snordgren

Penalizing landlords doesn't solve the problem that there are M people who want homes, N homes in places with decent job markets, and M > N.

Better transportation, more efficient use of scarce land, and more home building is needed. Converting rentals to non-rentals won't add a meaningful number of homes to the market.

That's very likely. If you eat for €2k in Sweden you're not eating out at any time during the month. You can add at least another €2k to that if you want to go out to eat every workday lunch, which many high-income people do.

Speaking from personal experience, its mostly younger workers who eat yesterday's leftovers in the cafeteria.

I hadn't thought about that before, but that might be one contributing reason why online communities are moving into discord chats instead. A subreddit or forum has its most toxic posts visible until they are deleted. A discord chat almost moderates itself because anything older than a few days is a pain to retrieve, and doesn't show up on Google.

I believe it comes down to the statistical frequency in the dataset used to build the tokenizer. If the dataset is 90% Burmese the resulting tokenizer will have single tokens representing common complex constructs in Burmese instead of English. Today's tokenizer might tokenize Burmese letter-by-letter or even byte-by-byte as it is not a very common language on the internet.

Those 30 year mortgages also makes the economy less sensitive to interest rate raises, allowing you to run a higher interest rate than a comparable country where 30-year mortgages is not the norm. This makes the dollar stronger.

I think this post suffers from anthropomorphizing LLMs. Humans have values which inform our behavior, these values permeate all our communication and choices (to varying extents).

LLMs are trained to minimize loss on a dataset, and the previous tokens are used to inform which token is most likely to come next. The LLM will reflect the values of its training data, which may be wildly inconsistent.

We can use a system prompt like OpenAI to align LLM output with human values, by simply asking for it. By framing a query with an ethical system prompt we instruct the LLM to pull its output from a different part of its latent space.

This "constitutional AI" is just doing the same, but encoding the system prompt into the training data. As long as the training data concerning ethical behavior is more ethical than the rest of the training data, it is fair to assume that this method will be quite effective at aligning the model.

This is not perpetual motion, because the required information was within the model the whole time. It just wasn't the most likely output with regard to the original dataset.

Americans worry about the dollar going to zero, meanwhile it was recently at a 20-year high against the euro, the pound, and the yen.

Inflation has dipped to 5% and unemployment is lower than in decades.

For all its faults, American monetary policy has been better than that of most other developed countries. The ECB went 11 years without raising interest rates.

China definitely "leaks" a lot of talent to American companies - most AI papers that I've seen from respected Western universities include at least one Chinese name.

One challenge for China has been the university enrollment rate. While in Western countries half of each cohort has been going to university for decades, China is not there yet. In 2019, just 17% of Chinese adults have degrees compared to 44% in the US.

So the large Chinese population is offset by its relative lack of access to education, while the US can draw from its own highly educated population in addition to attracting the best and the brightest from the rest of the world, including China.

Algorithmic feeds generate more user engagement, that is what they are for. That's why platforms built around following updates from people that you like can't help themselves but implement them.

The USD is valuable because the US government demands its taxes be paid in USD. If you want to operate in the world's largest consumer market, you need dollars.

They are social media sites, the textbook example of the network effect.

Them not being very profitable is almost like a moat in itself - not only is it hard to displace them, it's probably not profitable to do so either.

Semantic search using embeddings seems like the missing puzzle piece here to me. We can already generate embeddings for both text and images.

The vision subsystem generates an embedding when it sees a cat, which the memory subsystem uses to query the database for the N nearest entries. They are all about cats. Then we feed all those database entries - summarized if necessary - along with the context of the conversation to the LLM.

Now your AI, too, gets a subconscious rush of impressions and memories when it sees a cat.

Unless most tech workers are 50+, they would not be saving the majority of that money, but rather investing it in the stock market.

Investment rates and savings rates go hand in hand, and taking money from the upper middle class to give to the investor class is not sound economic policy. Savings don't depress investments, on the contrary they reduce capital costs and allow for more investment without inflation. Banks don't actually keep your money on hand, they lend almost all of it out.

I don't know Keynes well enough to know if your take on his work is valid, but I do know that parts of his theories were found lacking 50 years ago in the last bout of stagflation.

Accidentally becoming too muscular simply does not happen unless you start injecting. Don't worry about it.

Road transport of goods accounts for a majority of wear on the roads. Rail is naturally more cumbersome because railroads are less ubiquitous than roads, but it is much more efficient in terms of both energy usage and labor.

If we subsidize road transport by letting tax payers or commuters pay for their road wear and tear, we end up with more road transport at the cost of more labor and more energy use. That's deadweight loss caused by the government distorting the market with road subsidies.

Road transport should be used where its flexibility gives it a competitive advantage, not be the default because taxpayers subsidize it.

One way to compensate lower-income earners for higher food prices is to lower taxes for those with low incomes, making the tax rate more progressive. Another is to reduce VAT for basic needs products like food.

These firms are in the business of finding and exploiting opportunities in the markets, before too many people know about them and the opportunity disappears. One person who knows too much could be an existential threat to the firm.

If 99% of global supply fell into the hands of the single country or corporation that brought it home, they would have the power to set prices. We would be looking at a De Beers-diamond situation.

There are already AI assistants for doctors. But the demand for healthcare is unlimited, so salaries will probably remain high and job markets tight.

The few times I have been to the doctor it has seemed like relatively standard knowledge work, gathering information and reasoning based on known and unknown factors.

The differentiating factors that make it high status and difficult are seemingly the same as for most knowledge work jobs: high stakes, high stress, high knowledge and skill requirements, limited pool of qualified people.

So unless I am mistaken, it seems like the kind of work that is very suited for automation. At the start it could be as simple as an AI automatically summarizing patient meetings.

And at the cost of the character of their neighborhood changing, their property valuation rose to the point that they can now live almost wherever they would like to. Countless people have to see their neighborhood change without the financial windfall they got.

According to that line of thinking, older generations should also stop receiving support from welfare and social security. Let them sell their houses and leave their communities if they can't afford to acquire the goods and services they need through consensual exchange.

And yet very few will accept elderly starving in the streets en masse because their pensions were gutted or went bankrupt. And so some measure of intergenerational agreement is necessary, where each generation provides for the next.

Your body builds the strength necessary to support your activity level. Unless you already have an active exercise regimen, an overweight person has generally only built the strength to support walking shorter distances. So the more overweight you are, the more careful you need to be when increasing distance while running.

Alternatives like cycling and swimming are better for the knees, especially for the old and overweight.

People will fight tooth and nail against new technology out of fear for cancer, but won't make the two adjustments that are most likely to save them from cancer: eat less processed food and exercise more. So evidently the risk of cancer is not the real reason for their fear.

Of course we need to cut down trees that are older than we are. The alternative to development is stagnation and poverty. It's medieval Europe where the social status of your parents decided your social status and that of your children.

Change is always hard, but the alternative is so much worse. There is no terra nullius that we can develop without anyone being bothered by it. Every house ever built does not have to be a museum to those who once lived there. We have too much of that thinking in Europe, where cultural conservation takes precedence over economic development, and as a consequence the economic wellbeing of the population suffers. It's great for tourists, but you can't live and do business in a city-sized museum.

Here in Europe peak electric prices have been far above $1/kWh, rising to around $7/kWh in the worst-hit regions.

Due to the build-out of wind power we have also had a few nights of negative electricity prices in recent months.

If we had a hydrogen energy storage facility, it could probably have recouped quite a portion of its capital costs this year, depending on its scale. Europe will not be building much base load power in the coming years, so the imbalance of the grid will only continue to rise, allowing for more business opportunities in the energy storage sector.

I've been wondering how much value could be generated if people could order their regular purchases months ahead of time. I am absolutely going to need to buy pasta three months from now. Why can't I purchase my pasta now and save the grocery store the effort of stocking the shelf just for me to pick it up? Long-term preordering could probably save labor and make demand forecasts more accurate. Just a thought.

People are definitely choosing cheaper alternatives where I live. Prices of pasta and ground meat have gone up 20-40%. Meanwhile prices for better cuts of meat, salmon and seafood are stagnant. I don't doubt that at least a part of this difference is because people are choosing cheaper foods.

As an example, if we use solar to generate the electricity to power the indoor farm, the panels will take up roughly 20-40% of the sun's energy, depending on the panel.

So to grow the crops indoors we need to use at least 2.5x as much land.

The sun is not a source of emissions, so why spend time and money replacing it?

Car companies were asleep at the wheel before Tesla. Selling EVs is perfect because they have lower OpEx at the cost of a higher CapEx. Higher CapEx means more revenue for the car manufacturers, and higher valuations in the long run if they can maintain their profit margin.

Car manufacturing will be a more valuable industry as EVs become the standard, but definitely not fast enough to quadruple the value of the industry in a handful of years. Not unless we start overvaluing every company like Tesla.

We already know that humans are human-level intelligent, so we don't need to devise a test to prove that they are.

If AI >= highest level of human intelligence, then AI > human intelligence. It is less accurate because it will miss any AI that achieves a precisely human level of intelligence, but it is probably also an easier test to define than others.

Of course simply increasing the scale of the model does not count as an improvement, there is nothing novel in that suggestion. But you are right that it opens a can of worms as to what an improvement is. Does an improvement count if it improves the model in the 80% of scenarios that only appear 20% of the time, while similarly reducing its performance in the 20% of scenarios that appear 80% of the time?

The number of highly disruptive studies remaining the same seems impressive enough to me. I expect diminishing returns to be the long-term norm in science. The low hanging fruit get picked first, and every breakthrough means one more concept that the next generation of researchers have to learn and incorporate into their work.