mg

Thinking about energy usage always puzzles me.

I'm not even sure if driving a car from one city to another really uses energy.

For example, let's look at it this way: The energy from the sun is heating earth. When we use some of it to drive a solar-powered electric car from one city to another, does that lower the temperature of planet earth? I would think it does not. So why did we get the energy to drive the car for free? I would think that the surprising answer is: Moving something from A to B does not use energy at all. But I'm not sure.

This reminds me of the discussions some years ago whether one should invest in an SSD when buying a new laptop or stay with a spinning disk.

Nowadays, barely any new laptops are released without an SSD.

When it comes to electric versus petrol, I often think about the fact that the sun sends more energy to earth in an hour than all of mankind uses in a year.

I guess that in theory, all that energy could be captured with parts that need no service at all and do not degrade? Of course, we will not practically reach that efficiency. But we will continue to improve towards this limit. When I think of what it takes to dig for oil and all the steps needed to turn that into a moving car, the way we are heading seems clear.

Is it the same route every time?

If not, they might take harder and harder routes as FSD gets better. Making it look like there is less progress than there really is.

What I would find really interesting: If someone drove the same route with every version of FSD and counted the interventions.

That would give a nice chart over time from which one could see the progress.

Is there anything online similar to this?

Another way to look at it would be to look at the revenue that is generated via FSD. Buyers probably do some due dilligence what FSD offers. If the number of buyers goes up, that could be an indicator that FSD is getting better. But a quick Google search did not bring up much data about this published by Tesla. In their earnings report, there is a chart showing the cumulative miles driven with FSD. Not sure if one can draw conclusions from that. It seems very smoothed out and only goes back to 2021.

Any suggestions for a fitness tracker (steps, sleep) which can be used without an Android/iOS app?

If it is possible to access the data from a Linux machine - that would be perfect.

Hmm... I would think the on-premise search for companies falls within the "less than 1% for everybody else" category I mentioned in regards to search.

The revenue of Elastic Search is less than 0.5% of Google's revenue.

Do LLM providers with smaller / less capable LLMs still have a USP or will it be like what we witnessed in the search space, where the #1 company has more than 90% market share, #2 has less than 5% market share and everyone else less than 1%?

Does Google rely on TSMC to build the TPU chips?

To test whether bard.google.com is already updated in your region, this prompt seems to work:

    Which version of Bard am I using?

Here in Europe (Germany), I get:

    The current version is Bard 2.0.3. It is
    powered by the Google AI PaLM 2 model

Considering that you have to log in to use Bard while Bing offers GPT-4 publicly and that Bard will be powered by Gemini Pro, which is not the version that they say beats GPT-4, it seems Microsoft and OpenAI are still leading the race towards the main prize: Replacing search+results with questions+answers.

I'm really curious to see the next SimilarWeb update for Bing and Google. Does anybody here already have access to the November numbers? I would expect we can already see some migration from Google to Bing because of Bing's inclusion of GPT-4 and Dall-E.

Searches for Bing went throught the roof when they started to offer these tools for free:

https://trends.google.de/trends/explore?date=today+5-y&q=bin...

And the average across all 2-3 year old cars is 5%.

It does not tall us much unless we would also get some information on how expensive it is to fix the problem.

A 5% problem rate can be worse than a 15% rate if the cars in the 5% category have issues that are more than 3x more expensive to fix.

When you click on the name, you get the "Pannenstatistik". Which sounds like it is how many cars insured by ADAC had a problem on the road.

The chart on the "TÜV-Report" seems to be what percentage of cars turned out to have a problem when routinely checked in a garage.

End of next year. So in theory I could 10x the number of data points.

But its one of the more annoying experiments to let a coin decide if you have to take another walk :) So I'm not sure how often I will do that. I'm pretty convinced already that walking more is a good thing.

I'm in a hurry right now, but I will probably write a full blog post about the steps-sleep relationship with p values and everything some time next year.

Is "take a measurement every morning" about sleep quality or about something else?

Is it even possible to sleep with the H10 strap?

Good point.

Is there a standardized way to measure sleep quality? Or are the numbers one measures only compareable as long as one uses the same device?

I would be interested in suggestions which watches measure the HRV / Sleep quality reliably and display the data on screen. Or let me get it from a Linux laptop. I hate having to install an app to get the info off the watch.

Maybe because of less physical activity?

Since 2020 I have manually tracked 35093 data points about my life [1]. And the one parameter that correlates most with my sleep time is how many steps I took during the day.

The median number of steps I take during a day is about 9500.

When I look at the sleep time of the following night, on average I slept an additional 9 minutes after a day with 9500 steps or more compared to a day with less than 9500 steps.

The gap increases to 16 minutes when I look at nights where I walked over 12000 steps in the previous day.

To see if that is just correlation or if there is causation at play, I also started to decide to take an additional walk or not by coin flip. So far, I did that 31 times. 16 times the coin decided to take a walk, 21 times it decided not to. On average, I slept 13 minutes longer after the coin decided that I take an additional walk. 31 is a very small N of course. I hope to have a larger sample by the end of next year.

1: https://www.gibney.org/a_syntax_for_self-tracking

Would it be possible to run Stable Diffusion in the browser via WebGPU?

One of my favorite Munger quotes is "The more you shout it out, the more you pound it in".

Meaning that if you talk with a strong opinion on a subject, then it is hard for you to change your mind by thinking it through further and coming to the conclusion that you were wrong.

I guess one could turn this into a recommendation like "Speak softly and stay flexible".

It's interesting to note that Munger himself often didn't follow this advice and was quite bold in expressing his thoughts.

That could be interesting. A recursive cat, so to say.

The problem would be this: In the picture at hand, the big cat is rather simple. Just a portrait of a smiling cat. While the 9 smaller cats are doing all kinds of poses to adjust to the form of the big cat portrait. So the subcats are more complex than the main cat.

When doing the recursive cat, it would be hard to make a subcat from 9 subsubcats because the subcat is already a complex image that is not as easy to recognise as the main cat.

I had a similar idea early last year and also dabbled with a checkerboard approach.

Here a cat is made from 9 paintings of cats in the style of popular painters:

https://twitter.com/marekgibney/status/1521500594577584141

You might have to squint your eyes to see it.

I made a few of them and then somehow lost interest.

Have you seen PWAs?

https://www.google.com/search?q=pwa

Websites already have access to offline storage via the localStorage API.

When the user installs a PWA, it gets additional permissions which make it even more similar to a native app.

Have you considered making it a website? Then it can be used on every device via the browser.

I'm running this comparison of free and open AI engines:

https://www.gnod.com/search/ai

Looks like they all currently work.

If there are more, let me know.

Surprisingly, it looks like Bluesky is outperforming Mastodon:

https://trends.google.com/trends/explore?date=today+5-y&q=bl...

https://www.similarweb.com/website/bsky.app/#traffic

If this holds true, it is a big win for interface over functionality and under-the-hood-technology.

Mastodon is based on a pretty good protocol (ActivityPub) and has many instances. But the usability is terrible. Bluesky on the other hand only has a single implementation and instance, and I find the protocol (The AT protocol) harder to grok. But the interface of the one instance is way better than any Mastodon instance I have seen.

No. I did not mention profit at all. Only revenue.

To me it seems that every task done by humans will be supported or even replaced by AI. I already see it happening, just 2 years into the new LLM area: Programming, designing, writing, research, driving. How could that not result in a doubling in software use?

I did not assume that one player is going to capture all value. I stated the number for 10% and 90% market share. Today, AWS has 30% market share of cloud computing and Google has 90% market share of search. Assuming the market leader in AI will achieve between 10 and 90 percent market share seems reasonable to me.

You don't expect the number of use cases for software to double with AI?

What are yours?

In the US, software is 10% of GDP. And with growing usefulness, that should increase.

The world's GDP is about a hundred trillion dollars. Growing a about 5% per year. If that will continue, it will be 250 trillion in 20 years.

In 20 years, what part of it will be AI revenue? 10%? 20%? 30%?

At 20% that would be a 50 trillion dollar market.

If Microsoft manages to capture 10% of that, that is $5T in revenue per year.

Of course, Google somehow managed to capture 90% of the search market. If Microsoft manages that, it would be $45T yearly revenue.

I remember my first experiences with touch screen phones. It was magic, and it was clear that this would change the world. Touch screen phones were the biggest and final "dent in the universe" Jobs made.

But it was not as magic and promising as my first experiences with AI that I had over the last 24 months or so.

AI is a much bigger step for mankind than home computers and smartphones. So Altman is involved in making much bigger dents in the universe than Jobs was.

It's interesting, that Google keeps publishing AI research papers. Is there a business rationale behind it?

OpenAI has become one of the fastest growing companies of all time. And much of it is based on Google's "Attention is all you need" and other papers.

Since Microsoft added the Dall-E 3 image creator to Bing, Bing saw a huge inflow of new users. Dall-E is also a technology rooted in Google papers.

I wonder how Google thinks about this internally.