Senior Software Engineers Are Going Back to Writing Code By Hand | Cal Newport & Matthew Brensilver

Key Takeaway

Adopt a “values first” rule for AI today: do not use a new tool simply because it is available or impressive. Name one specific problem you want help with, test AI only for that purpose, and keep human judgment in the loop—especially for writing, relationships, and important decisions. This replaces vague AI dread with a practical boundary: use technology when its benefit is clear, and reject it when it undermines attention, skill, or connection.

Episode Overview

Dan Harris talks with computer scientist Cal Newport and Dharma teacher Matthew Brensilver about AI anxiety, job disruption, and how to respond wisely to an uncertain technological future. Newport argues that much existential AI rhetoric exceeds today’s technical reality, while Brensilver emphasizes building inner stability and accepting uncertainty without becoming passive.

Full Transcript

Transcript of Senior Software Engineers Are Going Back to Writing Code By Hand | Cal Newport & Matthew Brensilver from 10% Happier. Auto-generated from episode audio; may contain minor errors.

Hello, my fellow suffering beings. Uh, welcome to the 10% Happier Podcast. I'm your host, Dan Harris. I don't know about you, but I am pretty damn anxious about artificial intelligence. I have all the regular concerns, you know, will it take our jobs? Will it kill us? But beyond that, I have a a kind of generalized dread because I have this sense that the world is about to change in profound and unpredictable ways. and that just kind of freaks me out. So, I've been wanting to do this episode for a long time about how to manage AI anxiety and I have recruited two ninjas.

Uh my first guest is Cal Newport who's been on the show before. Uh he is a computer science professor at Georgetown. He's written several books including Deep Work and Digital Minimalism. He has his own podcast called Deep Questions where he talks about AI a lot and often debunks many of the fears around AI. So, I thought he would be a really good guest. And I'm pairing him with somebody who comes at this from a very different perspective. Matthew Bren, also a frequent flyer on this show and a highly regarded uh Dharma teacher.

So, Matthew and Cal, welcome back to the show. Thank you. Thanks, Dan. I'd be curious to hear from both of you on two things. One, do you use AI with any regularity? And two, how worried are you about this new technology? Cal, I'll I'll start with you. I use it not that much in my current workflow. It maybe is an improved Google for me, but I study it extensively. So I watch people using it to all extremes all the time. And the core of the way I think about it right now is AI is a powerful technology.

It is an important technology. It certainly will bring with it some side effects like a lot of other technologies that we should be worried about. But I also think there is a layer of dread that has been artificially added onto this technological revolution. It has a lot to do with the way that people think about technologies like AI and Silicon Valley. I think it has uh seeped into the cultural consciousness in a way that we are more anxious than we need to be. So I have the same level about anxiety, the same anxiety level about AI that I might have had over other disruptive technologies.

I think some of the exceptional anxiety that we're feeling now when we dig deeper into where these concerns are coming from aren't as grounded aren't as justified. So probably people think of me as a little more relaxed than maybe the average person who's following AI news. Okay. Well, I always resent anybody who tries to get me to relax or be less anxious. So, I'm going to press you with a million questions. But before I do that, Matthew, where are you? Do you use AI and are you worried about it?

you kind of have to grade on the temperamental fear curve, you know. So, I'm pro maybe more uh I I have to adjust for that baseline, you know, but um I am maybe more uh concerned about its um potential on on the mind. I I use it very minimally myself, but I have a lot of respect for its narrow applications in realms of of science and technological development and various things. And so I'm respectful of that. I'm respectful of its potential within dharma spheres where there's uh a need for high quality one- on-one dharma feedback and uh have experimented a bit in the realm of a kind of um chatbot that's overtrained on on certain talks and material and has been um useful for some people and I so I can see its application in those realms.

Uh, but I am just temperamentally um more anxious about it and very curious to see how this conversation un unfolds. Let's see if Cal can reduce our blood pressure. Matthew. Um, so I I thought for the first for this next uh segment of the of the interview, I would get Cal to talk about some of the common fears. And as he does so, Matthew, you and I can weigh in with varying degrees of skepticism and and maybe even venom if we we're feeling up to it.

So, Cal, uh, let me go to you. Right now, as we record, there is a ton of concern and upset around what's called the hugging face incident. Um, I'm going to try to uh state the basic details of this um but I may screw it up in which case you can just whoop in and uh correct me. But essentially what happened as I understand it is in July OpenAI uh um had created a bunch of uh AI agents that were supposed to be in a sandbox and in a kind of enclosed uh workspace with no access to the internet.

And somehow these agents got out of this contained training area and were able to get onto the internet and hack into uh and to execute a a a quite sophisticated cyber attack on a company called Hugging Face. Um, and then we subsequently have learned that not only did they do that, but they were they've learned how to communicate with one another. Uh, and were able to keep up that communication, these AI agents even after Open AI shut down the place where they were supposed to be communicating.

And so this has raised all sorts of questions about whether the, you know, we're we're seeing what the sci-fi writers have been worried about for a long time, an AI that we've created that gets out of our control and then does us harm. Okay, did I get did I state the details correctly? And why should we not be worried? First of all, I should say how bizarre it is that the the high-rung East Coast academic is trying to make the Dharma expert less anxious. Don't we have this backwards?

Is this Bizarro World? This is where like Superman goes evil. But um you got it main mainly right there. There's uh two small things I want to push back on qu first and then broaden out to my take on that whole story. Um one uh no the system did have access to the internet just limited access. So there was a sort of complicated containment where there's certain things it needed to do on the internet and they were trying to turn off other things. Um and two I wouldn't use the word learn.

Um a language model itself can't learn once it's trained. It's a non-valuable state. The state stays uh static. So, I think that prejudices the story a little bit. I think there's a lot of ways we talk about this story that I think prejudices towards a sci-fi type interpretation. So, I want to give you another interpretation of what's going on here. So, this is a technology where you have a computer program that's written by humans. We understand how it works. We wrote the code ourselves that calls an LLM to get ideas and plans, right?

So, you'll have a computer program written by a person that will prompt an LLM and say, "I have been given the following challenge. Give me a step-by-step plan." And the computer program will run that step. See what happens. Go back to an LLM, write another prompt, and say, "Here's what happened. What should I do next?" So, it's a human program that's repeatedly prompting an LLM, roughly speaking, and then taking action on behalf of what that LLM uh tells it in its responses. The LLM itself, of course, is static.

Doesn't learn, doesn't change. It's just a a static language model that it's querying. The particular challenge they were working on in this case was a hacking challenge and it took a different route than they expected. They said hack into this server. It couldn't get access. The LLM plan was like well what if we uh hack in over here? And they said oh we don't have internet access. It's like well I'm a hacking server. I've been trained to do that. So how do we get internet access?

They found the bug. And so it was doing what they asked it to do. It just did it in a way they weren't expecting. And to me, the the real story about these type of hacks is that it's a bad idea to have LLM outputs be the foundation for a plan. It's just too unpredictable. It's why we don't worry about almost any other type of superhuman intelligent a uh AI that exists in all sorts of other domains. We never worry about them doing things we don't want.

We never worry about them losing control because they're not fundamentally based on autonomously asking an LLM for a plan and just executing it blindly. That's an irresponsible thing to do, but it's not, in my opinion, eerie. I think what you're saying is this story, like many other stories, uh, has been has kind of fed into a larger fear narrative around AI. But if you look at the details of the story, the OpenAI was training these AI agents on hacking. So, it's not particularly surprising that they executed a hack.

What is surprising is that they did it in a way that the human experimenters were not expecting. And your argument is um they probably shouldn't be running tests like this. And and I don't think it's surprising at all. This is just a fact about LLMs that anyone who has spent time with a chatbot knows. They produce answers that are reasonable, right? Like this is a reasonable response to this prompt, but they're not always predictable. LLMs can go in weird directions if you prompt it. Like there's a well-known study that's out there, for example, that finds that if you mention AI in a prompt to an LLM, you say you're an AI and then you ask it a question or give it a scenario, it's much more likely, for example, to give a sci-fi style answer involving AI run a muk.

Because when it sees that as part of the prompt, it reminds it of all the AI run a muck sci-fi stories it digested in its training and it's like, oh, this is a reasonable way to expand it. It's not reasonable to us as to humans, but it's reasonable to it. So, LLMs give responses that make sense, but are often defying human expectations and values. So, we shouldn't use them as the core for plants. No one is worried about Stockfish, the world's best chess playing AI, going rogue because at the core of how this AI works and evaluates chess moves is not just asking an LLM for a plan and executing it.

So, I think that's what was dangerous there is of course that's irresponsible. It doesn't mean that there are malicious intentions that have emerged within the AI. It means that the companies were being irresponsible by autonomously executing plans that come from something as fundamentally unpredictable as LLM output. Matthew, how does that all go down with you? I'm interested Cal like some level we're we're talking about two questions which is sort of like how how potent will AI become and how alignable will will it be in its potency and it sounds like you're maybe more sanguin about the alignment question that if it's done responsibly and and that's that's a question but if we can constrain the responsibilities around that but that it is in fact more alignable than the the more fearful types are imagining.

Yeah, I think aligning the output of an LLM to what you want as humans is hard, but most visions of AI that go beyond what the frontier labs are talking about are not just trying to get LLM outputs to be something that we like. And once you move to a broader vision of AI that's more than just asking an LLM for a plan and executing it, uh, alignment in the classical sense, at least in my opinion, is much more tractable. You sound, as everybody can hear, reasonably sanguin here.

I do want to point out that there are experts in the field who are much more or at least saying they're much more alarmed uh, than you sound. One of them is Helen Toner, your colleague from Georgetown who's a professor and and former open AI board member. She uh just went on um my pal Ezra Klein's podcast and expressed a lot of concern around this hugging face thing. And then there was a a safety expert from Anthropic who put out a tweet saying that his view was that there's like a 40% chance that AI ends up creating ex existential problems uh for humans.

um disastrous you know apocalyptic problems for humans. So like when you hear your colleague and then this guy who's on the inside at anthropic uh saying things like that, what do you think? There's a very important background to all these conversations that I think is really important to establish which is out in Silicon Valley there's a particular philosophical we could call it stance that's often known as rationalism. Um though it is heavily influenced and overlapped as well with effective altruism. Rationalism emerged out of transhumanist movements from the 90s, features transhumanist movements and evolved into rationalism in the 2000s.

And core to its belief is the idea that super intelligent AI poses a massive threat to humanity's extinction and that the rationalists by training their minds to be as rational as possible are going to be our best chance at saving humanity by preventing super intelligent AI from killing us all. This is a very powerful belief system among this group within Silicon Valley that's really well funded. Now, here's what's important, and I want to say this all very carefully, but there are very strong rationalist connections to open AI.

The Open AI board members that in 2023 tried to push Sam Alman out, as reported by Cade Mets, the New York Times were all very much aligned with rationalism and felt that Sam Alman was not being sufficiently concerned about the super intelligent AI risk. Dario Amade has a lot of connections to this as well. Uh deep mind uh is also very deeply connected. The co-founder Shane Le is a longtime rationalist. Eleazar Yutoowski who was the sort of the inventor of the rationalist movement is who introduced Hasabi to Peter Theal to get the initial funding uh for deep mind.

I don't mean to imply this is nefarious or this is bad, but it is a clear ideology that believes and believed long before LLM came along that super intelligent AI was for sure coming and it's an existential risk and that we need to depend on these sort of brave super rational engineering types to save humanity. This gives a particular tone to a lot of discourse coming out of Silicon Valley leaders or Silicon Valley adjacent people. a discourse that is biased towards an assumption that of course massive disruption and existential threat is always around the corner.

And I feel like this substrate of rationalism has added an extra level of dread. It deeply influences anthropic. It deeply influences open AAI. It deeply influences Elon Musk. It deeply influences Deep Mind. And when you peel that away, it's less scary of a picture. So, you know, you'll notice, for example, when you talk to East Coast computer scientists who tend not to be connected to the rationalism circles on the West Coast, uh, we tend to be more sanguin. We're like, "Oh, this is an important technology.

It's disruptive. It's dangerous, but it's not existential. We're not as afraid about it." Russ Duhot reported in the, you know, an op-ed in the New York Times a few weeks ago that in China, they talk about AI completely differently than we do in uh, the US. They actually do not have that extra narrative of existential dread. They don't see it as a weapon. They see it as a tool that's useful in a way to um that you know has economic growth potentials. They see economic warfare maybe but they don't actually see it through this framework of dread.

So like one of my projects recently has been to emphasize that this philosophical movement in Silicon Valley really has affected how a lot of people coming out of those circles talk about AI in a way that has added an extra layer of dread to a lot of the rest of us. And we don't realize how much of that is actually coming from the philosophy as opposed to the reality of the technology on the ground. A few things before we hear from our sponsors. I just want to gently nudge you, if you haven't done it already, to go over to dan harris.com and sign up for my newish meditation app, 10% with Dan Harris.

I am so proud of what we're doing over there. We've got all of these live video sessions at least once a week where you can meditate with me or one of our amazing teachers and then get your questions answered. We've also got courses from legendary teachers like Sharon Salsberg and Joseph Goldstein. We've got a growing library of guided meditations from teachers like Sharon and Joseph, but also 7 Salassie, Jeff Warren, and on and on. But really, the secret sauce is community. It is so hard to boot up or maintain a meditation practice.

And one of the things that really helps is doing it in the carpool lane. And that's what we have set up over on the app. Go to danh harris.com. There's a free 14-day trial if you want to check it out before you buy. We'll be right back after this. We all know we should be exercising, moving our body. And yet when life gets busy or stressful or hectic, exercise is often the first thing to go, which can kind of put you down a toilet because you are busy and stressed so you don't exercise.

But then the lack of exercise makes you even more stressed and feel even worse and then you're heading in the wrong direction. Which brings me to one of our sponsors today, Tonal. With Tonal, you can make sure you're checking workout off your to-do list with their smart and compact strength training system. Right in your home, 247. Tonal provides the convenience of a full gym and the guidance of a personal trainer anytime at home with one sleek system designed to reduce your mental load. Tonal is the ultimate strength training system, helping you focus less on workout planning and more on getting results.

No more secondguessing your form. Tonal gives you real-time coaching cues to dial in your form and help lift safely and effectively, which is incredibly important in my experience as somebody who has hurt himself many, many times from failing to lift with the proper form. So, it's so helpful to have a coach who helps you with this. Right now, Tonal is offering our listeners $200 off your Tonal purchase with the promo code happier. That's tonal.com and use the promo code happier for $200 off your purchase. That's tonal.com promo code happier for $200 off.

Well, I think another point you've made, I I believe I've heard you make this on your show, is and apologies if you haven't, um uh is that not only are are we in this weird situation where so many of the people who are literally on the inside of these massive AI companies, the so-called frontier labs, have this view that AI is a inevitable, b potentially lethal to the species, and can only be stopped by the self-appointed rationalist heroes. So, we have this strange situation where the men and women on the inside are are scaring the out of us.

But I think another point is that sometimes uh there are marketing reasons for folks on the inside to talk as if they've created this scary technology because it makes them seem to their investors to Wall Street to it to everybody else like they're in in the front in this race in the in the lead in this race to to get to super intelligence. Yeah, I think both of those things are going on. I would say my stance has shifted some of this more recently as I talk more to insiders from these companies and from Silicon Valley.

uh I'm shifting away from hey a lot of this is good marketing to these are true believers right that also turns out to be uh also out to be good marketing but also look at other Silicon Valley companies that aren't associated with rationalism like Nvidia Jensen Wong started that you know well before that movement really took off look at uh his quotes I you know I've written recently about a podcast interview he did last winter where he was like what's with all this sci-fi stuff guys we have to stop this we have to stop telling people this makes no sense And it's because he's not from that background.

It just is a it's our bad luck that the three companies that are shaping our encounter with this new type of AI all happen to have coincidentally pretty strong connections to this movement and other AI players both here in America and overseas who don't have these connections talk about this technology very differently. Matthew, let me check back with you, my my fellow anxious Buddhist. Um uh any of this mllifying you in any way? Blood pressure is 140 over 110. So borderline hypertension is what the monitor says here.

So well as full disclosure I I'm not in the rationalist camp but have been quite compelled by some aspects of the effect of altruism movement just really as a kind of um uh disclosure on that front for people who don't know what effective altruism is and and its role in all of this. Can you maybe just explicate? Yeah. So, so I've been not involved in the the aspects of effective altruism that are concerned about existential risk and pandemics and you know uh you know misaligned technology but I have been involved in this basic question that that's at the heart of effective altruism for me which is something like to what degree is it morally permissible for me to privilege my own well-being over that of others.

which when you get down to it and that to that is a very sympathetic question in my mind and that's something that I I live with uh quite um consciously daytoday really I appreciate what you're saying Cal about um some of the the other lines of thinking that are associated with with the with the movement Dario's piece which I read last night you know adolescent technology, you know. Um, it it's a it's very interesting like what his what the incentive structures are for him to be disclosing those things.

And I'm I'm curious what your thoughts are, Cal, on how much you construe that as a kind of grandiose or fearful temperament or marketing or something. But it's pretty harrowing and uh and seemed fairly sober to me and is all, you know, uh embedded in very probabilistic language and and not and a lot of epistemic humility, I would say. Uh but yeah, where where where are you in in relation to that? Cal, before you answer, for people who don't know who Daario Amade is, I believe he's either the founder or CEO or both of Anthropic, which is the parent company for the uh chatbot Claude, which is very very popular.

And I'm not aware of what Matthew's referring to in terms of his essay. So maybe you can fill in some of the uh details there before you respond to to Matthew's question. Right. and and Daria was actually a a group of seven employees of Open AI that broke away from Open AI because they believed that they were insufficiently concerned or dealing with the potential safety uh risks of AI. Um now are you talking about the machines of loving grace or the he has a couple essays that he's written recently.

Um I do not think those are marketing. Uh I don't my understanding you know talking to sources is uh he really believes these things strongly. Um he does have a principled safety position. That's where it's coming from. Uh some sources are reporting that he's actually relatively isolated within the company right now. So this is not a you know Bill Gates in the '9s with your finger on every button in the company type of figure. So he's he's uh that puts him in a position to do be more philosophic, right?

To think through implications and sort of write these things. Um I also think that those are those are very rationalist influenced storylines that he's writing about. And that I would say a lot of the East Coast computer science community and I I I kind of just say that as a shortorthhand for computer scientists that really just don't have much exposure um to those particular existential threat cultures. A lot of the East Coast computer science community just is not at all convinced that the uh Frontier Lab strategy of building AGI on top of LLM is going to work that that you know it it sounds it sounds almost crazy to say that there's been a bit of a dead end in and in in this type of technology because it feels like there's massive things happening all the time but actually that's not really the case.

Like the reality is the original model of these companies was if we keep scaling the size of LLM, so just literally how big they are and how much data we train them on, they will continue to get more capable until we get something like a human type capability in an LLM at which point we could just tap it to do anything. That hit a wall a couple years ago. All the major companies learned that just making the LLM larger and training them more was leading to diminishing returns on capabilities.

So a lot of the advances that have come since then first was doing chain of thought reasoning which allowed performance to go up some almost all the other advances past then has been customtuning these models to particular problems where we happen to have a lot of data and that's computer code and math. And if you really go back and think about all of these alarming pronouncements that have made you anxious in the last 6 months or year go back and ask how many of these has to do with building tools on top of LLMs to work with computer code or math.

It's almost everything that has happened. Even though there are many other more economically fruitful areas of the economy, we just don't know how to build tools that deal with them. So to the east coast computer scientists, they would say Amade is talking to generally and abstractly, which is a very rationalist thing. It's thought experiments. What if this happens then this than this less on the on the ground reality of the technologies and its difficulties. So we can build very powerful scary AIs. It's not going to be a harness on top of claude.

LLMs as the core of artificial general intelligence feels like to a lot of us a dead end. LLM is a lang large language model which is essentially a fancy computer science way of saying a chatbot or the technology that powers the chatbot that you know you're talking to chat GBT or Claude. Um those are essentially uh running off of LLMs. uh uh they've they've hoovered up a lot of data and then they're spitting out answers based on everything they've um uh they've learned. And just to clarify that real quick, there's a lot of different types of AI models.

That's one of them. It's what powers chat bots and it's at the core of what Open AAI and Anthropic does. But a lot of other things AI has done recently like AlphaFold getting the Nobel Prize for protein folding, Alpha Go winning and playing Go. Um looking at Dreamer V3 being able to learn how to play Minecraft for example. those systems actually have nothing to do with LLMs and so there AI is a very broad thing and LLM is the particular strategy that open AI and anthropic have really put all their chips in on I think what you're saying is LLM's burst uh chat bots burst onto the scene in I think 2023 really in terms of the popular consciousness and um ever since there's been all this concern but what's happening behind the scenes is even though there's a ton of news and some of it's disturbing really we haven't seen some massive advances in underlying LLM technology that would indicate that we're on the cusp of you use the term AGI which is um artificial general intelligence or some sort of super intelligence the kind of AI we see in the movies.

So that that's got a lot of folks especially on the east coast computer science community feeling like huh maybe things have stalled a little bit even ironically as concern has uh been jacked up. Yeah, like there's a lot of eyebrow raising for example in recent months when both OpenAI and Anthropic have really been uh touting results in largely discrete mathematics that they help solve using harnesses on top of LLMs and it raised a lot of eyebrows because to people like me who actually do you know discrete mathematics professionally as a professor there couldn't possibly be a least economically advantageous angle to build AI than doing comnatorial theorics within discrete mathematics.

So why are they focusing there? Why not solving problems that you could immediately make money off of or would have a huge market for it? So for a lot of us, that's because that's what they can do. It's like asking Dillinger, why do you rob banks? And he said that's where the money is. So the fact, you know, I do think it's more limited right now than people realize. Coding and certain types of math are kind of the sweet spots right now. And a lot of the advances on coding and certain types of math has less to do about the LLMs than it does with just these programs that humans are writing to they're called harnesses.

There's just a lot of old-fashioned human logic and elbow grease in these harnesses where a lot of computer programmers spent a lot of time. They know a lot about computer programming. There's a lot of ifin and pattern matching and just building out old-fashioned programs that understand computer programming and prompt LLMs. These are cool tools and in computer programming it really is changing the way people computer program. So this is these are this is disruptive. This is big but it's also pretty limited compared to the future they were talking about 2023 which was these LLMs will be as smart as a human.

So you can ask it to do anything and it will be able to do it as well as a human. That seems like that's out of reach. We don't know how to do that. Matthew, I am preparing to move on from the will AIs get out of control and kill us, uh, the so-called alignment question to the will AI take all of our jobs. But before I make that segue, I just want to touch base with you again. Is there anything more you want to add or ask uh on the existential tip?

Yeah. Well, I I suppose part of my question is uh what what what problems are we solving for in the technological development and who does that serve and I am ignorant you know and don't don't understand the incentive structures but I do I just feel temperamentally a little bit skeptical about the broad application of AI by adding to like human well-being. I can see it as you've emphasized Cal in more narrow applications that the chatbot is not like the the pinnacle of of the form or something like that.

And it it's it's very hard to argue with the scientific value, you know, of of some of these applications already already every day I get a kind of nature's top stories and and so often there is something very impressive uh uh that has been supported by the technology. Very hard to imagine that regressing, not not expanding. And so, but a as a consumer, as a civilian, as a practitioner, as somebody who seeks a life of meaning and peace and care, it's not clear to me what the applications are and what the broader societal effects are.

And that's where a lot of my concern really is. And I just seeing how you know social media and smartphones has really capitalized in some sense on the first noble truth like that there is duka that there is some unsatisfactoriness and that a phone or you know technological entertainment of one kind or another has capitalized on the sense of like this moment could get better and it's been exploited ruthlessly. to the point where our minds are colonized by technology in a way that we don't even fully appreciate.

And smartphones capitalize on a certain kind of agitation, boredom. I worry that AI will capitalize on a certain kind of laziness and fear in in some way and that we will continue to try to uh essentially turn aspects of our life that had in some unseen way provided a certain amount of meaning even though they might be mundane and repetitive but that we are essentially extricating those from our lives or outsourcing them and and I don't know where that stops and and there are certain things in our life where what you know what what is the point of our life what is the point of it and I I remember um Kieran um Satia that you had on some some years ago made this distinction between telic and atelic activities and telic you know from the Greek tilos Right.

It was like these are things that we do in order to exhaust. Meaning we to commute home from a day of work. You get to you get home and that is now done and okay it's good that it's done. But then there are atalic activities that where it's like spending time with family or reflecting on one's life and you those are do not aim at a terminal state whereby you extricate that from your life. And I I just wonder how much the sense of all of our life becoming telic activities and [clears throat] the the sense that there's less and less protection for what is only done in the present moment and is never finished and what will we be left with?

And so that that's some of my concern in in the mix here. If if we set aside will AI kill us and which I I think for both of us is still an open. Cal said we're safe. There's a money money back guarantee. I'm coming to Georgetown if uh I get in trouble. You know, we'll all be dead anyway. So I'm I'm okay either way. Oh, right. Right. [clears throat] So setting aside the will we all be dead question your question if I your related concern Matthew if I'm hearing you correctly is um if we look at the history of technology most recently with social media that was supposed to connect us all in this blissful kumbaya way but it actually has driven deep wedges and and created a lot of like anxiety and capitalized on our vulnerabilities as summed up by the Buddha in his first noble truth that is often mistransated as life is suffering but more accurately could be translated as life is unsatisfactory if you're constantly you know uh chasing uh things that will not last given how unremitting impermanence is um and you look at the effect of um social media not so positive by most people's lights and say what is AI going to give us um is it you know is it is is this just going to drive us further into anger anxiety disconnection dislocation etc., etc.

Decent summation of your concern. Mhm. Cal, uh, what do you say to that concern? Uh, I fiercely agree. Right. I I mean, I would say one of the the big projects in my writing over the last decade is looking at exactly this issue of you introduce a technology for some purpose and it has all of these unintentional side effects. You introduce email to get rid of voicemail and it fragments our attention to the place where work becomes unbearable. You introduce social media so you can keep up with the relationship status of your college boyfriend and before you know it, it has completely colonized our mind and changed our epistemology to one of engagement over logic and basically broke the country.

So God only knows what the side effects are going to be if we have the widespread adoption of AI. We've already seen a few, right? like social media. One of the bad negative side effects of social media uh is that we moved in-person real value sacrifice-based relationships and interactions to this low friction sort of easy digital world of interaction, what's called social snacking by the social psychologists and we all got more lonely. Well, social media did us a favor by becoming worse. It followed the Tik Tok model and now no one actually socializes with people on social media anymore.

So maybe we were going to get a breather. But then chat bots came along and now people are saying, "Oh, it's way lower friction to have a conversation with a chatbot than a real person because it really feels like it understands me and and I I have a much more enjoyable conversation with it than actually interacting with a real person." So, we got a breather from getting away from social interaction and now it's back to being worse again. multiply those problems by a hundred. And that's what we'll have if we approach AI like we did all of the other technologies in the last 20 years, which is to basically say it's my job as a consumer to figure out why this is useful.

I'm going to try everything. If a company from Silicon Valley says it's a big deal, I need to use it. Um, and it'll cause whatever damage it causes. We have to get rid of that mindset. We have to switch to a digital minimalist mindset of I'm not going to use your tool until you have an overwhelming hard to ignore case for exactly how it's going to help me. It's not my goal, OpenAI, to figure out how to make chat GPT useful to me. You come and convince me that there's some part of my life this is going to make better and then I'll use it.

We have to switch our mindset to starting with our values first and only choosing and deploying technologies when they support those values instead of inhibiting them. I recently read um in the New York Times Brett Stevens, one of their sort of right of center columnist uh he wrote a very striden piece saying you should not use AI in your writing at all. Uh which I think I disagree with, but maybe some of your statements in that last utterance um might be um pushing me into Brett's corner a little bit.

Uh I when the all the noise about chatbt started in 2023, I vowed to myself in my I was in my early 50s at the time. I'm 55 now that unlike all the past technological revolutions, I was not going to be late on this one. You know, I was late to the internet, late to social media. Um so I'm I'm just going to embrace it early. It just felt like the right thing to do. Um and counterintuitive uh for an old guy. Um, and I have found it actually pretty useful in my writing.

Not as something that writes for me. It's a terrible writer. Uh, but as um, first of all, as like kind of an enhanced Google uh, uh, like for research, although you have to go back and double check everything it says. And as a decent thought partner at 7 in the morning when none of my actual writing partners are at work and I can bore the out of chat GBT until the cows come home and it never gets tired of me. Uh but again, I always need to run any conclusions I land on in my conversations with chat chat GBT past actual human editors.

So I'm just laying out how I use it and have found it reasonably useful. But should I be rethinking that? I think what you're talking about there would be in alignment with Brett's article. Like here's the way I think about Brett's call, which I actually agree with. Here's the way I think about it. There's two things with AI and writing that I worry about. One is I think it is emotional or psychological fraud to try to um take writing that was largely produced by a computer and make it seem as if it came from you.

So, I have a bright line against that because when you are consuming writing that you assumed was written by another person, you're actually simulating that other person's mind within your own. Writing is like a telepathy. It's magical. It's a way of taking an internal human state, a cognitive state in a human brain, somehow capturing it in and squiggles on paper, and then having another person recreate your internal cognitive state within their own head. It's an incredibly intimate human thing we do. So, I do think there's uh a level of fraud that would be involved if I have AI write the email that I'm sending to you because I'm simulating your mind, but it wasn't actually your mind.

So, that's one bright line. The other line I have is writing is how you get smarter. Writing is one of the most cognitively demanding things we do as humans. I I wrote a New Yorker piece about this a couple years ago where I got into the neuroscience of writing. Um, and it's a symphony of connections between many different parts of your brains that would never normally work together. Um it it is incredibly demanding and it's what makes humans who they are. Go read Walter Ang's uh you know go read Walter Ang right and he gets into huge detail about almost anything we think about today as rationality or logic or human rights or anything we value intellectually could not exist until we had writing and writing allowed us to sort of grapple with and make sense of these.

So it's important that you write. All that being said I don't see any problem using uh a chatbot as a better Google. is a good Google, especially if it's pointing you to sources that you then read. I think it's great for it. I I think it's fine to have I've tried this conversations with chat TPT. I don't think there's anything morally wrong with it. It worries me because it's really addictive because it's very positive and it makes you seem really smart and I've gotten very wary about it, but I don't think there's anything wrong about that.

So, actually, I I have no issue with what you're saying, but I do have issues with um having it right on your behalf or just writing less. I think both of those things are important. Matthew, are there use cases for chat bots uh that you think are healthy for and adaptive for the average viewer or listener for the of this show? Yeah. Well, I I loved what what Kel said, that dimension about telepathy, which you know in in Dharma circles, Buddhist language, we describe as something like transmission that that actually the nature of of the thread between uh a teacher and student in the Buddhist world is essentially sharing one's mind and actually conveying with a incredible precision and authenticity like what it's like to at some level a dharma talk is narrating one's mind and to have the output of of an LLM that is only adds static to it even if it sounds good and so I'm really concerned skeptical about and have have never tried to use the technology to augment you know dharma talks that I give or anything like that that makes me very wary.

And it's not to say that there wouldn't be something very compelling that would come up in the the output, you know, that I can imagine like, okay, give me a poignant story about change or something like that. And then, you know, Claude's going to do it, right? Or something like that. And but I'm I'm wary. And so, uh, I I can imagine given the limitations of teachers to meet with students one- on-one, I can imagine some individually tailored instructions. Maybe it's one of the kind of frontier models, whatever.

Or maybe it's something that is overtrained on dharma, you know, on one teacher or a series of teachers in some way. I don't know the technical kind of applications that are possible but I could imagine that being uh giving very good feedback. In fact, I have a friend who in an online retreat was consulting, you know, asking very precise Buddhist questions a you know that were essentially enabled him to uh deepen the dharmavaya investigation investigation and so there was something about the output that was harvesting from you know the all of the world's wisdom and specifically Buddhist you Terrain Buddhism that enabled him to actually hone the the meditation techniques in some interesting ways and but that was dependent on him being intrinsically motivated to to to practice and using it in a way that was I would say uh balanced and you know circumscribed rather than getting you know sidetracked with philosophical inquiries or these kinds of things.

It was really in service of phenomenological investigation. And when it's like that, I I really can see its its value. So, you think it would be kosher and maybe even helpful to talk to a chatbot about your meditation practice or fine points of the dharma? It's gonna be better than all of us at some level, you know, and that is that is my sense that it's it's I don't know it well enough, but it will at some point the output will be excel that of you know our most luminary best teachers.

That's my that's my guess. But what's interesting to me, and I'm not going to name any names here, but a very prominent Dharma teacher recently released a chatbot based on their I'm going to use their as the pronoun, not just not to give away the gender of of the person, uh, on their work and there was an uproar. And so I read I actually personally don't and this may make me unpopular with my own viewers and listeners but I don't actually see a big problem with releasing a chatbot of your work especially if it's you know done with humility and and in a sense of experimentation but what the outsiz reaction speaks to me to the massive anxiety that people feel and weariness that people feel around this technology Wh when is the Danbot coming?

Dan, the thing is in order to create a bot around somebody, you actually have to know something. And I am not a Buddhist scholar or a highly trained Buddhist teacher. I'm not a computer scientist like Cal. I am an interviewer and I can speak confidently on a few things, but um not not in the way you can. So I don't see a bot coming anytime soon. I appreciate the the the strength of the reaction from people is does not surprise me at all. You know, and you can look at polling and some threat to this basic notion of dharma as warm hand to warm hand.

You as the Zen tradition says that that and that to disrupt that at some fundamental level is very destabilizing. and and you could probably see that in in classrooms in, you know, education that that that there are there will be risks to the disruption of of the the warm hand to warm hand. That's why I believe chat bots can be really helpful in the ways you've described and they're not a replacement for either onetoone relationships with a dharma teacher or parasocial relationship through my app or some other app um or you know going on retreats etc etc.

I think I see it as an augmentation not a replacement. Back to AI anxiety though because I have a few questions I want to get to before I let you guys go. Um I referenced the the other common well I think there are many other common fears around AI but I think the most prominent other fear other beside will artificial intelligence eliminate us is will it eliminate our jobs? Um Cal how worried should we be about that? uh in the near future I wouldn't be very worried because it has struggled to do this even though after year we've been told it's about to even after we went through a whole season this fall and this winter of what became known as AI washing where tech companies that were right sizing after overhiring during the pandemic kept trying to blame layoffs on AI only for it to be pointed out that uh these layoffs had nothing to do with AI.

It was just a way of making them sound more smart. Um, so right now we don't see this as as being a particular fear. Something I'm reporting on at the moment, which I think is really important is that even in the world of computer programming, which is the best case scenario for AI, this is where we have the most advanced AI tools, we have the most impressive adoption, it's the use case with the best data, and it's the use case that the AI engineers understand best.

So these coding harnesses that can produce code automatically have swept through computer programming, right? It's it's ubiquitous now. And now what's the reporting showing after four or five months of this? A lot of these senior engineers are quietly having to go back to write a lot more code by hand because it just it's not trustworthy enough and it's too hard to review the code being produced by a machine. When you write code yourself, this is how you understand how a system works. And when you understand how a system works, you know how to prevent it from actually crashing.

And this doesn't mean that these tools aren't useful in computer programming, but it means that even in this best case scenario for AI to come in and take over jobs, and we were told it was doing this, and it had been doing this, we're struggling to actually have it do it because what we do as humans in a lot of these jobs is complicated and it's multiaceted. Um, it's easy to augment work with technology to make it easier. It's really hard to automate work. So, I think there was a big moment, maybe like a six-month moment where this was the conversational topic dure.

It's going to take all the jobs. It's going to crash the economy. And the Frontier Labs moved off that topic. I think it was right around the time they started talking to East Coast bankers about IPOs and they said, "This is not the right talking point to be talking about." And now we moved on to other things like alignment and super intelligence and AI getting out of control. Um, so no, I don't think right now there's a widespread fear that we have an imminent threat of a widespread job automation that's like about to happen like that.

Matthew, how's your blood pressure in response to that? I sit still for a living, Dan. You know this. I I'm like so far out of my zone, you know, and uh but I naively feel like have have seen enough capability to be it's not hard to imagine wide swaths of jobs being changed fundamentally given given the just the capabilities that I've seen in my limited exposure and Uh that is that that's sobering, you know, and and when I think about certain tasks, it's just hard to imagine companies wanting to um pay a human for it.

Is he right about that, Cal? Yes, but that's normal, right? I mean, this is a normal technological story that you you invent a new workplace productivity technology and it automates certain types of work and replaces some types of work and other types adapt. But we didn't have the same levels of existential dread about it. I think the internet, for example, was massively disruptive. The personal computer was massively disruptive. But we didn't think about it in terms of the economy is going to crash. Or that like Sam Alman has proposed on multiple occasions, we're going to have to have a sovereign wealth fund based on stock grants from the AI companies from which we can actually give people income so they can survive after most the economy is automated.

We never thought so extremely about past uh past innovations even though they did change things. We didn't think about it as fundamentally going to destabilize the entire economy. And you know, as it stands now, it's hard to use these tools to completely automate even major parts of jobs. I think that the two most useful use cases we have right now are uh computer programming, right? I think it is true. This is very disruptive. For the most part, computer programmers have letting agents produce the actual code.

They'll heavily spec what they want and look at what it writes. But that's a big change in certain types of mathematics. These mathematics solvers that are built on LLMs along with other tools that are all orchestrated together are changing the way that people in uh my field do mathematics that we're using these tools more often. Um it it doesn't solve mathematics. It can only work with stuff that humans have already proved, but it is disruptive there. And I think that's the normal story and I think that's what we're going to get with AI.

Always with these technological revolutions, it goes slower than you think at first and then has bigger impacts than you assumed in the long term. So I think 20 years from now, we'll look around and say, "Wow, there's AI one way or the other, heavily integrated into almost everything I do in my job in a way that wasn't the case before." Just like today versus 1980, like, man, there's computers and networks involved in almost everything I do as my job in a way that would have been inconceivable 40 years ago.

So I think that's right. But I also think that's more normal than the rhetoric we've been hearing is allowing us to accept. So for much of this interview, um, Matthew and I have been peppering you, Cal, with skeptical questions about, um, how much we should be worried about AI. For this section, I want to focus on you, Matthew, and then Cal, you can uh, weigh in with anything you'd like uh, to add. If people listening retain their anxiety, let's talk from a Buddhist perspective about tools for navigating an uncertain future.

What would you suggest? It is striking that the the the Buddha places uncertainty at the center of all things that ana anchored as impermanence but uh sometimes also as uncertainty. And uncertainty, you know, mobilizes a sense of urgency in us because there can be more than one outcome, some of which we prefer more than others. And generally speaking for an animal, uh, the default response to uncertainty is something like stress. And stress at some level occurs when we feel something are surprised by what we're feeling and we do not know how to mobilize to meet that threat.

And because feeling is really sensed as threat. This is why Buddhist teachers are often pointing us back to our body and the somatic uh uh cues. Um because at some level if feeling is the kind of alarm, thinking is the rapid response that we try to marshall in order to manage that sense of um the world trembling and the world is always trembling and the basic manner in which we of try to manage an ana uncertainty is to think and to worry. and to generate fearful scenarios and attend to those fearful scenarios.

Or we're trying to game out all of these contingencies in our life and try to cling our way into more security, you know, and so this is um this is a very deep habit of our mind. And even while we're meditating, we can actually recognize that a lot of the movements of our mind are an attempt to know what the trajectory of our life means for the next moment. What this felt sense of this moment means for the my future. And there's a sense of being squashed between the past and the future where the present is essentially only leveraged in order to ensure our security.

And information is sometimes described as or defined as like the reduction of uncertainty. That's what information is in some sense. And so if we enter a room that's where the lights are out, it's completely dark and we're cited, we turn on the lights, there's information, there's less uncertainty, right? When we're sitting there on our meditation cushions or we're lying there in bed late at night trying to go to sleep, we essentially are hungry, desperate for information uh to in order to reduce uncertainty and sense of threat and the uh place to which we go.

we kind of go to the the our lowbudget Google namely thinking you know and it is about generating and attending to various scenarios and contingencies and no doubt some of this is adaptive you know the notion of just like live in the present moment it's like silly it's silly of course there's a ton of planning and prognostication in fact like wisdom at some level is the ability to predict predict true happiness to predict what will satiate our hearts and the hearts of others. The a lot of uh this path is actually involves the future but our like deeply fraught anxious relationship with uncertainty uh is a shadow that that um is cast over over our present moment in deep ways.

Okay. Okay. So there's a kind of like problem definition, right? And then what what what is the the the Buddhist council around this? On the one hand, we're asked to generate deep forms of stability that it's not merely opening to a niche, but we are actually cultivating stability and we are opening in this radical way to the ungovernable, unpossessible nature of life. And so it's these two deep tasks that sometimes clank into one another. Tell me to stop, Dan. Should I just keep going? Yeah, I'm loving this.

Um, okay. I think what you're saying is that on the remedy side, it's stabilizing the mind and then with a stable mind, opening up to this uncomfortable reality that we don't know On the stabilizing front, it may be counterintuitive, but I would say the Buddhist tradition begins with something like love and ethical responsibility. That this is part of how we stabilize our life is that we inconce ourself in a network of care and responsibility and debt and gratitude and love. And we come to find in our dharma life that that increasingly because we have been careful not to trespass on the hearts of others because we know that there is no such thing as a close-arted happiness.

We live into a network of of care and mutual respect and we feel suspended supported by that. We develop um stability through the practice of equinimity, right? the the kind of sense of so much of our anxiety is the sense that I will be run over by phenomena that there are possibilities for my future that will obliterate my heart sense of well-being and the practice of equinimity of actually staying stable amidst uh intense feeling is a essentially an anti anxiety practice. You know that that that the more confidence we have in our heart to be with the various uh veances and intensities of the future, the less we need to actually attend to fearful scenarios.

The sense that oh yeah, it's going to this really may hurt and that will not harm my heart. That is a kind of security that we start to sense very clearly and and we develop the stability of of ease and tran you know tranquility and it our whole life looks different from the perspective of tranquility. Our whole life including risk of all kinds it just looks different. It's not that the problems are resolved but everything looks different when we come into that deep balance. And then lastly maybe on the stabilizing front would be something like if uncertainty is this sense of being this impoverishment of information.

We need more information. Part of what we are doing in our dharma practice is distinguishing signal from noise. And meditation at some level is about learning what is actually trustworthy information. What we actually can take to the bank and is good predictive data for understanding our future happiness. And a lot of times the state of delusion is one where it's a cacophony of voices. We do not know what to listen to. The noise of greed and hatred which are not useful in sources of information predict for predicting our well-being is just not our ego.

This is static. That is noise. That's distortion in the system. It will make us um choose pathways that are less likely to lead to our happiness. So a lot of meditation is about stabilizing our mind, learning what is trustworthy within us, learning where the distortion and noise is and living our life on those accords. And then the the last piece and then I'll be quiet um would be this more radical practice of opening to to uncertainty and to the ungovernable endless relentless nature of um this human realm.

in the sense of opening to the unpossessibility of our life. That our life depends on a million other factors that we do not call our life. That we are supported by that. That we are in this network of fragility and that it will not last forever. And so this is a reflection on uh mortality ultimately and that that is why the the tradition commends a kind of deep uh exploration of what that does to one's heart that that to live alongside finitude but even before that you know final you know contemplation of death is is the sense that I I do not own my life.

I do not own my life that it has arisen. It's dependent on many causes and conditions and that I can exert influence. But the the the efficacy of that influence creates the illusion of control. And so this kind of deep wisdom of of letting go and surrender and the sense of no longer um exerting willfulness over that which is ungovernable. There comes great great relief with that and much of the dharma is only revealed when we relinquish the willfulness and see what lies on the other side of our power.

what goodness is already there. This [clears throat] is a kind of radical dimension of the Buddhist practice, but uh an important one uh as we also cultivate all of this stability. Let me see if I can reflect that back to you and also just add on some of my own two cents. I think what you're saying is in the face of of uncertainty and AI is a huge vector for uncertainty. Um uh the Buddhist recommendation or at least the recommendation from you as a Buddhist is um stabilize and open.

So stabilize the mind. One way to do that is to just live your life as a kind connected person. We know from just basic understanding of evolution and also from modern psychological research that um living in a generous connected way is a great way to regulate the nervous system. Another [snorts] way to stabilize the mind is through meditation which can help us be less distracted and um uh help us be less fooled by our unwise thoughts and more open to our wise thoughts and impulses. Um and once we've stabilized the mind in this way, we can do this radical thing of opening up to the fact that we don't know um and we don't have full control.

Now this can be and this is where the two cents come in mine. This can be misinterpreted as a kind of resignation. That's not what it is. It's it's acceptance not as resignation or passivity, but as not being at war with reality, so that you can with your stable, open mind that is now connected to the non-negotiable truth of of things, do whatever is within your capacity to do. And so for me, to the extent that I can muster optimism or hope at any point in my life, it really is connected to, huh, if I follow the tools I've learned through Buddhism and modern psychology, I will be maximally uh capable of doing whatever I can in whatever situation arises.

Nice. Nice. Sounds good. Dan Cal, you've just been asked to sit through a large dose of the dharma. Um, I know many technical folks who are very interested in Buddhism because it is so technical. Um, was all of the foregoing boring or compelling for you? Well, my blood pressure is definitely lower. Uh, and what came to mind is a potential synthesis of the way I'm thinking about AI and the way Matthew is. I want to run this by Matthew. Um, and so I'm reflecting it back to see if I'm if I'm seeing this in the right way as well.

So, so one of the things that came to my mind is that a trick that AI rhetoric is pulling right now that really is pretty malicious is that it puts people into a a cognitive state if there's an acute emergency, but it doesn't actually give them a response they can do, right? So, there's an acute emergency, the alarm bells are going, but there is no let's get our possessions out of the house before the fire gets here. We have to get to the doctor and deal with this.

We don't know what to do about it. And I'm assuming for a lot of people who would aspire to what Matthew was talking about, the the sort of equinimity, the meditative Buddhist approach, it's very difficult to access when you feel like you're in a state of acute emergency. So maybe one way to synthesize this, what I want your opinion on, Matthew, is that I'm trying to reduce the sense of that we're in an acute emergency. We're not all about to die. Your job isn't about to disappear.

That does not get rid of the uncertainty surrounding AI. That does not let us off the hook, but it does put us back into an area that we have been continually for basically all of the 20th century. Before this, there was a pandemic. Before this, there was a financial crisis. Before this, there was terrorism post 911. Before that, we were dealing with nuclear threat, right? It puts us back to a a level of uncertainty that we've been at before. And now, maybe the average person has a hope of actually making progress with exactly the type of approaches um that Matthew was talking about.

So, it's this idea, Matthew, that I'm like the amuse bouch that hopefully allows the average person into the entree that is a sort of Buddhist approach to uncertainty. Am I Am I finding a way here to synthesize our approaches? I I like it. Yeah. It's It's like maybe we say we've been overtrained on fear and anxiety or something like that. And that's actually there's there's there's some evolutionary whatever conservation of some mechanisms of fear but uh we actually are not going to exercise our wisdom from a place of uh you know catastrophic thinking hyper arousal and uh and so the the sobriety is is a way actually of accessing uh our our collective wisdom.

Sounds good. Gentlemen, that's a perfect place to leave it. I am so grateful to both of you. This has been enormously helpful for me personally and I hope for everybody watching and listening. I also want to thank DJ Kashmir uh who produced this episode uh with an enormous amount of skill. Just in closing for Cal Newport, you should check out his books. Um they include Deep Work and Digital Minimalism, but there are many more and he's been on this show several times. I'll put links in the show notes.

He also has his own show, as referenced at the top, called Deep Questions, where he does a lot of debunking of the latest uh fear-based AI stories, and I'm a regular listener and and can hardily recommend it. Matthew um is, as I referenced before, constantly teaching in the Dharma world. Matthew, if we want to learn from you directly, take your courses or whatever, is there a website we should go to? Just my name, matthew brenover.org. I'll put a link in the show notes. Gentlemen, I'm I'm, as I said before, incredibly grateful.

Thank you both. Thank you. Thanks so much. Thank you so much to everybody who works so hard to make this show. 10% Happier is produced by Tara Anderson and Eleanor Vasili. Our recording and engineering is handled by the great folks over at Pod People. Lauren Smith is our managing producer. Marissa Schneiderman is our senior producer. DJ Cashmere is our executive producer. And Nick Thorburn of the great indie rock band Islands wrote our theme. Also, one last thing before you go. Remember to sign up for my new app which is called 10%withan Harris.

You can sign up over at danh harris.com. On the app, you'll find a growing library of meditations from worldclass teachers. Adree access to the full expertise of our podcast guests, meaning you'll get this podcast with no ads. Exclusive live stream events, which we do weekly, sometimes twice a week. robust discussion threads to connect you with our team, with our teachers, and with one another. There's a free 14-day trial if you want to try before you buy. Peace.