Approved Thoughts

1. The Kids Are Not All Right via MR: Reminder that humans, often the smartest humans, can believe anything. Wonder how much of the coventional wisdom today falls in this category.

   Two weeks after his birth, Jeffrey’s health took a turn for the worse: He developed a heart defect common in premature infants – patent ductus arteriosus, or PDA. Jeffrey was scheduled for open-heart surgery and transferred to the nearest children’s hospital.

   In those days, surgery for PDA was invasive. Holes were cut on either side of Jeffrey’s tiny neck and chest to insert a catheter into his jugular vein. His little body was opened from breastbone to backbone, his flesh lifted aside, ribs pried apart, and a blood vessel near his heart tied off, and then all the tissues were stitched back together.

   Baby Jeffrey felt everything – every incision, every internal repair, every stitch. The medical team had not given their fragile patient any drugs, any comfort, anything to protect him from the excruciating pain of open-heart surgery – just a paralyzing agent to keep him still during the procedure.

   Five weeks later, Jeffrey passed away.

   In the days before her child’s death, Jill Lawson learned a shocking fact: Anesthetizing babies for surgery was not common practice. After Jeffrey died, Lawson called his doctor for reassurance. Surely, she thought, her child had been given something for the pain.

   “The anesthesiologist informed me that she had not used any anesthesia or analgesia on Jeffrey,” Lawson wrote in an account of her son’s experience. The doctor told the grieving mother it hadn’t even occurred to her to do so because it had never been demonstrated that babies can feel pain.

2. @peelraja via MR:

   The paradox of India:

   Punjab is over 60% vegetarian, but Tandoori chicken and butter chicken are its most popular dishes outside the state.

   Tamil Nadu is less than 1% vegetarian, but its "pure veg" idly, dosa, sambhar, pongal, etc are its most popular dishes outside the state.

3. The Seneca via Daring Fireball:

   The Seneca is my middle finger to the aesthetic homogeneity and economic over-optimization of 21st century life.

   A keyboard for the low, low price of $3,600. Yes, a computer keyboard. Yes, US$. Although to be fair, it is not $3,600. Its "from $3,600".

4. My Thoughts on the Future of "AI" via Simon Willison:

   I have very wide error bars on the potential future of large language models, and I think you should too.

   Specifically, I wouldn't be surprised if, in three to five years, language models are capable of performing most (all?) cognitive economically-useful tasks beyond the level of human experts.

   And I also wouldn't be surprised if, in five years, the best models we have are better than the ones we have today, but only in “normal” ways where costs continue to decrease considerably and capabilities continue to get better but there's no fundamental paradigm shift that upends the world order.

5. The Anatomy of Marital Happiness via MR:

   Since 1972, the General Social Survey has periodically asked whether people are happy with Yes, Maybe or No type answers. Here I use a net “happiness” measure, which is percentage Yes less percentage No with Maybe treated as zero.

   Average happiness is around +20 on this scale for all respondents from 1972 to the last pre-pandemic survey (2018). However, there is a wide gap of around 30 points between married and unmarried respondents.

   This “marital premium” is this paper’s subject. I describe how this premium varies across and within population groups. These include standard socio demographics (age, sex, race education, income) and more. I find little variety and thereby surface a notable regularity in US socio demography: there is a substantial marital premium for every group and subgroup I analyze, and this premium is usually close to the overall 30-point average.

   This holds not just for standard characteristics but also for those directly related to marriage like children and sex (and sex preference). I also find a “cohabitation premium”, but it is much smaller (10 points) than the marital premium. The analysis is mainly visual, and there is inevitably some interesting variety across seventeen figures, such as a 5-point increase in recent years.

1. Australian man survives 100 days with artificial heart in world-first success via Nico McCarty:

   An Australian man with heart failure has become the first person in the world to walk out of a hospital with a total artificial heart implant.

   The Australian researchers and doctors behind the operation announced on Wednesday that the implant had been an “unmitigated clinical success” after the man lived with the device for more than 100 days before receiving a donor heart transplant in early March.

   The BiVACOR total artificial heart, invented by Queensland-born Dr Daniel Timms, is the world’s first implantable rotary blood pump that can act as a complete replacement for a human heart, using magnetic levitation technology to replicate the natural blood flow of a healthy heart.

2. Inflation Expectations by Political Party Affiliation: Entry #3692 in "Politics makes you stupid".

3. Broccoli, the Man – and Vegetable – Behind the Bond Franchise: What a story.

4. The shift from a mindset of scarcity to one of abundance is overdue.(NYT Paywall) The article itself has a strong partisan tone. I hope the book is different.

   In the time California has spent failing to complete its 500-mile high-speed rail system, China has built more than 23,000 miles of high-speed rail. China does not spend years debating with judges over whether it needs to move a storage facility. That power leads to abuse and imperiousness. It also leads to trains.

   I do not want America to become China. But I do want it to be able to build trains.

5. Some Vegans Were Harmed in the Watching of This Movie (NYT Paywall): Even as a vegetarian, this seems over the top.

   “People might think a glass of milk is innocuous,” she said. “It’s not. It’s full of violence.”

6. In Search of a Boring Business (NYT Paywall):

   On BizBuySell, the popular listings site where the Rizzos found the Smiths, “corporate refugees” ditching the 9-to-5 have surged to 42 percent of buyers, roughly double the 2021 figure. Meanwhile, nearly a quarter of American small businesses are owned by people 65 and older, making the Smiths part of a “silver tsunami” of sellers.

7. Cognitive security is now as important as basic literacy: Must read link of the week. Between this stuff and use of AI in scams, I am more than a little scared.

8. Is this good or bad? Policy can be hard.

1. James Harrison: The Man Who Saved 2.4 Million Babies: Hall of Fame.
2. Coincidentally, this week both Tyler Cowen and Karthik S shared memories of playing card games as kids. And then today my sister mentioned my younger newphew had started playing Dungeons and Dragons recently. We have played card games and board games with our boys, especially during COVID and it was fun. Clearly we should do more of this.
3. Another entry to the hall of fame. A different hall, sure, but come one, clearly this guy is a legend.
4. A different way of learning math? Via Zvi, who belongs in a third hall. Can't fault him for lacking ambition but perhaps this is the rare tweet that should have been an article.
5. A Cheeto Shaped Like the Pokemon Charizard Sells for Nearly $90,000. NYT Paywall but do you really need to read anything else?
6. Honey on a razor blade: Something about the actual visualization struck a chord with me.

1. A.I. and Vibecoding Helped Me Create My Own Software: The limitation here is my imagination.

2. Humanoid Robots are coming. Wild dogs will not stop me from paying good money for a bot that can iron and fold clothes.

3. First world problem?: Nebraska Man Struggles to Change Daughter’s Name From ‘Unakite Thirteen Hotel’. For two years.

4. Michelle Trachtenberg: I did not watch Buffy or any of her other work but it is always depressing to see so many child celebrities die early deaths. This job is hazardous. To be clear cause of death is not known and I do not want to speculate.

   What struck me was this bit in the report:

   In a 2012 interview with Complex magazine, she recalled a scene in the 2006 film “Beautiful Ohio,” starring William Hurt and Rita Wilson, that featured her “naked tush.” It was, she said, “probably one of the most horrendous moments of my life.”

   “It would take an army — or Martin Scorsese — to ever get me naked again,” she added.

As a viewer, I almost never think about the person behind the character. Sure if a Christian Bale loses nearly 30kg for a role, its visible enough and stark enough that one is forced to think about the how of it. But for the most part, I watch the program and move on. Maybe if there is something really interesting or shocking, a comment to my SO.

So it is a jarring to see that the impact of a brief scene stuck with the actor for years after the fact.

These days there is enough nudity on screen that most of us are desensitized to seeing it and I hope the actors are too, to performing them. Even in 2006 or 2012, when maybe it was less common, I would not have imagined an actor, and especially an American¹ actor, would be so traumatized by such a scene.

Maybe it is because she grew up in a time when that was a much bigger deal that it was harder on her. Perhaps actors like her were the trailblazers and paid a price and it is easier for actors now.

It is, however, possible to imagine a possibility where being naked on screen is traumatizing, no matter how common and acceptable. According to Claude, the "being naked in public" dream genre is common across cultures and age groups and more common in adoloscents.

If that is the case I shudder to imagine the cost of all the content on Netflix and other services in a few years time.

The report also has a bit on another scene she was not happy about. And guess what? At the time (and thankfully no longer), the report carried a photo from that scene.

I need to sit down and combine some of these thoughts into a coherent piece but for now, I just want to dump it all here for reference later. I doubt any of these are original and I have certainly seen some elsewhere.

1. Market structure for LLM makers may end up being like the airline market. High fixed cost to set up, hard to create a product differentiation that users care about (other than price), lots of competitors entering, at least in part due to the "prestige" of owning one. Lots of utility for consumers but hardly any profit for producers.

2. An LLM with an infinite context window, one that can contain all my life, will be an entirely different product than an LLM with a limited context window. You can never have enough when it comes to context windows.

3. Notwithstanding 1, "personality" makes a huge difference in the experience of working with an LLM and the ability to create the right one could determine whether a model can dominate a market or a niche. Claude Sonnet 3.5 absolutely had than special sauce in my experience. We need a lot more of it.

4. This.:

   I don’t see how we’re going to avoid a situation where the internet become lousy with AI-created, pseudo academic writing filled with made up facts and quotes, which will then get cemented into “knowledge” as those articles become the training fodder for future models.

   But combine it with the fact that if we all start getting our answers from LLMs, the online content & ad based business model goes caput and then what is the incentive for people to put up good content on the web? None.

5. But anyone who has good proprietary, verified, high-quality data & content will potentially control the value for the customer even as base LLMs become a commodity. Therefore does more data and content start going behind the paywall? If it doesn't it becomes training data and cannot be monetized.

1. 50 years of Travel Tips: Some great tips.

   For the best travel experiences you need either a lot of money, or a lot of time. Of the two modes, it is far better to have more time than money. Although it tries, money cannot buy what time delivers. You have enough time to attend the rare festival, to learn some new words, to understand what the real prices are, to wait out the weather, or to get to that place that takes a week in a jeep. Time is the one resource you can give yourself, so take advantage of this if you are young without money.

2. 100% HIV Prevention:

   ...we now have evidence that in real-world use it can do something that has never been seen before: twice-yearly injections in thousands of female trial volunteers in high-risk areas (South Africa and Uganda) showed a one hundred per cent prevention rate of HIV infection.

3. Stable Systems (X):

   I don’t think a lot of people appreciate how much of their overall lifestyle and relative certainty is backstopped by a steady, boring stability of systems they don’t understand or even realize exist.

4. AI or No I (X):

   Me using LLMs for fun little personal projects: wow this thing is such a genius why do we even need humans anymore.

   Me trying to deploy LLMs in messy real-world environments: why is this thing so unbelievably stupid and dumb.

   This fits with my experience. But also for doing "real" work, prompt engineering matters. Giving the right instructions in the right order does make a difference.

   See more scepticism in this thread. Along with some of the comments I am hearing Satya Nadella made in his podcast with Dwarkesh, I wonder if we are seeing a vibe shift on AI?

5. My LLM codegen workflow atm(via):

   I have been building so many small products using LLMs. It has been fun, and useful. However, there are pitfalls that can waste so much time. A while back a friend asked me how I was using LLMs to write software. I thought “oh boy. how much time do you have!” and thus this post.

Again, for real work, prompt engineering matters.

6. Consensus projections on AI (X):

   I think Grok 3 came in right at expectations, so I don't think there is much to update in terms of consensus projections on AI: still accelerating development, speed is a moat, compute still matters, no obvious secret sauce to making a frontier model if you have talent & chips.

7. Three Observations: Obligatory link to every Sam Altman post.

   Anyone in 2035 should be able to marshall the intellectual capacity equivalent to everyone in 2025; everyone should have access to unlimited genius to direct however they can imagine. There is a great deal of talent right now without the resources to fully express itself, and if we change that, the resulting creative output of the world will lead to tremendous benefits for us all.

1. Translating with LLMs: A fantastic guide from a professional translator on how he is using LLMs to do his job better.

   For my day to day queries, I simply ask Claude, like I would ask a colleague or a friend and it just works. But for professional work, correctly prompting the LLMs improves the quality of output significantly. I hope to find (and create) more such examples.

2. Wisdom:

3. Wisdom II: A lot of AI cheerleading takes this form. Yes, you can just do things. I have also thought and posted along those lines. But as Gergely points out, the hard thing in building a payment system is not the code - its fraud, AML, KYC, tax, reporting, reconcilliation...

   That is true for most other hard things too. The hard thing in pharma is not finding the compound but getting the trials and approvals done. The hard thing in politics in not the manifesto but the consensus.

4. Tarriffs and Modern Supply Chains: Tarriffs are more disruptive than I thought.

5. How My Dad Reconciled His God and His Gay Son: Beautiful.

   This is me with my dad, Bill White. For decades, he has been an evangelical pastor. Before I was born, he wrote a letter to my future wife. He didn’t know what we both do now: that I’m gay. When I came out nearly 16 years later, it shook his faith and fractured his church. But it never separated us. I wanted to understand how. So I read his journals.

6. Strongly endorse:

   One of the reasons making things is satisfying is that it's a concrete demonstration of the notion that you can have some amount of control over your environment and circumstances. Passive media consumption is the opposite of that.

7. Should you value the life of your brother more than that of a stranger?:

   But again, this is a distraction from any real issue! Oh, you should value the life of your brother more than a stranger? You don’t say? I’m hearing this for the first time! Now let’s kill five million foreign children to fund one sixth of a broadband boondoggle.

   I am happy to “concede” that if you face a choice between saving a stranger and saving your brother, save your brother! Or your cousin, or your great-uncle, or your seven-times-great-grand-nephew-twice-removed. I’ll “concede” all of this, immediately, because it’s all fake; none of your relatives were ever in any danger. The only point of this whole style of philosophical discussion is so that you can sound wise as you say “Ah, but is not saving your brother more important than saving a complete stranger?” then sentence five million strangers to death for basically no benefit while your brother continues to be a successful real estate agent in Des Moines.

   When will I be able to write so well?

Most books should have been blog posts, most blog posts should have been tweets and most tweets should never have been written. - Unknown Tweeter

Another set of AI-heavy links. I can’t help myself right now.

1. Set up a fake cricket league:

   These arrangements might have suggested that the league featured professional-grade cricket that an online audience would find worth watching, but in fact, the players weren’t established cricketers or even skilled amateurs. They were locals that Davda had recruited with the promise of paying 400 rupees (about £3.50) per day — twice what people in the area make in daily wages working on farms and for local businesses.

   Respect.

2. How to choose a religion (NYT Paywall):

   But for the general obligation imposed upon us all, as time-bound creatures in a world shot through with intimations of transcendence, a different Eliot line is apt: “For us, there is only the trying. The rest is not our business.”

   I didn’t know Eliot plagiarized The Gita. The book is called Believe: Why Everyone Should Be Religious. I hope to read this book.

3. I am (not) a Failure: Lessons Learned From Six (and a half) Failed Startup Attempts. I am a sucker for these stories, having a failed startup in my recent past. The conclusion conveniently helps me feel good about the failure.

4. Agency:

5. How to prepare for an AI future: Coincidentally, I ran into a number of pieces on this topic last week. The first one from Tyler Cowen and the second from Nate Silver. Both have the same advice in the first place: use LLMs more than you do and for more things than you do.

6. Using AI to build a nuclear fusor: Combining the wisdom of the previous two links.

7. Using AI to improve learning: I would love to see how the tutor was created. Perhaps I, too, need to take the advice from links 4 & 5 and do this project myself.

1. Why Is Homeowners Insurance Getting So Expensive? by Brian Potter via MR: A great piece objectively exploring a question getting a lot of coverage after the recent LA fires. There’s so much I love about this link. Original, deep analysis of an important question debunking the popular narrative (climate change, see next two links). Also, a great blog on Construction Physics? With 51K subscribers! Isn’t internet the best?

2. Gell-Mann Amnesia: First described by Michael Crichton and apparently nothing to do with Gell-Mann. He just gave it an important sounding name to have it taken seriously.

   Gell -Mann Amnesia effect is as follows. You open the newspaper to an article on some subject you know well. In Murray's case, physics. In mine, show business. You read the article and see the journalist has absolutely no understanding of either the facts or the issues. Often, the article is so wrong it actually presents the story backward—reversing cause and effect. I call these the "wet streets cause rain" stories. Paper's full of them.

   In any case, you read with exasperation or amusement the multiple errors in a story, and then turn the page to national or international affairs, and read as if the rest of the newspaper was somehow more accurate about Palestine than the baloney you just read. You turn the page, and forget what you know.

   I should make a spaced repetition card for this.

3. Bjorn Lomborg: A good follow if you want to see less depressing news about climate change, for instance that polar bear population has been increasing since the 1960s or that the Great Barrier Reef is doing just fine. Do note that his work is also subject to Gell-Mann Amnesia.

4. Only reason I am posting this is because I am right now in the land of the great UFO scare of 2025, New Jersey. Trump appointed CIA Director? On Fox News? Former Texas Congressman? There should be an emoji for that. Here’s a palate cleanser (via Alex Tabarrok at MR).

5. Jokic. Watch till the end for the reaction.

6. Is it better to bribe Trump by purchasing his memecoin or his stock? Not financial advice.

No AI in the links at least, although I cannot confirm or deny if an AI was involved in the selection of these links.

Programming note: Since I am traveling for the next few weeks, some of those letters may only contain the links section.

This is the website in question. Bookmark it. Or don’t.

Many people are asking if this can even be called a website. These people are losers. Sad. The finest people I know think its a great website, maybe the best website in the history of the world. Nobody knows websites like I do.

Anyhoo.

It all began when I decided to shift this letter to a website last week. Having purchased a few web domains before, it took me only a few minutes to buy approvedthoughts.com from Cloudflare. Unlike GoDaddy etc. who offer a promotional price bundles and then jack it up on renewals, Cloudflare offers at-cost registration + renewals and I have good experience with some of their other services, so it was an easy choice.

For all my talk about using AI models, I hadn’t really used them for any type of programming project / tech support roles. I figured this was the perfect opportunity for me to try out if I could actually create something useful with the help of one of these models.

For context, while I have an undergraduate degree in Computer Applications, I have never programmed professionally and frankly did barely any programming during the course itself. I am probably a bit better than someone who has never done any programming, but not by a lot.

For context also, I could obviously just spend $20 on a paid service that would be absolutely trivial to set up and manage but where’s the fun in that?

This is how I started on Claude 3.5 Sonnet:

I have recently purchased a domain for a website. I used cloudflare for purchasing the domain. I would now like to set up a Wordpress or other website at the domain. Walk me through the steps of doing this.

The Sonnet is the “most intelligent model” by Anthropic. It is the go to model at the moment for a lot of the people I follow and the only model I pay for.

The project went through multiple chats because after a while, in each chat, I would start getting this message:

Here, for the first time, I used a trick I had recently seen (I think) on Twitter.

ok - I have saved and deployed on cloudflare. However I am getting the message that this chat has become too long. Can you provide a status update that I can use with another AI chatbot to continue this work?

And it did! I pasted the update in another chat and the conversation continued there as if it was the same chat!

You can check out the first chat [here. I think / hope it has no confidential information. I won’t be sharing the remaining chats because they do have some of my account credentials. This one should provide enough context for most of what I am discussing here though.

So How’d it Go?

Well you’ve seen the “website”, so clearly not as well as I’d hoped. That said, I am continuing with the project and based on what I have learned so far, I am hopeful I will eventually get it working.

It is still remarkable that I got this far, considering that I have zero experience with the technologies / services involved. Claude successfully walked me through setting up 3 different services (Cloudflare, GitHub and Hugo) and had them work with each other to serve up the site.

All of this happened within a couple of hours and with me blindly following its instructions - copying and pasting code where it asked me to, choosing the options it asked me to and only making high level decisions, like which theme I liked for the blog, which require no technical sophistication.

For the most part, all of this was “one-shot”, meaning it told me to do something, in plain English, I did the thing and it all worked. Where it required a second shot, it was because I hadn’t provided the full context e.g. that I was using a Mac and not a PC. And once I clarified that, it adjusted the directions for all subsequent steps.

In fact, because I use Claude almost exclusively, I had forgotten how much better it is that some of the other models at being almost “human”. This chat is not very different from how I would interact with IT support engineer at work. Just randomly dropping bits of context or asides and having it be incorporated correctly into subsequent interactions as relevant.

In one of the subsequent chats, it seemed to me like Claude was stuck or going around in circles, without fixing the problem, so I decided to switch to the newly launched DeepSeek-R1 reasoning model. While the R1 made immediate progress where Claude was stuck, the experience of working with the R1 was just so much harder.

Unlike Claude, which “understood” that I was not technically sophisticated, and only gave tasks of a manageable size at each stage, the R1 would just throw a laundry list of complex tasks at me.

For e.g. there’s a set of 3 commands I needed to run on the Terminal app on my Mac, after every change to the code. This would “push” the changes to GitHub. Every time there would be only minor changes to these commands. Claude would provide these commands every time, unprompted. R1 would not.

Claude also seemed to just understand better where we were in the overall project at all times. So if it asked me to do three things and if I went back with a question about the first thing, it would help me fix that and then go back to the remaining two items in the list. R1 did not seem to have as good an “understanding” of the overall project and did not seem to “appreciate” that if I was asking a question about step 2 of 5, that steps 3-5 still needed to be done after it answered the question about step 2.

Of course, the answer to step 2 would have 5 steps, so now I have to scroll back and forth to figure out which step of which step I am on. So despite being a “stronger” model, I found it discouraging and just much harder with the R1 to have to keep track of where we were in the project.

What went wrong?

I suspect the website would be looking a lot better if I had just followed Claude’s guidance. In fact, even now, I think it is at a stage where I can post these weekly updates there with some extra effort.

But I knew I was going to be traveling for a few weeks and so after the site was ready for me to start posting content, I asked Claude if we could also set up a CMS, content management system, for the website. That way I could post from anywhere instead of just from my desktop and include links and images in the posts with minimal effort.

Things went off the rails pretty much immediately. Without getting into a lot of detail, we encountered a bug while setting up the authorization setup for the CMS. Claude kept suggesting various tweaks but nothing worked. It felt to me like we were going around in circles, trying the same things over and over again.

I even asked Claude to look at the entire conversation and figure out if it was going around in circles. It gave itself a clean chit. Of the 6 or so hours I have spend on this project, my guess is 4-5 have been trying to resolve this issue.

My suspicion is that Claude is trained on an older version of the various services / projects I am using and therefore could not fix these errors.

Eventually I gave up with Claude and took the project over to DeepSeek R1. It made some progress, identifying some errors Claude had made, e.g. key files located in wrong folders, wrong parameters etc. However since neither of the models can really take over and look at the code themselves, they may not be able to fix issues caused by my mistakes in implementing their directions.

Maybe AI agents will fix this issue in future. In any event, the project is on pause as I travel and I am wondering whether I should start from scratch with DeepSeek instead of trying to fix the current version.

What did I learn?

A lot.

Despite the hiccup, I feel comfortable that these models can help me do projects that I just couldn’t have done before. I intend to try more projects, and of course, finish this one.

Different models have very different “skills” and it’s important to figure out which model is the best for a given task.

While Claude is the easiest to talk to, almost like a person, it really helps to understand how to prompt other models to get the best out of them. Providing more context and better instructions can result in much better results.

Larger context windows will change everything. A model that can keep your entire life in context will be multiple orders of magnitude more helpful than the current models.

I am a little more skeptical than before about LLMs becoming generally intelligent or superintelligent. Their ability to deal with problems that are not in their training data seems limited. Maybe agentic models will overcome some of this but breakthroughs in for e.g. fundamental physics seem unlikely. Obviously I am extrapolating from a tiny sample but that’s the direction in which I have updated my priors.

What’s next?

I have a few ideas I want to work on once (not if!) I finish the website. I wonder if I can set up a web server on my old Mac and run a few basic services off it. For e.g. creating a URL shortening service (like Bitly) or a QR code generation service.

This from Simon Willison is my inspiration. 14 projects in 1 week. No rocket science but just small, useful things here and there. Or, if I am more ambitious, that custom Spaced Repetition app built by Andy Matuschak, I mentioned couple of weeks ago.

Is truly customized software possible? I hope it is.