tmnvdb

The openAi version costs 200$ and takes a lot longer, not sure if it is fair to compare?

For R1-Zero they did RL on two properties:

a) there is a 'thinking' box and an 'answer box' (format constraint)

b) the answer box has a correct answer

Note that there is nothing about the contents of the thinking box in the reinforcement learning. Only that there is such a box.

They then observe that during RL, the model will start generating more and more stuff in the thinking box.

In essence this is the emergence of using more test-time compute to improve answers.

When reading out the thinking box, they found the model reflecting on its own answers, going back and changing its mind, and reflecting on the question, etc, similar to what we can see with R1 now. These are fully emergent phenomenon, without any prompting to do such a thing!

(The reasoning output would sometimes be a bit garbled and switch languages randomly, so for R1 they added some constraints to make the thinking content intelligble to humans. This actually made the model slightly worse at answering questions correctly.)

The study on which the article is based seems somewhat speculative.

The assumption for the full renewable scenario are the existence of the hydrogen economy. I'm quite positive about the potential of hydrogen but there are quite a few unsolved problems at the moment and it seems the hydrogen part of the energy transition has slowed down a bit.

Certainly the total cost of such a system is not well known at this time. So the cost calculation for the renewable scenario is quite uncertain - other energy storage tech might be more expensive.

The carbon capture calculation is based on the assumption that there is no other renewables and we go 100% capture.

This is not really that interesting an scenario, what matters more is the marginal costs of each technology at different points in the future.

In general a healthy dose of scepticism is warented when it comes to long term projections or cost of technologies, though or course policy does require we take a stab at it.

Fair enough, I guess as with all tools YMMV and things depend on the kind of work. I also have days where I have no use for them.

GPS still works when power fails.

Like it or not, you live in an industrial society that depends on a million things you can not possibly recplicate or repair if they fail.

Or you could learn to use an LLM for what is it good at so that it actually saves you time (arguing with an LLM for a long time a sure sign of not having a lot of experience with this tool..)

This example is from 1 year ago. With current LLM progress that means it's basically a historical document.

Both 4o with internet and o3 without internet gave the right answer.

Somebody in the comments did a deep research run which produces an extremely detailed History of the naming.

Of course reliability is not a solved problem in general but people here would do well to check both blog posts and generative model output for factuality and relevance.

Can you give an example of when that happened?

You are partly correct to note that local markets are often not economical, but the reasons for that are precisely low density, car dependence and restrictive zoning.

For a local market to be economical it needs a catchment area containing enough customers.

Once customers start driving they prefer to drive a bit longer to go to a mega market or mall with good parking options, i.e. not a local market.

Local markets thus have a catchment area that is defined mostly by walking and biking.

(Not having a lot of parking can also help to make local markets more economical)

Some important conditions for local markets are then:

1. Zoning needs to allow the shop to be very close to residential areas, and with only few parking spaces.

2. Residential density needs to be high enough to have a large number of customers within walking or biking distance.

3. Walking and biking should be sufficiently safe and pleasant.

It is easy to see the why there are few local markets in suburban North America, but they are very common in urban places all over the world, as well as in thoughtfully designed and slightly denser suburbs in many countries.

That is not obvious to me. Despite the many flaws I see in Trump, rolling over does not seem to be his first instinct. And European leaders surely are not above playing his ego either.

I understand the criticism, Trump has essentially given away some core issues before negotiation starts, usually not what you want to do. However, a more favorable interpretation is that all knew these things were not really on the table, and it might help to build trust with Putin. My feeling is that it is hard to tell how this will play out for now..

Observation: just saturating all the benchmarks is now a lower bound for some.

Those things are not the same though. Reproducible just means it will break again if you configure your system in the same way.

Fantasy is precisely the right word.

France and the UK have nukes precisely because they did not believe that a country would be willing to risk total destruction to protect another country.

We should probably consider the policies and weigh their pros and cons.

Those are fair points but I think there are some important details that are missing.

1. There is no European VC market. This is important. When people talk about EU VC funding they are just adding together national numbers. In reality EU countries have seperate capital markets with different rules serving seperate national markets - when it comes to services there is no single market. This means none of the European markets come close to having the scale of the Chinese or American ecosystems.

Even though the amount of VC funding of the EU as a whole may seem to be of similar order of magnitude, being around 1/3rd of US VC (although as I wrote above, this is a virtual number) there are several important differences. About two-thirds of European funding is provided by governments or (risk-averse) banks, whereas about two-thirds of US finding is from private sources that are comfortable with higher risks and long periods of unprofitable money shoveling (see Amazon, Uber, Tesla).

Furthermore, there is a large difference in the size of funding (about 5x bigger in the US per company) and the amount of it being for late stage growth (very little in the EU). These are big barriers to creating 'big winners' in winner-takes-all markets.

You raise a valid question about the direction of causality here. It could be that European companies are just inefficient because they are overburdened by regulation, and investors know this. I think that is probably partly the case, and definitely in some sectors.

However, there is some evidence that regulatory burden and productivity are currently not the limiting factor. First of all there is quite a lot of diversity in regulatory burden and productivity between countries in Europe. But regularitory burden and productivity does not seem to correlate much with the size of the startup markets in Europe. The biggest startup markets are found in the capitals of the countries with the biggest markets (Germany, France and the UK). These countries are not known for being particularly business friendly or efficiënt, but they are the biggest countries. It seems likely that network effects and market scale are dominant forces here.

Indeed, it seems that the higher cost and complexity of having to scale internationally across different regulatory domains are a bigger problem for potential European big winners. A company that wins scaling in the US dominates the biggest market in the world and is likely to be the global winner. National big winners in Europe do not have the size to compete, unlike those in China. And establishing Europe-wide winners is difficult.

(In fact, it is mainly European business that push for moving regulation from the national to the European level, precisely to make Europe-wide activity easier)

One more piece of evidence is that the US VC ecosystem is centred in SF, California, probably the place in the US most burdened by regulation. They even suffer from a GDPR-light.

2. The big correlate for profits is size and Europe has very few very large companies that are of recent vintage. However, those big companies that do exist have similar profits to large US companies. If you think there is some hard limit I'm curious to hear what you think is the mechanism.

3. Maybe, but this kind of funding always comes with the art of writing the right words, doing the right compliance, and sometimes knowing the right people and having a big and well known name. But maybe I was a bit too black-and-white.

Software safety is an abstract concept. You have experience with specific instances of problems that fall under the abstract concept of software security. This is not to say it is not important.

Not being able to move your hands is not an abstract concept. It can be directly experienced.

The big difference is that this is a top-down gov run iniative, it will by nature be very different than a free market iniative.

I disagree, I did an EU subsidy request when I was 18, it was quite simple and well organized.

German bureaucracy is terrible but that is not the same as the topic here.

While I understand this sentiment, we do need to balance the short term and the long term. And in the long term, economic growth is compound growth.

Consider an example. Country 1 grows at 2.5% (like the US) Country 2 grows at 1.5% (like the EU)

What is the difference after 100 years?

The first country will have grown by 1200%. The second country by 440%.

The faster growing country is now almost 3x richer than the second country.

This is comparable to the difference between Romania and Austria currently.

As these growth effects compound, I believe that it will become hard to compete on quality of life long term.

Your analysis points out some real problems, but I think they are more systemic rather than just cultural. (This is not to say that the cultural issues are not real)

1. Startups mostly leave because Europe doesn not have a private capital ecosystem and hence entrepreneurs can not get sufficient funding for growth.

2. There is some truth to this statement in the sense that some people have this mentality but is not true in general. There are profitable companies in Europe and there are jurisdictions (Denmark, Ireland, Estonia, etc) where it is reasonably attractive to have a company. The data does not show that this makes a difference: the biggest startup markets are still in the larger countries that have a larger (capital) market, even though these countries are usually more hostile to business.

3. Yes, that is the case with all government subsidies, however often their founding documents claim to wish to support small and new things.

This does not seem like a very good basis for rational policy.

A lot smaller. The city of Amsterdam employs more bureaucracts than the EU.

There is no European (capital) market and the national markets are too small to compete with the US or China. European companies are therefore unlikely to gain trust from investors that they will be able to compete.

Hence these kinds of top-down iniatives are in a sense the best they can do given rather poor ircumstances for private investment.

It is very brave of you to make this totally hypothetical sacrifce for us - only in your mind of course.

You seem awfully sure about what you would do but you again do not seem to have considered about what it is like to be fully paralyzed. Let me be clear, I understand the risks of these devices. But my impression is that you're having trouble emphathizing with people with such medical conditions and you're not really considering how it feels to live like that. In fact, you again spend your whole message talking about abstract considerations, but you do not talk about the experience of being unable to do almost anything - how that shapes a persons willingness to take risks and weigh them differently. That is my point about trade-offs, consider the personal and emotional as well as the technical.

You have again ignored the trade-off in your word-play.