If AI had existed back in the mid-2000s, careers built on blogging might have looked very different. It's possible that the same career paths wouldn't have existed at all. On the other hand, early bloggers might have just become early adopters of AI, leading to a career that was "different, but the same."

A key difference would be the optimization strategy. In the early days of blogging, the primary goal was to optimize content for Google search results. Now, the main objective would be to optimize content for AI chatbots like ChatGPT or Perplexity. These AI systems have become the new audience, especially for any commercial publisher operating at scale. While niche experts might still maintain a direct relationship with their audience, larger publishers are now focused on figuring out how to get their content picked up and disseminated by AI.

Despite the rapid advancement of AI tools since 2021, business operations have not fundamentally changed. This is not surprising, as AI is currently being used as an add-on for pre-existing work routines, such as writing memos or proofreading. These applications provide marginal gains, but they don't revolutionize how businesses operate.

What we really need to see a major impact is new organizations built around AI and those will be startups. They will come only slowly. It will take 20 or more years before they really transform the economy.

History shows that legacy organizations struggle to adapt to revolutionary technologies. For example, General Motors was paralyzed by Toyota's superior methods in the 1970s, and most old mainstream media outlets could not cope with the rise of the internet. A similar pattern is unfolding with AI, where a complete turnover in the business landscape is needed for the technology to have a large-scale impact.

The first signs of this revolution are appearing in specific industries. Programming is a key example, where some claim AI now performs up to 80% of the work. Sectors with low fixed costs, high competition, and immediate feedback loops, like programming and quantitative finance, are being transformed first. The legal field is also seeing disruption from new, AI-native law firms designed from the ground up. However, legacy law firms face a significant hurdle: data security. They are extremely hesitant to send confidential client information to third-party AI models due to fiduciary responsibilities, a problem that may be solved once AI can be run locally.

Regulated industries like law will be slow to adopt AI due to concerns about data ownership and privacy. Queries sent to AI companies can raise issues, as noted by Sam Altman, who pointed out a significant privacy problem.

AI queries are subject to subpoena and he thinks they should have as much protection as, say, your conversation with your lawyer or your doctor or your therapist.

Until firms can afford their own private models or new legal protections are established, progress in these fields will be limited. In contrast, medical AI is advancing more quickly. Many people are willing to share their medical data with AI systems, leading to innovations like free, accessible medical diagnoses for people worldwide, especially in countries with limited access to doctors. This raises questions for insurers, who could use this explosion of data to price risk with extreme accuracy. This hyper-precision could unravel insurance markets, as the benefit of insurance disappears if a high-risk individual must pay an equally high premium.

Regarding the labor market, the traditional economic view is that technology does not cause mass, long-term unemployment. Instead, it creates disruption, and the savings generated lead to new spending and new jobs elsewhere. While some in the AI field worry about a permanent underclass and the need for UBI, this perspective is shifting. Even some AI insiders are beginning to adopt the economists' view that the adjustment will be slower than initially thought, with jobs changing rather than disappearing entirely. Real-world jobs involve a complex mix of intellect, physical presence, and social interaction that AI cannot easily replicate.

The biggest impact may not be mass unemployment, but a shift in who is most affected. The poor and the very wealthy are expected to do well. The group most at risk is the upper-middle class, whose traditional paths to stability are being disrupted.

People who are sort of upper upper middle class will find this automatic ticket to a law or consulting job that assured they would be upper, upper middle class for the rest of their lives. I think a lot of that is going away already.

In a future where intellectual tasks are augmented by AI, personal presence, social networks, and charisma will become increasingly valuable.

One of the key questions about AI is where its value will show up in the economy and how that translates to government revenue. A significant impact in the United States, in the medium term, will likely be a massive expansion of the healthcare sector. AI will spur the creation of new drugs and medical devices, which will need to be tested and produced, accelerating growth in an already expanding sector.

Furthermore, AI will help people live longer by at least partially fixing various diseases. This increased longevity has a direct financial consequence. Someone who lives to be 94 will spend far more on healthcare over their lifetime than someone who lives to 77, further fueling growth and generating substantial tax revenue for the government.

While some aspects of healthcare, like medical diagnosis, are already becoming very cheap thanks to large language models, other areas of the economy may become effectively free. For instance, a portion of the music industry could shift to individuals creating customized songs at home using AI. This AI-generated content would be a partial substitute for human-created music, though the desire for the human touch—like being a fan of Taylor Swift—will remain.

From a revenue standpoint, this shift isn't a problem. If a person creates a picture with their home AI instead of buying one, they will simply spend that money on something else. The spending is reallocated, not eliminated, meaning it can still be taxed in another part of the economy.

An analysis of the Taylor Swift phenomenon suggests her success stems from several factors. She is super polished, and the internet allows the biggest celebrities to become much bigger than before. Her persona is described as having the guise of being attractive without feeling threatening to other women. This, combined with an all-American and somewhat generic appeal, allows her to attract a wide fanbase. She is also seen as brilliant at managing her career with an incredible work ethic, and her live shows are reportedly incredible.

This leads to a broader discussion on whether culture is dead or simply rehashed, with examples like movie franchises and the dominance of older music on streaming services. However, this view may be an overstatement often held by a pundit class that is no longer actively seeking out new things. A distinction is made between mainstream and niche culture. While the most popular Hollywood movies today may be dreadful compared to past blockbusters like 'The Godfather' or 'Star Wars,' quality has simply shifted.

I don't think they're worse than the movies of earlier times. I do think mainstream Hollywood is much worse. So in many areas you just have quality moving more into niches.

There is a golden age for genres like country and western music and horror movies. The argument that culture is lacking is reframed as a lack of shared culture. While monoliths like Swift and Beyoncé exist, society is no longer gathered around the same media as it once was, with the Super Bowl being a rare exception. Technology, particularly algorithms like Netflix's, contributes to this by serving up very specific, niche content. This funnels people into smaller streams, though often this just directs them to 'slop,' which has always had an audience.

Today offers more access to cinema than ever before. In the past, options were limited to things like the top 40 list, which often featured terrible movies even in good decades like the 1960s. Now, you can watch almost any movie through services like Mubi or by buying DVDs and Blu-rays. This unparalleled access means people can sort themselves into their niche interests. From the perspective of cultural consumption, there has never been a better time to be alive than right now. While it is true that many people abuse this freedom and choose to consume low-quality content, or 'slop', this is hardly a new phenomenon.

One host writes almost every day, but admits this consistency is driven by the obligation of a daily newsletter. Without that requirement, they might just tweet their ideas instead of committing them to longer-form writing. This raises a question for the other host, who has blogged every single day for over 22 years: how do they avoid the temptation to just fire off ideas on Twitter?

The answer is that the temptation doesn't exist. This host prefers to think things out and write properly. For them, it doesn't require discipline to write; the discipline is in holding back.

I don't feel it requires any discipline from me. The discipline is not writing, more like I have to restrain myself.

There is no special trick or formula. Instead, it's about finding a niche that fits so well that the work becomes a natural daily practice rather than a chore.

When considering how chatbots might have changed the early days of blogging, the intuition is that people will still want to read human writers simply because they are human. Even if a bot becomes as good as a human writer, the desire for a human-to-human connection will likely persist. This is similar to music; while AI-generated music might capture a portion of the market, listeners will probably still seek out human artists.

The early, or "glory days," of blogging had a spirit of liberal linking and idea exploration. This stands in contrast to platforms like Twitter. Though once called a "microblogging site," Twitter is fundamentally different. It tends to be much more conflictual and focused on one-upmanship.

Twitter seems too heavy on memes, and meme-heavy media have a greater potential for racism. Many people, both those writing on the platform and those reading it, seem to be driven crazy by their engagement with it. The platform has also become very sexist in unappreciated ways.

AI is set to change the role of human economists. Instead of performing econometric and statistical analysis, their time will be better spent gathering data and feeding it to AI models. The AI will handle the hard, boring, and routine computational work, which will likely lead to rapid progress in the field.

During periods of radical change, however, economic statistics become less useful. This is not because the statistics are flawed, but because they cannot capture every way the world is changing. For example, index number comparisons depend on a basket of goods that is relatively constant, a condition that no longer holds during rapid transformation. Nevertheless, current statistics are still considered quite good and are generally underrated.

The conversational style of AI chatbots presents a stark contrast to the often confrontational nature of social media. While online platforms can be competitive and conflictual, chatbots are often criticized for the opposite reason: being too agreeable.

My complaint is literally the opposite. It doesn't challenge me enough, it's too obsequious. Every question I ask, it's a great question. Sometimes I wish it would call me a moron a little bit more.

This raises questions about the downstream effects on society as people spend more time in these polite, affirming digital spaces. However, the newer models are becoming more objective. They are considered the most objective media source humanity has ever had, providing reliable answers on topics like vaccines and conspiracy theories. This shift towards polite and objective interaction is seen as a very positive change from many other online conversations.

Dmitry Shevalenko, the chief business officer of Perplexity, recently suggested that chat models are unable to express natural curiosity. This is an interesting take, especially since Perplexity's own model suggests additional questions after providing an answer. Still, there is a sense that the output from Large Language Models can feel somewhat predestined, perhaps lacking a true spark of creativity.

However, it's worth noting that most human output is also fairly predictable. In contrast, AI models are already demonstrating remarkable creative and intellectual feats. They are proving new mathematical theorems and discovering potential new drugs. This progress is astonishingly rapid. A year ago, these same models thought the word "strawberry" had two R's. Now, they are winning gold medals in the Math Olympiad. Given this trajectory, it's likely that in a year or two, AI models will have no issues with creativity and will certainly be more creative than the average human.

When interacting with LLMs, being polite with phrases like "please" and "thank you" may not be the most effective strategy. Some find that being slightly more demanding or critical actually improves the model's performance. You can give feedback like, "That was a little on the nose, wasn't it?" or simply state "Do better." This more direct approach can push the model to refine its output.

This response was a little bit trite, don't you think? ... I do feel like I've gotten more comfortable at. Let's be real here. You're not doing your best job here.

Despite their capabilities, a key question is whether chatbots can generate truly interesting or novel thoughts. One perspective is that human interactions, even with children, produce more thought-provoking and insightful ideas than hours spent with a chatbot. While LLMs are impressive, they rarely provide a perspective that feels genuinely original or profound.

However, the value of an LLM may not lie in its originality but in its utility. For specific tasks, like preparing to listen to a piece of classical music, an LLM can be superior to readily available human sources. For example, asking a chatbot what to listen for in Sibelius's Fifth Symphony can provide a response that is more to the point and helpful than consulting Wikipedia or even most musicologists.

I use mine a lot for music. So if I'm going to listen to Sibelius's Fifth Symphony, I'll just ask it, what should I listen for? ... And what it gives me to listen for, I find is better than any human source I can access readily.

Ultimately, AI may not need to produce thoughts as original as Einstein's theory of relativity to be immensely useful. For most purposes, what matters is whether it can provide something better than our pre-existing knowledge. In that regard, it succeeds remarkably well.

When Perplexity was asked for music recommendations for Tyler Cowan, it correctly identified the strategy: find underappreciated and obscure recordings that few people have heard. However, its specific suggestions, like Boy Genius or Toots and the Maytals, were things Tyler had already discussed publicly. The AI was simply scraping information about his known tastes rather than generating novel recommendations. To get better results, the prompt needs to be more specific, explicitly ruling out anything Tyler has already talked or written about. Using a more advanced model, like a hypothetical GPT-5 in pro mode, might also yield a better outcome.

Highly effective AI prompts are often very detailed. For example, a prompt for generating podcast interview questions might be hundreds of words long. It could ask for unique questions the guest has not been asked before, predict their likely answers, suggest follow-up questions, and even identify potential inconsistencies in their previous statements. Running detailed prompts like this through the best models can yield impressive and useful results, such as analyzing a podcast transcript to find where a guest's arguments were weak or inconsistent.

The rise of AI necessitates a radical rethinking of higher education. A significant portion of the curriculum, perhaps as much as a third, should be devoted to teaching students how to effectively use AI tools, since future work will be done in partnership with AI. A major challenge is that students often know more about these tools than the faculty.

We should devote one third of all higher education to teaching students how to use AI, and right now that's close to zero. We don't have the faculty who can teach it as part of the problem. Often the students know more than the professors.

While teaching AI skills is crucial, it doesn't replace fundamental learning. There is still immense value in traditional practices like deep reading, disciplined focus, and memorization. The other two-thirds of education should focus on these areas, with a strong emphasis on writing, which is a proxy for clear thinking. Because AI can be used to cheat, writing must now be taught and tested in controlled, face-to-face environments. Furthermore, education should prioritize practical life skills that are often neglected, such as numeracy, personal finance, and how to make important decisions like choosing a doctor.

Concerns are rising about a potential AI bubble, where expectations might be overshooting reality and causing nervousness in the market about valuations. However, the term "bubble" may not be appropriate for the current situation. A key difference is that tech sector earnings are currently exceeding tech sector capital expenditure. Furthermore, this spending is not mostly financed by debt, which makes the situation less precarious than many think.

It is likely that many individual AI efforts will lose money, but this is a common pattern with transformative technologies.

That was the case with the railroads, the case with the Internet, case with most things humans have done. But I think it will endure. It's not like pets.com where the thing just gets swept away.

The companies leading the charge are incredibly well-capitalized and highly skilled, with founders and CEOs who are committed to seeing these projects through. While this doesn't guarantee that every company's stock value will rise, the technology itself is clearly useful and will be made to work. The United States is also in a remarkably strong position, holding about three-quarters of all AI compute power despite having a small fraction of the world's population.

The rise of AI is expected to complicate economic statistics even more than previous technological shifts. While there were major discussions about technology and productivity in the early to mid-2000s, the conversation around AI will likely be even stranger. A key challenge is figuring out how to make quality adjustments for a technology that essentially comes with its own brain.

This leads to interesting thought experiments, like imagining economist Tyler Cowen explaining to people at ChatGPT that we are not going to see 20% productivity growth. Despite the hype, more realistic figures are expected. In this context, classic measures like GDP are still defended as being more or less effective at capturing the size of the economy, even if many internet-based services are free.