it answered most of the questions that night very well. The one it got totally wrong, we were … because it was so good, we kept thinking, Oh, we must be wrong. It turned out it was a math weakness [LAUGHTER] that, you know, we later understood that that was an area of, weirdly, of incredible weakness of those early models. But, you know, that was when I realized, OK, the age of cheap intelligence was at its beginning.  LEE: Yeah. So I guess it seems like you had something similar to me in that my first encounters, I actually harbored some skepticism. Is it fair to say you were skeptical before that?  GATES: Well, the idea that we’ve figured out how to encode and access knowledge in this very deep sense without even understanding the nature of the encoding, …  LEE: Right.   GATES: … that is a bit weird.   LEE: Yeah.  GATES: We have an algorithm that creates the computation, but even say, OK, where is the president’s birthday stored in there? Where is this fact stored in there? The fact that even now when we’re playing around, getting a little bit more sense of it, it’s opaque to us what the semantic encoding is, it’s, kind of, amazing to me. I thought the invention of knowledge storage would be an explicit way of encoding knowledge, not an implicit statistical training.  LEE: Yeah, yeah. All right. So, Seb, you know, on this same topic, you know, I got—as we say at Microsoft—I got pulled into the tent. [LAUGHS]  BUBECK: Yes.   LEE: Because this was a very secret project. And then, um, I had the opportunity to select a small number of researchers in MSR [Microsoft Research] to join and start investigating this thing seriously. And the first person I pulled in was you.  BUBECK: Yeah.  LEE: And so what were your first encounters? Because I actually don’t remember what happened then.  BUBECK: Oh, I remember it very well. [LAUGHS] My first encounter with GPT-4 was in a meeting with the two of you, actually. But my kind of first contact, the first moment where I realized that something was happening with generative AI, was before that. And I agree with Bill that I also wasn’t too impressed by GPT-3.  I though that it was kind of, you know, very naturally mimicking the web, sort of parroting what was written there in a nice way. Still in a way which seemed very impressive. But it wasn’t really intelligent in any way. But shortly after GPT-3, there was a model before GPT-4 that really shocked me, and this was the first image generation model, DALL-E 1.  So that was in 2021. And I will forever remember the press release of OpenAI where they had this prompt of an avocado chair and then you had this image of the avocado chair. [LAUGHTER] And what really shocked me is that clearly the model kind of “understood” what is a chair, what is an avocado, and was able to merge those concepts.  So this was really, to me, the first moment where I saw some understanding in those models.   LEE: So this was, just to get the timing right, that was before I pulled you into the tent.  BUBECK: That was before. That was like a year before.  LEE: Right.   BUBECK: And now I will tell you how, you know, we went from that moment to the meeting with the two of you and GPT-4.  So once I saw this kind of understanding, I thought, OK, fine. It understands concept, but it’s still not able to reason. It cannot—as, you know, Bill was saying—it cannot learn from your document. It cannot reason.   So I set out to try to prove that. You know, this is what I was in the business of at the time, trying to prove things in mathematics. So I was trying to prove that basically autoregressive transformers could never reason. So I was trying to prove this. And after a year of work, I had something reasonable to show. And so I had the meeting with the two of you, and I had this example where I wanted to say, there is no way that an LLM is going to be able to do x.  And then as soon as I … I don’t know if you remember, Bill. But as soon as I said that, you said, oh, but wait a second. I had, you know, the OpenAI crew at my house recently, and they showed me a new model. Why don’t we ask this new model this question?   LEE: Yeah. BUBECK: And we did, and it solved it on the spot. And that really, honestly, just changed my life. Like, you know, I had been working for a year trying to say that this was impossible. And just right there, it was shown to be possible.   LEE: [LAUGHS] One of the very first things I got interested in—because I was really thinking a lot about healthcare—was healthcare and medicine.  And I don’t know if the two of you remember, but I ended up doing a lot of tests. I ran through, you know, step one and step two of the US Medical Licensing Exam. Did a whole bunch of other things. I wrote this big report. It was, you know, I can’t remember … a couple hundred pages.   And I needed to share this with someone. I didn’t … there weren’t too many people I could share it with. So I sent, I think, a copy to you, Bill. Sent a copy to you, Seb.   I hardly slept for about a week putting that report together. And, yeah, and I kept working on it. But I was far from alone. I think everyone who was in the tent, so to speak, in those early days was going through something pretty similar. All right. So I think … of course, a lot of what I put in the report also ended up being examples that made it into the book.  But the main purpose of this conversation isn’t to reminisce about [LAUGHS] or indulge in those reminiscences but to talk about what’s happening in healthcare and medicine. And, you know, as I said, we wrote this book. We did it very, very quickly. Seb, you helped. Bill, you know, you provided a review and some endorsements.  But, you know, honestly, we didn’t know what we were talking about because no one had access to this thing. And so we just made a bunch of guesses. So really, the whole thing I wanted to probe with the two of you is, now with two years of experience out in the world, what, you know, what do we think is happening today?  You know, is AI actually having an impact, positive or negative, on healthcare and medicine? And what do we now think is going to happen in the next two years, five years, or 10 years? And so I realize it’s a little bit too abstract to just ask it that way. So let me just try to narrow the discussion and guide us a little bit.   Um, the kind of administrative and clerical work, paperwork, around healthcare—and we made a lot of guesses about that—that appears to be going well, but, you know, Bill, I know we’ve discussed that sometimes that you think there ought to be a lot more going on. Do you have a viewpoint on how AI is actually finding its way into reducing paperwork?  GATES: Well, I’m stunned … I don’t think there should be a patient-doctor meeting where the AI is not sitting in and both transcribing, offering to help with the paperwork, and even making suggestions, although the doctor will be the one, you know, who makes the final decision about the diagnosis and whatever prescription gets done.   It’s so helpful. You know, when that patient goes home and their, you know, son who wants to understand what happened has some questions, that AI should be available to continue that conversation. And the way you can improve that experience and streamline things and, you know, involve the people who advise you. I don’t understand why that’s not more adopted, because there you still have the human in the loop making that final decision.  But even for, like, follow-up calls to make sure the patient did things, to understand if they have concerns and knowing when to escalate back to the doctor, the benefit is incredible. And, you know, that thing is ready for prime time. That paradigm is ready for prime time, in my view.  LEE: Yeah, there are some good products, but it seems like the number one use right now—and we kind of got this from some of the previous guests in previous episodes—is the use of AI just to respond to emails from patients. [LAUGHTER] Does that make sense to you?  BUBECK: Yeah. So maybe I want to second what Bill was saying but maybe take a step back first. You know, two years ago, like, the concept of clinical scribes, which is one of the things that we’re talking about right now, it would have sounded, in fact, it sounded two years ago, borderline dangerous. Because everybody was worried about hallucinations. What happened if you have this AI listening in and then it transcribes, you know, something wrong?  Now, two years later, I think it’s mostly working. And in fact, it is not yet, you know, fully adopted. You’re right. But it is in production. It is used, you know, in many, many places. So this rate of progress is astounding because it wasn’t obvious that we would be able to overcome those obstacles of hallucination. It’s not to say that hallucinations are fully solved. In the case of the closed system, they are.   Now, I think more generally what’s going on in the background is that there is something that we, that certainly I, underestimated, which is this management overhead. So I think the reason why this is not adopted everywhere is really a training and teaching aspect. People need to be taught, like, those systems, how to interact with them.  And one example that I really like, a study that recently appeared where they tried to use ChatGPT for diagnosis and they were comparing doctors without and with ChatGPT (opens in new tab). And the amazing thing … so this was a set of cases where the accuracy of the doctors alone was around 75%. ChatGPT alone was 90%. So that’s already kind of mind blowing. But then the kicker is that doctors with ChatGPT was 80%.   Intelligence alone is not enough. It’s also how it’s presented, how you interact with it. And ChatGPT, it’s an amazing tool. Obviously, I absolutely love it. But it’s not … you don’t want a doctor to have to type in, you know, prompts and use it that way.  It should be, as Bill was saying, kind of running continuously in the background, sending you notifications. And you have to be really careful of the rate at which those notifications are being sent. Because if they are too frequent, then the doctor will learn to ignore them. So you have to … all of those things matter, in fact, at least as much as the level of intelligence of the machine.  LEE: One of the things I think about, Bill, in that scenario that you described, doctors do some thinking about the patient when they write the note. So, you know, I’m always a little uncertain whether it’s actually … you know, you wouldn’t necessarily want to fully automate this, I don’t think. Or at least there needs to be some prompt to the doctor to make sure that the doctor puts some thought into what happened in the encounter with the patient. Does that make sense to you at all?  GATES: At this stage, you know, I’d still put the onus on the doctor to write the conclusions and the summary and not delegate that.  The tradeoffs you make a little bit are somewhat dependent on the situation you’re in. If you’re in Africa, So, yes, the doctor’s still going to have to do a lot of work, but just the quality of letting the patient and the people around them interact and ask questions and have things explained, that alone is such a quality improvement. It’s mind blowing.   LEE: So since you mentioned, you know, Africa—and, of course, this touches on the mission and some of the priorities of the Gates Foundation and this idea of democratization of access to expert medical care—what’s the most interesting stuff going on right now? Are there people and organizations or technologies that are impressing you or that you’re tracking?  GATES: Yeah. So the Gates Foundation has given out a lot of grants to people in Africa doing education, agriculture but more healthcare examples than anything. And the way these things start off, they often start out either being patient-centric in a narrow situation, like, OK, I’m a pregnant woman; talk to me. Or, I have infectious disease symptoms; talk to me. Or they’re connected to a health worker where they’re helping that worker get their job done. And we have lots of pilots out, you know, in both of those cases.   The dream would be eventually to have the thing the patient consults be so broad that it’s like having a doctor available who understands the local things.   LEE: Right.   GATES: We’re not there yet. But over the next two or three years, you know, particularly given the worsening financial constraints against African health systems, where the withdrawal of money has been dramatic, you know, figuring out how to take this—what I sometimes call “free intelligence”—and build a quality health system around that, we will have to be more radical in low-income countries than any rich country is ever going to be.   LEE: Also, there’s maybe a different regulatory environment, so some of those things maybe are easier? Because right now, I think the world hasn’t figured out how to and whether to regulate, let’s say, an AI that might give a medical diagnosis or write a prescription for a medication.  BUBECK: Yeah. I think one issue with this, and it’s also slowing down the deployment of AI in healthcare more generally, is a lack of proper benchmark. Because, you know, you were mentioning the USMLE [United States Medical Licensing Examination], for example. That’s a great test to test human beings and their knowledge of healthcare and medicine. But it’s not a great test to give to an AI.  It’s not asking the right questions. So finding what are the right questions to test whether an AI system is ready to give diagnosis in a constrained setting, that’s a very, very important direction, which to my surprise, is not yet accelerating at the rate that I was hoping for.  LEE: OK, so that gives me an excuse to get more now into the core AI tech because something I’ve discussed with both of you is this issue of what are the right tests. And you both know the very first test I give to any new spin of an LLM is I present a patient, the results—a mythical patient—the results of my physical exam, my mythical physical exam. Maybe some results of some initial labs. And then I present or propose a differential diagnosis. And if you’re not in medicine, a differential diagnosis you can just think of as a prioritized list of the possible diagnoses that fit with all that data. And in that proposed differential, I always intentionally make two mistakes.  I make a textbook technical error in one of the possible elements of the differential diagnosis, and I have an error of omission. And, you know, I just want to know, does the LLM understand what I’m talking about? And all the good ones out there do now. But then I want to know, can it spot the errors? And then most importantly, is it willing to tell me I’m wrong, that I’ve made a mistake?   That last piece seems really hard for AI today. And so let me ask you first, Seb, because at the time of this taping, of course, there was a new spin of GPT-4o last week that became overly sycophantic. In other words, it was actually prone in that test of mine not only to not tell me I’m wrong, but it actually praised me for the creativity of my differential. [LAUGHTER] What’s up with that?  BUBECK: Yeah, I guess it’s a testament to the fact that training those models is still more of an art than a science. So it’s a difficult job. Just to be clear with the audience, we have rolled back that [LAUGHS] version of GPT-4o, so now we don’t have the sycophant version out there.  Yeah, no, it’s a really difficult question. It has to do … as you said, it’s very technical. It has to do with the post-training and how, like, where do you nudge the model? So, you know, there is this very classical by now technique called RLHF [reinforcement learning from human feedback], where you push the model in the direction of a certain reward model. So the reward model is just telling the model, you know, what behavior is good, what behavior is bad.  But this reward model is itself an LLM, and, you know, Bill was saying at the very beginning of the conversation that we don’t really understand how those LLMs deal with concepts like, you know, where is the capital of France located? Things like that. It is the same thing for this reward model. We don’t know why it says that it prefers one output to another, and whether this is correlated with some sycophancy is, you know, something that we discovered basically just now. That if you push too hard in optimization on this reward model, you will get a sycophant model.  So it’s kind of … what I’m trying to say is we became too good at what we were doing, and we ended up, in fact, in a trap of the reward model.  LEE: I mean, you do want … it’s a difficult balance because you do want models to follow your desires and …  BUBECK: It’s a very difficult, very difficult balance.  LEE: So this brings up then the following question for me, which is the extent to which we think we’ll need to have specially trained models for things. So let me start with you, Bill. Do you have a point of view on whether we will need to, you know, quote-unquote take AI models to med school? Have them specially trained? Like, if you were going to deploy something to give medical care in underserved parts of the world, do we need to do something special to create those models?  GATES: We certainly need to teach them the African languages and the unique dialects so that the multimedia interactions are very high quality. We certainly need to teach them the disease prevalence and unique disease patterns like, you know, neglected tropical diseases and malaria. So we need to gather a set of facts that somebody trying to go for a US customer base, you know, wouldn’t necessarily have that in there.  Those two things are actually very straightforward because the additional training time is small. I’d say for the next few years, we’ll also need to do reinforcement learning about the context of being a doctor and how important certain behaviors are. Humans learn over the course of their life to some degree that, I’m in a different context and the way I behave in terms of being willing to criticize or be nice, you know, how important is it? Who’s here? What’s my relationship to them?   Right now, these machines don’t have that broad social experience. And so if you know it’s going to be used for health things, a lot of reinforcement learning of the very best humans in that context would still be valuable. Eventually, the models will, having read all the literature of the world about good doctors, bad doctors, it’ll understand as soon as you say, “I want you to be a doctor diagnosing somebody.” All of the implicit reinforcement that fits that situation, you know, will be there. LEE: Yeah. GATES: And so I hope three years from now, we don’t have to do that reinforcement learning. But today, for any medical context, you would want a lot of data to reinforce tone, willingness to say things when, you know, there might be something significant at stake.  LEE: Yeah. So, you know, something Bill said, kind of, reminds me of another thing that I think we missed, which is, the context also … and the specialization also pertains to different, I guess, what we still call “modes,” although I don’t know if the idea of multimodal is the same as it was two years ago. But, you know, what do you make of all of the hubbub around—in fact, within Microsoft Research, this is a big deal, but I think we’re far from alone—you know, medical images and vision, video, proteins and molecules, cell, you know, cellular data and so on.  BUBECK: Yeah. OK. So there is a lot to say to everything … to the last, you know, couple of minutes. Maybe on the specialization aspect, you know, I think there is, hiding behind this, a really fundamental scientific question of whether eventually we have a singular AGI [artificial general intelligence] that kind of knows everything and you can just put, you know, explain your own context and it will just get it and understand everything.  That’s one vision. I have to say, I don’t particularly believe in this vision. In fact, we humans are not like that at all. I think, hopefully, we are general intelligences, yet we have to specialize a lot. And, you know, I did myself a lot of RL, reinforcement learning, on mathematics. Like, that’s what I did, you know, spent a lot of time doing that. And I didn’t improve on other aspects. You know, in fact, I probably degraded in other aspects. [LAUGHTER] So it’s … I think it’s an important example to have in mind.  LEE: I think I might disagree with you on that, though, because, like, doesn’t a model have to see both good science and bad science in order to be able to gain the ability to discern between the two?  BUBECK: Yeah, no, that absolutely. I think there is value in seeing the generality, in having a very broad base. But then you, kind of, specialize on verticals. And this is where also, you know, open-weights model, which we haven’t talked about yet, are really important because they allow you to provide this broad base to everyone. And then you can specialize on top of it.  LEE: So we have about three hours of stuff to talk about, but our time is actually running low. BUBECK: Yes, yes, yes.   LEE: So I think I want … there’s a more provocative question. It’s almost a silly question, but I need to ask it of the two of you, which is, is there a future, you know, where AI replaces doctors or replaces, you know, medical specialties that we have today? So what does the world look like, say, five years from now?  GATES: Well, it’s important to distinguish healthcare discovery activity from healthcare delivery activity. We focused mostly on delivery. I think it’s very much within the realm of possibility that the AI is not only accelerating healthcare discovery but substituting for a lot of the roles of, you know, I’m an organic chemist, or I run various types of assays. I can see those, which are, you know, testable-output-type jobs but with still very high value, I can see, you know, some replacement in those areas before the doctor.   The doctor, still understanding the human condition and long-term dialogues, you know, they’ve had a lifetime of reinforcement of that, particularly when you get into areas like mental health. So I wouldn’t say in five years, either people will choose to adopt it, but it will be profound that there’ll be this nearly free intelligence that can do follow-up, that can help you, you know, make sure you went through different possibilities.  And so I’d say, yes, we’ll have doctors, but I’d say healthcare will be massively transformed in its quality and in efficiency by AI in that time period.  LEE: Is there a comparison, useful comparison, say, between doctors and, say, programmers, computer programmers, or doctors and, I don’t know, lawyers?  GATES: Programming is another one that has, kind of, a mathematical correctness to it, you know, and so the objective function that you’re trying to reinforce to, as soon as you can understand the state machines, you can have something that’s “checkable”; that’s correct. So I think programming, you know, which is weird to say, that the machine will beat us at most programming tasks before we let it take over roles that have deep empathy, you know, physical presence and social understanding in them.  LEE: Yeah. By the way, you know, I fully expect in five years that AI will produce mathematical proofs that are checkable for validity, easily checkable, because they’ll be written in a proof-checking language like Lean or something but will be so complex that no human mathematician can understand them. I expect that to happen.   I can imagine in some fields, like cellular biology, we could have the same situation in the future because the molecular pathways, the chemistry, biochemistry of human cells or living cells is as complex as any mathematics, and so it seems possible that we may be in a state where in wet lab, we see, Oh yeah, this actually works, but no one can understand why.  BUBECK: Yeah, absolutely. I mean, I think I really agree with Bill’s distinction of the discovery and the delivery, and indeed, the discovery’s when you can check things, and at the end, there is an artifact that you can verify. You know, you can run the protocol in the wet lab and see [if you have] produced what you wanted. So I absolutely agree with that.   And in fact, you know, we don’t have to talk five years from now. I don’t know if you know, but just recently, there was a paper that was published on a scientific discovery using o3- mini (opens in new tab). So this is really amazing. And, you know, just very quickly, just so people know, it was about this statistical physics model, the frustrated Potts model, which has to do with coloring, and basically, the case of three colors, like, more than two colors was open for a long time, and o3 was able to reduce the case of three colors to two colors.   LEE: Yeah.  BUBECK: Which is just, like, astounding. And this is not … this is now. This is happening right now. So this is something that I personally didn’t expect it would happen so quickly, and it’s due to those reasoning models.   Now, on the delivery side, I would add something more to it for the reason why doctors and, in fact, lawyers and coders will remain for a long time, and it’s because we still don’t understand how those models generalize. Like, at the end of the day, we are not able to tell you when they are confronted with a really new, novel situation, whether they will work or not.  Nobody is able to give you that guarantee. And I think until we understand this generalization better, we’re not going to be willing to just let the system in the wild without human supervision.  LEE: But don’t human doctors, human specialists … so, for example, a cardiologist sees a patient in a certain way that a nephrologist …  BUBECK: Yeah. LEE: … or an endocrinologist might not. BUBECK: That’s right. But another cardiologist will understand and, kind of, expect a certain level of generalization from their peer. And this, we just don’t have it with AI models. Now, of course, you’re exactly right. That generalization is also hard for humans. Like, if you have a human trained for one task and you put them into another task, then you don’t … you often don’t know. LEE: OK. You know, the podcast is focused on what’s happened over the last two years. But now, I’d like one provocative prediction about what you think the world of AI and medicine is going to be at some point in the future. You pick your timeframe. I don’t care if it’s two years or 20 years from now, but, you know, what do you think will be different about AI in medicine in that future than today?  BUBECK: Yeah, I think the deployment is going to accelerate soon. Like, we’re really not missing very much. There is this enormous capability overhang. Like, even if progress completely stopped, with current systems, we can do a lot more than what we’re doing right now. So I think this will … this has to be realized, you know, sooner rather than later.  And I think it’s probably dependent on these benchmarks and proper evaluation and tying this with regulation. So these are things that take time in human society and for good reason. But now we already are at two years; you know, give it another two years and it should be really …   LEE: Will AI prescribe your medicines? Write your prescriptions?  BUBECK: I think yes. I think yes.  LEE: OK. Bill?  GATES: Well, I think the next two years, we’ll have massive pilots, and so the amount of use of the AI, still in a copilot-type mode, you know, we should get millions of patient visits, you know, both in general medicine and in the mental health side, as well. And I think that’s going to build up both the data and the confidence to give the AI some additional autonomy. You know, are you going to let it talk to you at night when you’re panicked about your mental health with some ability to escalate? And, you know, I’ve gone so far as to tell politicians with national health systems that if they deploy AI appropriately, that the quality of care, the overload of the doctors, the improvement in the economics will be enough that their voters will be stunned because they just don’t expect this, and, you know, they could be reelected [LAUGHTER] just on this one thing of fixing what is a very overloaded and economically challenged health system in these rich countries.  You know, my personal role is going to be to make sure that in the poorer countries, there isn’t some lag; in fact, in many cases, that we’ll be more aggressive because, you know, we’re comparing to having no access to doctors at all. And, you know, so I think whether it’s India or Africa, there’ll be lessons that are globally valuable because we need medical intelligence. And, you know, thank god AI is going to provide a lot of that.  LEE: Well, on that optimistic note, I think that’s a good way to end. Bill, Seb, really appreciate all of this.   I think the most fundamental prediction we made in the book is that AI would actually find its way into the practice of medicine, and I think that that at least has come true, maybe in different ways than we expected, but it’s come true, and I think it’ll only accelerate from here. So thanks again, both of you.  [TRANSITION MUSIC]  GATES: Yeah. Thanks, you guys.  BUBECK: Thank you, Peter. Thanks, Bill.  LEE: I just always feel such a sense of privilege to have a chance to interact and actually work with people like Bill and Sébastien.    With Bill, I’m always amazed at how practically minded he is. He’s really thinking about the nuts and bolts of what AI might be able to do for people, and his thoughts about underserved parts of the world, the idea that we might actually be able to empower people with access to expert medical knowledge, I think is both inspiring and amazing.   And then, Seb, Sébastien Bubeck, he’s just absolutely a brilliant mind. He has a really firm grip on the deep mathematics of artificial intelligence and brings that to bear in his research and development work. And where that mathematics takes him isn’t just into the nuts and bolts of algorithms but into philosophical questions about the nature of intelligence.   One of the things that Sébastien brought up was the state of evaluation of AI systems. And indeed, he was fairly critical in our conversation. But of course, the world of AI research and development is just moving so fast, and indeed, since we recorded our conversation, OpenAI, in fact, released a new evaluation metric that is directly relevant to medical applications, and that is something called HealthBench. And Microsoft Research also released a new evaluation approach or process called ADeLe.   HealthBench and ADeLe are examples of new approaches to evaluating AI models that are less about testing their knowledge and ability to pass multiple-choice exams and instead are evaluation approaches designed to assess how well AI models are able to complete tasks that actually arise every day in typical healthcare or biomedical research settings. These are examples of really important good work that speak to how well AI models work in the real world of healthcare and biomedical research and how well they can collaborate with human beings in those settings.  You know, I asked Bill and Seb to make some predictions about the future. You know, my own answer, I expect that we’re going to be able to use AI to change how we diagnose patients, change how we decide treatment options.   If you’re a doctor or a nurse and you encounter a patient, you’ll ask questions, do a physical exam, you know, call out for labs just like you do today, but then you’ll be able to engage with AI based on all of that data and just ask, you know, based on all the other people who have gone through the same experience, who have similar data, how were they diagnosed? How were they treated? What were their outcomes? And what does that mean for the patient I have right now? Some people call it the “patients like me” paradigm. And I think that’s going to become real because of AI within our lifetimes. That idea of really grounding the delivery in healthcare and medical practice through data and intelligence, I actually now don’t see any barriers to that future becoming real.  [THEME MUSIC]  I’d like to extend another big thank you to Bill and Sébastien for their time. And to our listeners, as always, it’s a pleasure to have you along for the ride. I hope you’ll join us for our remaining conversations, as well as a second coauthor roundtable with Carey and Zak.   Until next time.   [MUSIC FADES]

Rene Byrd did IVF to have her baby. Courtesy of Rene Byrd 2025-06-14T21:23:01Z Save Saved Read in app This story is available exclusively to Business Insider subscribers. Become an Insider and start reading now. Have an account? Rene Byrd is a 49-year-old singer-songwriter in London who had her first baby at 48. She had held on to hope for a baby throughout her 40s, undergoing IVF for over two years. Being an older mom has had several benefits, like financial security and contentment. This as-told-to essay is based on a conversation with Rene Byrd. It has been edited for length and clarity.When I turned 40, I went on a seven-day retreat full of meditation and massage to fall in love with myself. I'm a strong believer that to find love, you first have to love yourself.I had wanted to settle down with someone and build a family, but it just hadn't happened. Three years prior, I had frozen my eggs because I knew that I wanted a family someday.On the retreat, I felt deep in my spirit that I would one day find my person and hold my child in my hands. I wouldn't give up hope.I met someone at a barReturning home, I continued dating, but it wasn't until a chance meeting at a bar that I finally found the man who would become my husband. I hadn't quite turned 41, and he was 34.I remember not wanting to scare him off by talking too much about my desire for kids, but we did have discussions about the future. When love started to bloom between the two of us, we started looking at what our options were for having a child together.After trying holistic methods to no avail, we decided to go down the IVF route. I'd heard horror stories about IVF — that it was never straightforward — but as I already had my eggs frozen, it was the best option for us at the time.I felt guilty for waiting so longTwo-and-a-half long years later, I was given the news from the IVF clinic — I was pregnant. I fell apart, phoning my husband to tell him we would be having a baby. Rene Byrd got pregnant at age 48 thanks to IVF. Courtesy of Rene Byrd Throughout my pregnancy, I remember being scared of what this new life as a mother would look like. I had little panic attacks considering how different life would be, as compared to the decades of life without a child. And then I felt guilty, telling myself I had waited so long for this. There was a lot of grappling with these thoughts until I realized my child would just be an extension of me.Once our little boy, Crue, was born in November 2024, I felt ready for his arrival in theory. Having spent years hearing from friends with children, I had an idea of what to expect. Even still, those early days were a lot to deal with. All these things were being thrown at me about what I should and shouldn't do with a baby.Being a mom in my late 40s has so many beautiful benefitsI joined online mother and baby communities and in-person baby groups, finding my tribe of mothers like me, ones that were "older."There is a stillness within me that grounds me as I take care of Crue. I have this playbook of mothering, developed from years of research and observation, that has given me assurance that even when things don't seem to be going to plan — like breastfeeding or sleeping — I was OK, and so was he.Having built up financial security, I didn't worry about how I was going to provide for a baby. Established in a career, I could plan for all baby-related expenses, including IVF.And since I had gotten so much out of my system in my younger years — corporate working, parties, nice restaurants — I felt content to settle in at home with my baby and husband. I never feel like I'm missing out.The only concern I've heard quietly whispered in different circles is that of my health. I know that as I get older, little issues with my body could pop up — issues that I might not have had as a younger mother. This has forced me to look after my body more than I ever have so that I can fully enjoy time with Crue as he gets older.Becoming a mother had always been a dream of mine. I trusted the process, holding on to hope, and although delayed, my dream finally came true.

The epidermis (skin) is the body’s largest organ, so it would make sense that toxins found in fabrics that sit on the skin’s surface could be absorbed by the skin and make their way into the bloodstream. And polyester has been considered a particularly suspect fabric because it’s made from a chemical called polyethylene terephthalate, a plastic polymer used in various products.One study published in 1993 followed 24 dogs who were divided into two equal groups, one group wore cotton underpants and the other polyester. At the end of the study period, there was a significant decrease in sperm count and an increase in sperm abnormalities in the dogs who wore the polyester pants. But that said, this study is three decades old, done on dogs, and has had little additional research to show for it since.So, the jury is certainly still out as to whether fabrics decrease fertility, but there are some things that we do know. Chemicals Found in PolyesterAccording to Audrey Gaskins, an associate professor of environmental health at Emory University, most studies are focused on specific chemicals that might be found in fabrics rather than the fabrics themselves, and those chemicals are usually measured in blood or urine. But fabrics like polyester can contain a number of chemicals that might impact fertility. PFAS, short for per- and polyfluoroalkyl substances, are a group of chemicals found in thousands of products, and they’re difficult for the body to eliminate.“PFAS are commonly found in water-resistant clothing,” says Gaskins. However, drinking water is likely the most common avenue of exposure, as well as non-stick cookware, and many others.Research has shown that PFAS can reduce fertility in women by some 40 percent. According to NIH’s National Institute for Environmental Health Sciences, high levels of PFAS found in the blood were linked to a reduced chance of pregnancy and live birth. Other research has shown that PFAS are linked to increased instances of endometriosis and polycystic ovary syndrome (PCOS), both of which reduce fertility.Poor Pregnancy OutcomesPolyester (when combined with spandex) may also contain bisphenol A (BPA), another chemical compound that has been shown to potentially impact fertility. A December 2022 study published in the Journal of Clinical Medicine found a higher prevalence of PCOS in women with high amounts of BPA in their blood.Finally, polyester can contain phthalates, a chemical commonly used in things like sports bras and other pieces of clothing. These, too, have been shown to have a negative impact on fertility. A study published in the September 2021 issue of the journal Best Practice & Research Clinical Endocrinology & Metabolism found that higher concentrations of the chemical have been associated with decreased rates of pregnancy, increased incidences of miscarriage, and other pregnancy complications.“We’ve found suggestive associations between higher concentrations of bisphenol and phthalate metabolites and worse markers of reproductive health like poor success with IVF,” says Gaskins. “What we don’t know is where the source of exposure is coming from.”Exposure to Fertility-Decreasing ChemicalsStill, the obvious implication if you’re trying to get pregnant is to try to decrease your exposure to any of these chemicals through any route possible, especially when you have control over exposure. If we know there are chemicals in these fabrics, decreasing use of them would be more achievable for many people compared to, say, changing your drinking water, says Gaskins.There’s definitely no downside to decreasing your exposure to these chemicals, and while clothing is likely not the largest means of exposure to things like PFAs, phthalates, and BPA, if you’re trying to get pregnant, they’re certainly a good place to start.This article is not offering medical advice and should be used for informational purposes only.Article SourcesOur writers at Discovermagazine.com use peer-reviewed studies and high-quality sources for our articles, and our editors review for scientific accuracy and editorial standards. Review the sources used below for this article:National Institute of Environmental Health Sciences. PFAS Exposure Linked to Reduced Fertility in Women Center for Environmental Health. What You Need to Know About BPA in ClothingJournal of Clinical Medicine. Bisphenol-A and Female Fertility: An Update of Existing Epidemiological StudiesBest Practice & Research Clinical Endocrinology & Metabolism. Phthalates, ovarian function and fertility in adulthoodSara Novak is a science journalist based in South Carolina. In addition to writing for Discover, her work appears in Scientific American, Popular Science, New Scientist, Sierra Magazine, Astronomy Magazine, and many more. She graduated with a bachelor’s degree in Journalism from the Grady School of Journalism at the University of Georgia. She's also a candidate for a master’s degree in science writing from Johns Hopkins University (expected graduation 2023).

US Secretary of Health and Human Services Robert F Kennedy JrTasos Katopodis/Getty One of the top vaccine experts at the US Centers for Disease Control and Prevention (CDC), Lakshmi Panagiotakopoulos, resigned on 4 June – a week after Robert F Kennedy Jr announced that covid-19 vaccines would no longer be recommended for most children and pregnancies. The announcement set off several days of confusion around who will have access to covid-19 vaccines in the US going forward. In practice, there hasn’t been a drastic change to access, though there will probably be new obstacles for parents hoping to vaccinate their children. Still, Kennedy’s announcement signals a troubling circumvention of public health norms. “My career in public health and vaccinology started with a deep-seated desire to help the most vulnerable members of our population, and that is not something I am able to continue doing in this role,” said Panagiotakopoulos in an email to colleagues obtained by Reuters. Panagiotakopoulos supported the Advisory Committee on Immunization Practices (ACIP), which has advised the CDC on vaccine recommendations since 1964. But last week, Kennedy – the country’s highest-ranking public health official – upended this decades-long precedent. “I couldn’t be more pleased to announce that, as of today, the covid vaccine for healthy children and healthy pregnant woman has been removed from the CDC recommended immunisation schedule,” he said in a video posted to the social media platform X on 27 May. Despite his directive, the CDC has, so far, only made minor changes to its guidance on covid-19 vaccines. Instead of recommending them for children outright, it now recommends vaccination “based on shared clinical decision-making”. In other words, parents should talk with a doctor before deciding. It isn’t clear how this will affect access to these vaccines in every scenario, but it could make it more difficult for children to get a shot at pharmacies. Get the most essential health and fitness news in your inbox every Saturday. Sign up to newsletter The CDC’s guidance on vaccination in pregnancy is also ambiguous. While its website still recommends a covid-19 shot during pregnancy, a note at the top says, “this page will be updated to align with the updated immunization schedule.” Kennedy’s announcement contradicts the stances of major public health organisations, too. Both the American College of Obstetricians and Gynecologists (ACOG) and the American Academy of Pediatrics (APP) have come out opposing it. “The CDC and HHS encourage individuals to talk with their healthcare provider about any personal medical decision,” an HHS spokesperson told New Scientist. “Under the leadership of Secretary Kennedy, HHS is restoring the doctor-patient relationship.” However, Linda Eckert at the University of Washington in Seattle says the conflicting messages are confusing for people. “It opens up disinformation opportunities. It undermines confidence in vaccination in general,” she says. “I can’t imagine it won’t decrease immunisation rates overall.” Research has repeatedly shown covid-19 vaccination in adolescence and pregnancy is safe and effective. In fact, Martin Makary, the head of the US Food and Drug Administration (FDA), listed pregnancy as a risk factor for severe covid-19 a week before Kennedy’s announcement, further convoluting the government’s public health messaging. Kennedy’s announcement is in line with some other countries’ covid policies. For example, Australia and the UK don’t recommend covid-19 vaccines for children unless they are at risk of severe illness. They also don’t recommend covid-19 vaccination during pregnancy if someone is already vaccinated. Asma Khalil, a member of the UK Joint Committee on Vaccination and Immunisation, says the UK’s decision was based on the reduced risk of the omicron variant, the cost-effectiveness of vaccination and high population immunity. However, these factors can vary across countries. The UK population also tends to have better access to healthcare than the US, says Eckert. “These decisions need to carefully consider the risks and benefits relative to the national population,” says Khalil. The HHS didn’t answer New Scientist’s questions about whether a similar analysis guided Kennedy’s decision-making. What is maybe most troubling, however, is the precedent Kennedy’s announcement sets. The ACIP – an independent group of public health experts – was expected to vote on proposed changes to covid-19 vaccine recommendations later this month. But Kennedy’s decision has bypassed this process. “This style of decision-making – by individuals versus going through experts who are carefully vetted for conflicts of interest, who carefully look at the data – this has never happened in our country,” says Eckert. “We’re in uncharted territory.” She worries the move could pave the way for Kennedy to chip away at other vaccine recommendations. “I know there are a lot of vaccines he has been actively against in his career,” she says. Kennedy has previously blamed vaccines for autism and falsely claimed that the polio vaccine caused more deaths than it averted. “What it speaks to is the fact that [Kennedy] does not see value in these vaccines and is going to do everything he can to try and devalue them in the minds of the public and make them harder to receive,” says Amesh Adalja at Johns Hopkins University. Topics:

Nature, Published online: 04 June 2025; doi:10.1038/d41586-025-01456-7In mice, a lack of maternal iron impairs an iron-dependent enzyme that activates the male sex-determining gene, causing some XY embryos to develop ovaries.

If you wanted to get a COVID vaccine during pregnancy, to protect yourself and your future baby from the virus, that may soon be difficult to impossible. According to a short video posted on X, U.S. Department of Health and Human Services secretary Robert F. Kennedy, Jr, who is also a noted anti-vaccine activist, said that the COVID-19 vaccine “has been removed” from the list of vaccines recommended in pregnancy, as well as the list of vaccines recommended for healthy children. This announcement sidesteps the usual regulatory process, and it’s not clear exactly what will happen next—but here’s what we know. The announcement may not be entirely validRFK, Jr made the announcement in a video where he stood alongside the NIH director Jay Bhattacharya and FDA commissioner Marty Makary. Notably, nobody from the CDC was present. The FDA approves vaccines, but it’s the CDC that is in charge of recommendations. (It is not clear who the CDC’s acting director actually is, or whether there is one.) Normally, the CDC has an advisory panel called ACIP (the Advisory Committee for Immunization Practices) that reviews scientific evidence to make recommendations for vaccines. They’ll vote on whether a given vaccine should be recommended for everybody in a group of people. Their decisions are then passed to CDC leadership, who make the final call as to whether the vaccine gets officially recommended for that group. Vaccines are not usually added or removed to the recommended list by the CDC without consulting with ACIP, and they definitely aren’t usually added or removed by tweeting a video. Dorit Reiss, a law professor who specializes in vaccine policy, posted on LinkedIn that the announcement may not be legally valid if it’s not immediately followed by supporting documentation. She says: “Under administrative law, to avoid being found arbitrary and capricious, an agency's decision has to meet certain criteria, including explaining the agency's fact finding, a connection between the facts and the decisions, etc. A one minute video on Twitter doesn't quite get you there.” So far, the CDC’s web page on vaccines recommended in pregnancy still says that “A pregnant woman should get vaccinated against whooping cough, flu, COVID-19, and respiratory syncytial virus (RSV).” The adult and child vaccine schedules still include COVID vaccines.Strangely, this move on behalf of the CDC contradicts the one we reported about recently from the FDA. The FDA plans to require extra steps (possibly unethical and/or impractical ones) to approve new COVID vaccines for healthy children and adults. But these steps don’t apply to people who are at high risk for complications of COVID. The FDA’s policy announcement included a list of those high risk health conditions—which includes pregnancy.Why it matters which vaccines are “recommended”Recommending a vaccine doesn’t just mean expressing an opinion; the Affordable Care Act requires that vaccines recommended by ACIP must be covered by most private insurance and Medicaid expansion plans without any cost sharing. That means no deductible and no copay—so these vaccines must be free to you out of pocket if you fall into a group of people for whom they are recommended. The recommended vaccines include all the standard childhood vaccines, plus your seasonal flu shot, and other vaccines that are recommended for adults, for people who are pregnant, and so on. The full schedules are here. If you’ve gotten a COVID shot, a flu shot, a tetanus shot, a shingles shot—the shot’s inclusion on this list is why you were able to (probably) get it for free.So taking a vaccine off the recommended list means that it could be prohibitively expensive. GoodRX, which keeps tabs on pharmacy prices, reports that COVID shots may cost $200 or more out of pocket, plus any applicable administration fee that the provider might charge.Taking a vaccine off the recommended list may also mean it won’t be covered by the Vaccines for Children program, which provides free vaccines to children who don’t have coverage for them through health insurance.Whether or not the vaccine actually gets taken off the list, the recent HHS announcement has another impact: The American College of Obstetricians and Gynecologists said in a statement that “Following this announcement, we are worried about our patients in the future, who may be less likely to choose vaccination during pregnancy despite the clear and definitive evidence demonstrating its benefit.” The ACOG statement also pointed out a few ways in which removing the vaccines from the recommended list is not “common sense and good science,” as the HHS announcement claimed. ACOG writes: “As ob-gyns who treat patients every day, we have seen firsthand how dangerous COVID infection can be during pregnancy and for newborns who depend on maternal antibodies from the vaccine for protection. We also understand that despite the change in recommendations from HHS, the science has not changed. It is very clear that COVID infection during pregnancy can be catastrophic and lead to major disability, and it can cause devastating consequences for families.”

More chaos CDC updates COVID vaccine recommendations, but not how RFK Jr. wanted Mixed messages only add to uncertainty about vaccine access for kids, pregnant individuals. Beth Mole – May 30, 2025 4:28 pm | 74 A nurse gives a 16-year-old a COVID-19 vaccine. Credit: Getty | Sopa images A nurse gives a 16-year-old a COVID-19 vaccine. Credit: Getty | Sopa images Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more The Centers for Disease Control and Prevention on Thursday updated its immunization schedules for children and adults to partially reflect the abrupt changes announced by health secretary and anti-vaccine advocate Robert F. Kennedy Jr. earlier this week. In a 58-second video posted on social media on Tuesday, May 27, Kennedy said he was unilaterally revoking the CDC's recommendations that healthy children and pregnant people get COVID-19 vaccines. "I couldn’t be more pleased to announce that, as of today, the COVID vaccine for healthy children and healthy pregnant women has been removed from the CDC recommended immunization schedule," Kennedy said in the video. The health agency's immunization schedules were not, in fact, updated at the time of the announcement, though. The Washington Post subsequently reported that the CDC was blindsided by the announcement. Five hours went by after the video was posted before CDC officials said they received a one-page "secretarial directive" about the changes, which was signed by Kennedy and puzzlingly dated May 19, according to the Post. Late Thursday, the CDC updated the immunization schedules. Contradicting what Kennedy said in the video, the CDC did not remove its recommendation for COVID-19 vaccines for healthy children in the child and adolescent immunization schedule. Instead, it added a stipulation that if a child's doctor agrees with the vaccination and parents "desire for their child to be vaccinated," healthy children can get vaccinated. In practice, it is unclear how this change will affect access to the vaccines. Health insurers are required to cover vaccines on the CDC schedules. But, it's yet to be seen if children will only be able to get vaccinated at their doctor's office (rather than a pharmacy or vaccine clinic) or if additional consent forms would be required, etc. Uncertainty about the changes and requirements alone may lead to fewer children getting vaccinated. In the adult immunization schedule, when viewed "by medical condition or other indication" (table 2), the COVID-19 vaccination recommendation for pregnancy is now shaded gray, meaning "no guidance/not applicable." Hovering a cursor over the box brings up the recommendation to "Delay vaccination until after pregnancy if vaccine is indicated." Previously, COVID-19 vaccines were recommended during pregnancy. The change makes it less likely that health insurers will cover the cost of vaccination during pregnancy. The change is at odds with Trump's Food and Drug Administration, which just last week confirmed that pregnancy puts people at increased risk of severe COVID-19 and, therefore, vaccination is recommended. Medical experts have decried the loss of the recommendation, which is also at odds with clear data showing the risks of COVID-19 during pregnancy and the benefits of vaccination. The President of the American College of Obstetricians and Gynecologists (ACOG) put out a statement shortly after the Tuesday video, saying that the organization was "extremely disappointed" with Kennedy's announcement. "It is very clear that COVID-19 infection during pregnancy can be catastrophic and lead to major disability, and it can cause devastating consequences for families," ACOG President Steven Fleischman said. Beth Mole Senior Health Reporter Beth Mole Senior Health Reporter Beth is Ars Technica’s Senior Health Reporter. Beth has a Ph.D. in microbiology from the University of North Carolina at Chapel Hill and attended the Science Communication program at the University of California, Santa Cruz. She specializes in covering infectious diseases, public health, and microbes. 74 Comments

This article contains spoilers for The Last of Us season 2. Season 2 of The Last of Us was undeniably a huge swing, as was the video game it’s based on. The Last of Us Part II features the death of the first game’s protagonist early on and forces the player to play as his killer not only before the deed is done, but for about half of the game part way through the story. It’s a narrative about cycles of violence and the lengths that people will go to protect who they love, but it’s also an exercise in empathy. There’s a difference between embodying a character for hours at a time in a video game and watching a character do the same actions in a TV show. When you spend hours living and breathing and fighting for your life as a character, it’s easy to form an attachment to them, to prescribe our own ideas onto them as our morals inform theirs. Even though there’s not really anything the player can do to affect the overall outcome of the story in The Last of Us Part II, your playstyle is going to affect your experience. One player may try to sneak by the W.L.F. and Seraphite adversaries as Ellie, trying to kill as few people as possible. Another may go in knives and guns blazing, leaving an even larger trail of bodies in their wake. Neither method is “wrong,” but it is going to affect how you interpret the story and the characters as a player. Translating this story and its structure to television was never going to be easy. The first season of The Last of Us had the luxury of adapting a beginning, middle, and end from the story of the first game. Season 1 also had nine episodes to tell the story of a roughly 10-hour (give or take) game and its approximately two-hour DLC meaning that we got to spend close to the same amount of time with the characters in the show as players do in the game. Season 2, on the other hand, is only adapting part of a game that can take upwards of 24 hours to play through, and only had seven episodes to tell this part of the story.  A lot of criticisms people have shared surrounding season 2 of the show are valid. There are parts of the story, especially when Ellie (Bella Ramsey) and Dina (Isabela Merced) get to Seattle, that feel rushed. There are some character choices that are or may seem different from those that are made in the game. But arguably, the heart of The Last of Us Part II’s story is still here, even if this season missed the mark with some aspects. Of course Ellie’s Seattle arc is going to feel rushed when we only get three approximately hour-long episodes to cover it versus the close to eleven hours of gameplay Ellie’s Seattle arc gets in the game. We’re not going to be able to see how Ellie got all of the cuts and bruises that Dina is tending to in the season finale or watch her traverse Seattle in-depth – there’s simply not enough time.  It would have been great to get more time with Ellie and Dina in Seattle. But unfortunately, 13 or even 10 episodes for one season is a luxury that most studios don’t seem to want to afford in the streaming era. Even though The Last of Us co-showrunners Craig Mazin and Neil Druckmann have said that they chose to end the season at this specific point in the story and felt like seven episodes was enough to do so, I still don’t fault them entirely. Trying to do more with less feels more like a symptom of the state of TV and the industry as a whole than something to only blame The Last of Us writers for doing. At some point you get used to doing more with less and less. With the structure of season 2, Mazin says that they “considered everything.” They thought about interlacing the stories of Ellie and Abby, but ultimately realized that switching perspectives halfway through the story is “part of the genetics of how this story functions.” But now that means “we have to take risks as a television show, and HBO is backing us taking risks. But then again, we just did kill Pedro Pascal. Like [HBO] understands that this show is going to be a different show every season, which is a tricky thing to do when you’re a hit show. You keep asking people like, ‘I know you love this, we’re taking it away and giving you this now.’” Understandably not everyone has been on board with these changes. Season 2 of The Last of Us has a consistently lower IMDb score than season 1, and it’s hard to look through any form of social media without finding a mix of reactions from fans who are enjoying the story as it is and others who think that the writers have massacred their favorite characters. But at the same time, Mazin, Druckmann, and TLOU Part II co-writer Halley Gross clearly have a deep love for this story, even if their interpretation of certain character’s decisions doesn’t always align with the audience’s. The characters in the TV show are different than the characters in the game because they experience these events differently. In the show, Ellie has to sit in a hospital recovering for three months before she can even think about chasing Abby and her crew to Seattle. Setting aside that time for recovery is not necessarily something that a video game has to think about – a physical therapy level isn’t exactly something that players of a game like this are going to be excited about.  Join our mailing list Get the best of Den of Geek delivered right to your inbox! It’s not that this version of Ellie is less angry than she is in the game. She’s just had three months to practice burying her anger so it’s more palatable for others. She has to convince the hospital, and Gail (Catherine O’Hara), that she’s fit enough to be released. She has to try and convince the council that she’s fit enough to lead a group to Seattle for justice. She has to convince a pregnant Dina that no matter what happens while they’re in Seattle, that this is the morally right thing for them to do. Because we don’t spend 11-plus hours literally in Ellie’s shoes while watching the TV show, her grief has to be explored in different ways. It’s shown in the brief moment she plays the guitar while waiting for Dina to triangulate a route. Even though Ellie may not be throwing the guitar across the room, there’s still clearly anger mixed with the grief on her face as she plays her and Joel’s song. We see it when she lashes out at Jesse and chooses to go to the aquarium instead of following him to find Tommy. We see it when she screams out in pain in a hospital bed in Jackson. And we see it when Dina tends to her wounds. It’s not that she’s not angry or grieving, we just don’t get to see every single moment of it that we do in the game. And of course Ellie is going to tell Abby that she didn’t mean to hurt her friends and beg her to spare their lives. Abby just shot Jesse dead in front of her and is standing over Tommy with his life in her hands just as she did with Joel. Even if this isn’t exactly how Ellie reacts in the game, it’s a logical trauma response to finally seeing Abby again. Abby was able to kill Joel – someone Ellie looked up to and probably thought was unstoppable as most kids do with their parents in their youth. It makes sense that seeing her again would trigger this kind of response in Ellie too. It’s not that she doesn’t want to kill Abby in this moment – she’s just trying to keep her and her loved ones alive for as long as she can.  We saw her do something similar with David (Scott Shepherd) in season 1. She made herself as non-threatening as possible to get him to let his guard down and then proceeded to viciously attack him. Ellie isn’t a stranger to lying and manipulating to get what she wants, even in stressful circumstances. Why should this be any different? Mazin doesn’t deny that they took some risks with season 2, admitting to The Hollywood Reporter that “I don’t think television is supposed to work like this. We’re clearly breaking quite a few rules, and I love that. And I love it because that is the point. This is not something we’re doing as a gimmick.” Mazin argues that The Last of Us forces us to interrogate what we believe about heroes and villains and see the flaws in that kind of black and white thinking, and he knows that this is “a challenging thing to keep track of emotionally” and that people are going to feel provoked by it. “But part of this story,” he says, “is about examining why we’re so comfortable with following one person’s point of view about everything.” The Last of Us season 2 was never going to be exactly like the game, and that’s okay! When you’ve already made a story that resonates with so many people, it’s not going to be easy to recreate that story in another medium – especially in the streaming era when shows don’t always know if they’re going to be able to get all the seasons they want to tell the story. Time is a luxury that television doesn’t always have. The show may not have hit a home run with every swing they took, but overall the story still lands. The heart of the game and its story of grief and loss and love and violence are still there. Hopefully fans won’t give up on the show just yet and trust that the show’s writers really do care about this story enough to do it justice.

Let’s start with one unambiguous fact: More children are diagnosed with autism today than in the early 1990s. According to a sweeping 2000 analysis by the Centers for Disease Control and Prevention, a range of 2–7 per 1,000, or roughly 0.5 percent of US children, were diagnosed with autism in the 1990s. That figure has risen to 1 in 35 kids, or roughly 3 percent.The apparent rapid increase caught the attention of people like Robert F. Kennedy Jr., who assumed that something had to be changing in the environment to drive it. In 2005, Kennedy, a lawyer and environmental activist at the time, authored an infamous essay in Rolling Stone that primarily placed the blame for the increased prevalence of autism on vaccines. (The article was retracted in 2011 as more studies debunked the vaccine-autism connection.) More recently, he has theorized that a mysterious toxin introduced in the late 1980s must be responsible. Now, as the nation’s top health official leading the Department of Health and Human Services, Kennedy has declared autism an “epidemic.” And, in April, he launched a massive federal effort to find the culprit for the rise in autism rates, calling for researchers to examine a range of suspects: chemicals, molds, vaccines, and perhaps even ultrasounds given to pregnant mothers. “Genes don’t cause epidemics. You need an environmental toxin,” Kennedy said in April when announcing his department’s new autism research project. He argued that too much money had been put into genetic research — “a dead end,” in his words — and his project would be a correction to focus on environmental causes. “That’s where we’re going to find an answer.”But according to many autism scientists I spoke to for this story, Kennedy is looking in exactly the wrong place. Three takeaways from this storyExperts say the increase in US autism rates is mostly explained by the expanding definitions of the condition, as well as more awareness and more screening for it.Scientists have identified hundreds of genes that are associated with autism, building a convincing case that genetics are the most important driver of autism’s development — not, as Health Secretary Robert F. Kennedy Jr. has argued, a single environmental toxin.Researchers fear Kennedy’s fixation on outside toxins could distract from genetic research that has facilitated the development of exciting new therapies that could help those with profound autism.Autism is a complex disorder with a range of manifestations that has long defied simple explanations, and it’s unlikely that we will ever identify a single “cause” of autism.But scientists have learned a lot in the past 50 years, including identifying some of the most important risk factors. They are not, as Kennedy suggests, out in our environment. They are written into our genetics. What appeared to be a massive increase in autism was actually a byproduct of better screening and more awareness. “The way the HHS secretary has been walking about his plans, his goals, he starts out with this basic assumption that nothing worthwhile has been done,” Helen Tager-Flusberg, a psychologist at Boston University who has worked with and studied children with autism for years, said. “Genes play a significant role. We know now that autism runs in families… There is no single underlying factor. Looking for that holy grail is not the best approach.”Doctors who treat children with autism often talk about how they wish they could provide easy answers to the families. The answers being uncovered through genetics research may not be simple per se, but they are answers supported by science.Kennedy is muddying the story, pledging to find a silver-bullet answer where likely none exists. It’s a false promise — one that could cause more anxiety and confusion for the very families Kennedy says he wants to help. Robert F. Kennedy Jr. speaks during a news conference at the Department of Health and Human Services in mid-April to discuss this agency’s efforts to determine the cause of autism. Alex Wong/Getty ImagesThe autism “epidemic” that wasn’tAutism was first described in 1911, and for many decades, researchers and clinicians confused the social challenges and language development difficulties common among those with the condition for a psychological issue. Some child therapists even blamed the condition on bad parenting. But in 1977, a study discovered that identical twins, who share all of their DNA, were much more likely to both be autistic than fraternal twins, who share no more DNA than ordinary siblings. It marked a major breakthrough in autism research, and pushed scientists to begin coalescing around a different theory: There was a biological factor.At the time, this was just a theory — scientists lacked the technology to prove those suspicions at the genetic level. And clinicians were also still trying to work out an even more fundamental question: What exactly was autism? For a long time, the criteria for diagnosing a person with autism was strictly based on speech development. But clinicians were increasingly observing children who could acquire basic language skills but still struggled with social communication — things like misunderstanding nonverbal cues or taking figurative language literally. Psychologists gradually broadened their definition of autism from a strict and narrow focus on language, culminating in a 2013 criteria that included a wide range of social and emotional symptoms with three subtypes — the autism spectrum disorder we’re familiar with today.Along the way, autism had evolved from a niche diagnosis for the severely impaired to something that encompassed far more children. It makes sense then, that as the broad criteria for autism expanded, more and more children would meet it, and autism rates would rise. That’s precisely what happened. And it means that the “epidemic” that Kennedy and other activists have been fixated on is mostly a diagnostic mirage. Historical autism data is spotty and subject to these same historical biases, but if you look at the prevalence of profound autism alone — those who need the highest levels of support — a clearer picture emerges. (There is an ongoing debate in the autism community about whether to use the terminology of “profound autism” or “high support needs” for those who have the most severe form of the condition.) In the ’80s and ’90s, low-support needs individuals would have been less likely to receive an autism diagnosis given the more restrictive criteria and less overall awareness of the disorder, meaning that people with severe autism likely represented most of the roughly 0.5 percent of children diagnosed with autism in the 1990s. (One large analysis from Atlanta examining data from 1996 found that 68 percent of kids ages 3 to 10 diagnosed with autism had an IQ below 70, the typical cutoff for intellectual disability.)By 2025, when about 3 percent of children are being diagnosed with autism, about one in four of those diagnosed are considered to have high-support needs autism, those with most severe manifestation of the condition. That would equal about 0.8 percent of all US children — which would be a fairly marginal increase from autism rates 30 years ago. Or look at it another way: In 2000, as many as 60 percent of the people being diagnosed with autism had an intellectual disability, one of the best indicators of high-support needs autism. In 2022, that percentage was less than 40 percent.As a recently published CDC report on autism prevalence among young children concluded, the increase in autism rates can largely be accounted for by stronger surveillance and more awareness among providers and parents, rather than a novel toxin or some other external factor driving an increase in cases.Other known risk factors — like more people now having babies later in their life, given that parental age is linked to a higher likelihood of autism — are more likely to be a factor than anything Kennedy is pointing at, experts say. “It’s very clear it’s not going to be one environmental toxin,” said Alison Singer, founder of the Autism Science Foundation and parent of a child with profound autism. “If there were a smoking gun, I think they would have found it.”While Kennedy has fixated on vaccines and environmental influences, scientists have gained more precision in mapping human genetics and identifying the biological mechanisms that appear to be a primary cause of autism. And that not only helps us understand why autism develops, but potentially puts long-elusive therapies within reach. It began with an accident in the 1990s. Steven Scherer, now director of the Center for Applied Genomics at the Hospital for Sick Children in Toronto, began his career in the late 1980s trying to identify the gene that caused cystic fibrosis — in collaboration with Francis Collins, who went on to lead the Human Genome Project that successfully sequenced all of the DNA in the human genome in the early 2000s. Scherer and Collins’s teams focused on chromosome 7, identified as a likely target by the primitive genetic research available at the time, a coincidence that would reorient Scherer’s career just a few years later, putting him on the trail of autism’s genetic roots.After four years, the researchers concluded that one gene within chromosome 7 caused cystic fibrosis. Soon after Scherer helped crack the code on cystic fibrosis in the mid-1990s, two parents from California called him: He was the world’s leading expert on chromosome 7, and recent tests had revealed that their children with autism had a problem within that particular chromosome.That very same week, Scherer says, he read the findings of a study by a group at Oxford University, which had looked at the chromosomes of families with two or more kids with autism. They, too, had identified problems within chromosome 7.“So I said, ‘Okay, we’re going to work on autism,’” Scherer told me. He helped coordinate a global research project, uniting his Canadian lab with the Oxford team and groups in the US to run a database that became the Autism Genome Project, still the world’s largest repository of genetic information of people with autism.They had a starting point — one chromosome — but a given chromosome contains hundreds of genes. And humans have, of course, 45 other chromosomes, any of which conceivably might play a role. So over the years, they collected DNA samples from thousands upon thousands of people with autism, sequenced their genes, and then searched for patterns. If the same gene is mutated or missing across a high percentage of autistic people, it goes on the list as potentially associated with the condition. Scientists discovered that autism has not one genetic factor, but many — further evidence that this is a condition of complex origin, in which multiple variables likely play a role in its development, rather than one caused by a single genetic error like sickle-cell anemia.Here is one way to think about how far we have come: Joseph Buxbaum, the director of the Seaver Autism Center for Research and Treatment at the Icahn School of Medicine at Mount Sinai in New York, entered autism genetics research 35 years ago. He recalls scientists being hopeful that they might identify a half dozen or so genes linked to autism.They have now found 500 genes — and Buxbaum told me he believed they might find a thousand before they are through. These genetic factors continue to prove their value in predicting the onset of autism: Scherer pointed to one recent study in which the researchers identified people who all shared a mutation in the SHANK3 gene, one of the first to be associated with autism, but who were otherwise unalike: They were not related and came from different demographic backgrounds. Nevertheless, they had all been diagnosed with autism.Researchers analyze the brain activity of a 14-year-old boy with autism as part of a University of California San Francisco study that involves intensive brain imaging of kids and their parents who have a rare chromosome disruption connected to autism. The study, the Simons Variation in Individuals Project, is a genetics-first approach to studying autism spectrum and related neurodevelopmental disorders. Michael Macor/San Francisco Chronicle via The Associated PressPrecisely how much genetics contributes to the development of autism remains the subject of ongoing study. By analyzing millions of children with autism and their parents for patterns in diagnoses, multiple studies have attributed about 80 percent of a person’s risk of developing autism to their inherited genetic factors. But of course 80 percent is not 100 percent. We don’t yet have the full picture of how or why autism develops. Among identical twins, for example, studies have found that in most cases, if one twin has high-support needs autism, the other does as well, affirming the genetic effect. But there are consistently a small minority of cases — 5 and 10 percent of twin pairs, Scherer told me — in which one twin has relatively low-support needs while the one requires a a high degree of support for their autism.Kennedy is not wholly incorrect to look at environmental factors — researchers theorize that autism may be the result of a complex interaction between a person’s genetics and something they experience in utero. Scientists in autism research are exploring the possible influence when, for example, a person’s mother develops maternal diabetes, high blood sugar that persists throughout pregnancy. And yet even if these other factors do play some role, the researchers I spoke to agree that genetics is, based on what we know now, far and away the most important driver.“We need to figure out how other types of genetics and also environmental factors affect autism’s development,” Scherer said. “There could be environmental changes…involved in some people, but it’s going to be based on their genetics and the pathways that lead them to be susceptible.”While the precise contours of Health Department’s new autism research project is still taking shape, Kennedy has that researchers at the National Institutes of Health will collect data from federal programs such as Medicare and Medicaid and somehow use that information to identify possible environmental exposures that lead to autism. He initially pledged results by September, a timeline that, as outside experts pointed out, may be too fast to allow for a thorough and thoughtful review of the research literature. Kennedy has since backed off on that deadline, promising some initial findings in the fall but with more to come next year.RFK Jr.’s autism commission research risks the accessibility of groundbreaking autism treatmentsIf Kennedy were serious about moving autism science forward, he would be talking more about genetics, not dismissing them. That’s because genetics is where all of the exciting drug development is currently happening.A biotech firm called Jaguar Gene Therapy has received FDA approval to conduct the first clinical trial of a gene therapy for autism, focused on SHANK3. The treatment, developed in part by one of Buxbaum’s colleagues, is a one-time injection that would replace a mutated or missing SHANK3 gene with a functional one. The hope is that the therapy would improve speech and other symptoms among people with high-needs autism who have also been diagnosed with a rare chromosomal deletion disorder called Phelan-McDermid syndrome; many people with this condition also have Autism spectrum disorder.The trial will begin this year with a few infant patients, 2 years old and younger, who have been diagnosed with autism. Jaguar eventually aims to test the therapy on adults over 18 with autism in the future. Patients are supposed to start enrolling this year in the trial, which is focused on first establishing the treatment’s safety; if it proves safe, another round of trials would start to rigorously evaluate its effectiveness.“This is the stuff that three or four years ago sounded like science fiction,” Singer said. “The conversation has really changed from Is this possible? to What are the best methods to do it? And that’s based on genetics.”Researchers at Mount Sinai have also experimented with delivering lithium to patients and seeing if it improves their SHANK3 function. Other gene therapies targeting other genes are in earlier stages of development. Some investigators are experimenting with CRISPR technology, the revolutionary new platform for gene editing, to target the problematic genes that correspond to the onset of autism.But these scientists fear that their work could be slowed by Kennedy’s insistence on hunting for environmental toxins, if federal dollars are instead shifted into his new project. They are already trying to subsist amid deep budget cuts across the many funding streams that support the institutions where they work. “Now we have this massive disruption where instead of doing really key experiments, people are worrying about paying their bills and laying off their staff and things,” Scherer said. “It’s horrible.” For the families of people with high-needs autism, Kennedy’s crusade has stirred conflicting emotions. Alison Singer, the leader of the Autism Science Foundation, is also the parent of a child with profound autism. When I spoke with her, I was struck by the bind that Kennedy’s rhetoric has put people like her and her family in. Singer told me profound autism has not received enough federal support in the past, as more emphasis was placed on individuals who have low support needs included in the expanding definitions of the disorder, and so she appreciates Kennedy giving voice to those families. She believes that he is sincerely empathetic toward their predicament and their feeling that the mainstream discussion about autism has for too long ignored their experiences in favor of patients with lower support needs. But she worries that his obsession with environmental factors will stymie the research that could yield breakthroughs for people like her child.“He feels for those families and genuinely wants to help them,” Singer said. “The problem is he is a data denier. You can’t be so entrenched in your beliefs that you can’t see the data right in front of you. That’s not science.”See More:

This review contains spoilers for The Last of Us season 2 episode 7. It’s hard to follow the heartbreakingly beautiful and emotional performances we saw in last week’s flashback episode of The Last of Us. Pedro Pascal and Bella Ramsey brought their A+ game as this episode revealed the ups and downs of Joel and Ellie’s relationship over the last few years, right up until the night before Joel died. And yet, despite this episode arguably being the best of the season, the season finale still manages to take us on a devastating wild ride. Back in Seattle, Dina (Isabela Merced) and Jesse (Young Mazino) deal with the aftermath of getting caught up in the war between the W.L.F. and the Seraphites. While they wait for Ellie to catch back up with them, they deal with Dina’s arrow to the leg and Jesse gets her all patched up. Even though emotions are high, they are still able to have a touching moment together, and it’s clear that the two still care about each other, even if their feelings are no longer romantic. We get to see another moment of care amidst the chaos once Ellie returns to the theater. She goes to check on Dina, who almost immediately starts taking care of Ellie and her battered and bruised body. Ellie is clearly still processing what she did to Nora (Tati Gabrielle) and what she learned about Abby’s (Kaitlyn Dever) location (which isn’t much). In this vulnerable moment, as Ellie verbally grapples with what she did to Nora, she tells Dina about Salt Lake City, including the fact that Joel killed Abby’s dad. Dina seems surprised by this, realizing that Abby may have been more justified in her revenge than she thought. At this point, it seems like everyone but Ellie is ready to put an end to this revenge mission and head back home to Jackson. Jesse is focused on finding Tommy (Gabriel Luna), especially after finding out that he’s going to be a father. But Ellie can’t let her vengeance go. While they’re looking for Tommy, she realizes that Nora’s clues “whale” and “wheel” probably mean that Abby’s at the aquarium – so she ditches Jesse and sets off in that direction. Ellie has so many close calls – if the W.L.F. weren’t planning an invasion of Seraphite territory at the same time, she would have been hung and fileted like the guy they saw in the park. She finds Owen (Spencer Lord) and Mel (Ariela Barer) seemingly arguing over Abby and whether or not to chase after her. When they spot Ellie, she tries to get them to tell her where Abby is, like we’ve seen Joel do before.  Owen moves for a gun, and Ellie reacts, shooting him in the neck and accidentally nicking an artery in Mel’s. Barer gives a devastating performance in Mel’s last moments as she begs Ellie to cut her open and try to save her baby. Ellie just kind of sits there holding her knife in shock over the fact that Mel was pregnant and she just killed her until Jesse and Tommy come rushing in. Tommy immediately goes to her, holding her and helping her up, while Jesse stares at the carnage for a moment after they leave. Even though he doesn’t say anything, it seems like he’s shaken up too – like he’s staring at himself and Dina if they don’t get out of Seattle soon. They get back to the theater and Ellie finally seems ready to leave. Whether it’s the shock of what she just did to Owen and Mel, her near death experience with the Seraphites, or the fact that her leads to find Abby have simply dried up isn’t fully clear, but at least she’s starting to recognize how dangerous it is for everyone else she cares about if they stay. Tommy tries to calm Ellie’s fears, he reassures her that Owen and Mel were still complicit in Joel’s death even if they didn’t hold the golf club themselves, and then leaves her and Jesse to try and reconcile. Ellie thanks Jesse for coming back for her and the two have a heart to heart. Despite their differences, Jesse knows that Ellie would set the world on fire to save him. Their moment is unfortunately short-lived when they hear a commotion in the lobby. They rush out and Jesse is immediately shot dead. We then see Abby for the first time since she killed Joel, now standing over Tommy pointing a gun at his head. Ellie pleads with her, telling her that she’s the one that she wants. There’s true fear in Ellie’s eyes as she worries that Abby will kill more people that she loves, more of her community. Abby then says to Ellie with an intense ferocity, “I let you live. I let you live, and you wasted it,” before turning her gun to Ellie. The screen goes black as we hear a gunshot in the background, not knowing who fired the gun nor who was potentially hit. The episode then goes back in time, taking us to the start of Abby’s journey these last few days. Some may feel like this ending is odd or abrupt, but given how The Last of Us Part II’s story is split up between the two protagonists, Ellie and Abby, and the fact that this season only had seven episodes, this is arguably the best place to end season 2 and a great way to tease what’s to come in season 3. It may not feel as concise as season 1’s ending, but that’s because season 1 had a clear beginning and end to the story. The source material for season 2 is a lot heftier and there’s plenty more of The Last of Us Part II left to be adapted for the show. This episode is a great way to effectively end Ellie’s Seattle arc and this chapter of the story while preparing the audience for next season to shift gears, and protagonists. Hopefully fans of the show will keep an open mind as we prepare to see Abby’s side of the story next season. I know I’ll be seated and ready whenever that time comes. Join our mailing list Get the best of Den of Geek delivered right to your inbox! All seven episodes of The Last of Us season 2 are available to stream on (soon-to-be HBO) Max now. Learn more about Den of Geek’s review process and why you can trust our recommendations here.