NickHoff

When I use these models in codex, there are two axes for me to control - the model and the reasoning level. I can use Astra, Sol, or Luna. And I can choose between 5 reasoning levels (light, medium, high, extra high, and ultra). What's the difference? As the problems that I want codex to solve get easier, should I turn down the model or the reasoning level? What's the difference between Astra medium and Luna high? So far I've just been leaving it on Astra and then turning the reasoning level up or down based on how hard I think the problem is.

I enjoyed some of the incorrect options. For "Debilitate" one of the options was "Remove a bill from the tab".

Look neat. FYI clicking "See All Connectors" is a 404.

A UUID v4 is a 128 bit number, but 4 bits are reserved to specify the version number and 2 more bits are reserved to specify the variant, which leaves 122 bits of randomness. That means it can take on 5 x 10^36 possible values. Following the birthday math, you'd have to generate about 103 trillion UUIDs to have a one-in-a-million chance of having a collision.

Neat. What about power density?

An H100 has a TDP of 700 watts (for the SXM5 version). With a die size of 814 mm^2 that's 0.86 W/mm^2. If the cerebras chip has the same power density, that means a cerebras TDP of 37.8 kW.

That's a lot. Let's say you cover the whole die area of the chip with water 1 cm deep. How long would it take to boil the water starting from room temperature (20 degrees C)?

amount of water = (die area of 46225 mm^2) * (1 cm deep) * (density of water) = 462 grams

energy needed = (specific heat of water) * (80 kelvin difference) * (462 grams) = 154 kJ

time = 154 kJ / 39.8 kW = 3.9 seconds

This thing will boil (!) a centimeter of water in 4 seconds. A typical consumer water cooler radiator would reduce the temperature of the coolant water by only 10-15 C relative to ambient, and wouldn't like it (I presume) if you pass in boiling water. To use water cooling you'd need some extreme flow rate and a big rack of radiators, right? I don't really know. I'm not even sure if that would work. How do you cool a chip at this power density?

I'm an American living in Germany. When I first arrived, the way Germans write the digit 1 surprised me. They write it with the upper hook thing very long, almost like a capital lambda (Λ), which sometimes makes 1 and A visually ambiguous. This isn't really a problem, just something funny about moving to a new country.

> Superb Owl

That's a great typo.

ASCII arrow: own work.

LOL

I clicked their link to Knative so I could read about it. On the Knative webpage, the cookie banner pops up. I don't want to accept so I click "learn more". That expands the cookie banner to include a button that says "I understand how to opt out, hide this notice" as well as a link to a lengthy explanation of cookies. Well I don't want to click the button so I click the link to the cookie explanation. At the bottom of that page are more links to browser documentation that might eventually explain how to opt out, but I can't click those links because everything on the page is disabled - because the same cookie popup is on this page too. It blocks interaction, including clicking on their links about how to opt out, until you opt in. This stuff is getting worse.

There's a recent Tom Scott video about an office inside an elevator. It has a desk and chairs, the whole thing. It's on the corner of the building too. It seems, from his video, that nobody is really sure why it was built and what it was used for.

https://www.youtube.com/watch?v=1yfX84RMQ3M

My question here is about underlying fab capacity. This chip is made on TSMC 4N, along with the H100 and 40xx series consumer GPUs. I assume Nvidia has purchased their entire production capacity. I also assume that Nvidia is using that capacity to produce the products with the highest margins, which probably means the H100 and this new GH200. So when they release this new chip, does it mean effectively fewer H100s and 4090s? Or is that not how fabrication capacity works?

I'm asking because whenever I look at ML training in the cloud, I never see any availability - either for this architecture or the A100s. AWS and GCP have quotas set to 0, lambda labs is usually sold out, paperspace has no capacity, etc. What we need isn't faster or bigger GPUs, it's _more_ GPUs.

See also the International Obfuscated C Code Contest. [0]

This program [1], for example. It just accepts some input on stdin and returns the same input, but mirrored along the diagonal. So the first row of the input becomes the first column of the output, second row becomes second column, etc. But the program is functionally invariant when given itself as input. In other words, you can flip the source code of the program along its diagonal and the result is a program which has the same functionality - it flips stdin.

Or this one [2], which parodies java in C. It's functional C code that looks like Java, including the

  class LameJavaApp
  {
    /** The infamous Long-Winded Signature From Hell. */
    public static void main(String[] args)
    ...

Or this one [3] that calculates pi by estimating its own surface area.

Or this one [4]. It's a lovers quarrel, written simultaneously in C and English. It's incredible, seriously, read it.

[0]: https://www.ioccc.org/

[1]: https://github.com/ioccc-src/winner/blob/master/1994/schnitz...

[2]: https://github.com/ioccc-src/winner/blob/master/2005/chia/ch...

[3]: https://github.com/ioccc-src/winner/blob/master/1988/westley...

[4]: https://github.com/ioccc-src/winner/blob/master/1990/westley...

There's also this one. https://www.youtube.com/watch?v=F3FkAUbetWU

He shoots the Prince Rupert's drop with a bullet. The bullet wins. In high speed, you can see the bullet splat against the glass and break into shards, long before the glass breaks. Really neat.

I've read about this. Studying the psychology of group dynamics and conflict in a confined environment was a part of the experiment. Sure enough, the crew of 8 quickly split into two factions. It still seems like a success to me though, because even though some of them were barely on speaking terms despite being friends before, they continued to work together.

> I don't like some of them, but we were a hell of a team. That was the nature of the factionalism... but despite that, we ran the damn thing and we cooperated totally.

So, what do people think of Vivaldi? I'm a long-time Firefox user but I've been scanning for a new browser for a while now. Even if it weren't for stuff like this, I'll have to change anyway when Firefox dips below ~3% and websites stop supporting Gecko.

I hope that Brother doesn't succumb to the same forces that drove HP (and everyone else) to these decisions. In the short term it makes sense for HP because

1. At first the printer works well. This causes reviewers, who only use the printer a few times when it's new, to write a good review.

2. It lets them keep the up-front price of their printers low. Most people only consider up-front cost and not total cost of ownership. It also helps with getting a good review.

3. They get some recurring revenue from customers who don't want to deal non-name-brand ink. They just want their printer to work and they're willing to pay.

Brother, on the other hand, is slightly more expensive but is better built and works with off-brand ink. We have a brother laser printer and it's great. The ink doesn't dry out like an inkjet, we can buy ink from wherever, and it doesn't break down every few months.

I just hope that brother understands that going HP-mode would probably give a short-term bump but hurt their name in the long run.

I took a linear algebra class in high school and we used two textbooks for the class. One gave a simple and soft presentation which we used most of the time. The other one was much more rigorous, referred to in hushed and fearful tones simply as "Strang". When I got to university, I realized with fear and excitement that the guy teaching the linear algebra class is the same guy who wrote the textbook we were all afraid of in high school. His lectures are excellent. Just as clear and rigorous as the book, but not at all scary. Just like Strang himself.

> retail GPUs are much more powerful

I agree with your basic point about building vs renting, but retail GPUs aren't simply more powerful. Retail GPUs are just as fast but have a lot less VRAM and memory bandwidth. The RTX4090 has 24 GB of VRAM with 1 TB/s of bandwidth. The A100 has 80 GB at 2 TB/s. For some tasks, the memory and bandwidth are the strongest constraints.

One of the issues is availability. Lambdalabs and Paperspace never have instances available. AWS and GCP have the per-account quota set to 0 by default for GPU instances and deny requests to raise it, even to 1. (One of my AWS accounts was granted 1 GPU after some back-and-forth with support.) I'm not upset, I understand that demand is high, but how will oblivus handle availability?

AI is moving too fast to regulate. A lot has happened in the field just during the time that these "last minute amendments" have been discussed. It's more likely that the EU will end up with laws that are obsolete by the time they're implemented, take forever to revise or repeal, and just sit there constraining innovation for no good reason. For example, if an EU-based startup wants to build a generative AI system to make interior design renders to show off furniture for a magazine (or whatever), how much time and legal expense will the "transparency and risk assessment requirement" add?

To be fair some of this sounds like a reasonable idea, like prohibiting "remote biometric identification systems in publicly accessible spaces". The issue is that this law would only prohibit using AI to do that. Let's outlaw the things we really don't want (like algorithmic voter influence) in a technology-agnostic way and then let AI flourish.

I spent a while trying to find the 'but' in that sentence. I assume you meant 'not'. :)

> City Hall said on its website about 103,000 people voted...

The population of Paris is about 2.1 million according to Wikipedia, meaning ~5% turnout.

Some of the permissions this app requires:

    read your contacts
    modify your contacts
    access approximate location only in the foreground
    send and view SMS messages
    receive text messages (SMS)
    receive text messages (MMS)
    read your text messages (SMS or MMS)
    take pictures and videos

I'm aware of these laws but I've always wondered - what if a German startup really does want to operate in a new and "unexpected" way, but doesn't want to trick their customers? They put the unexpected stuff clearly on their website, not buried in the ToS. I don't see any ethical problem with that - people can make an informed choice about whether they want to do business with the company. BUT - it would still have to be in the official ToS and now that unusual stuff would be void.

Doesn't this law prevent startups from developing innovative business models even when they are being upfront about it?

> What beats Steam will not be a better Steam. It will be something else...

Right. It will probably still have to be better than Steam though.

Like a game streaming service in which all the games are available to play. Pay $20 per month, play any game whenever you want. The economics might even work out. They could pay each publisher $x for each hour that each person plays their game. More money for small studios since more people will be trying their game for a few hours, still lots of money for the big guys.

From the players perspective it would be like buying a AAA game every 3 months. Expensive, but you get all the other games too and you don't need an expensive gaming rig.

Or another idea - no monthly subscription fee, but you pay $X/hr for each game. Big games cost $0.10 per hour, free-to-play costs $0.005 per hour. Make the numbers work out so that publishers get the same overall revenue, but players like the flexibility. You could even limit your monthly spend. Cloud pricing for games?

Bitwarden already does one thing well. It's everything I'm looking for - open source, costs money but not much ($10/yr), 2FA, clean interface. I'm happy for the new investment, but I hope they don't start adding new things just for the sake of growing.

Also - to the people who analyze funding rounds - $100M sounds like a huge amount to me. Why would a password manager need so much money?

In this [1] blog post, they say it's "between 2 and 20 times slower than MySQL (but committed to getting down to 4x slower)".

[1]: https://www.dolthub.com/blog/2021-03-09-dolt-use-cases-in-th...

Syncthing is great, I've used it for years to sync all my stuff and it has always been solid. I even use it on my kindle fire tablet (there's a Syncthing app in F-droid).

My only annoyance with Syncthing is when I reinstall an OS on one of my machines. Let's say I'm syncing files between my main workstation, two laptops, a phone, and a cloud backup. Now I want to reinstall the OS on one of the laptops. When I install syncthing on that machine, it gets a new ID. I can make it join the sync swarm but all the other members will think it's a new machine and won't trust it so I have to go to each of the other machines and manually remove the "old" laptop and bring in the "new" one.

This guy also made a similarly acerbic video about MuseScore, a good open source scorewriter [1]. They actually took his criticisms and started making changes. The two ended up working together and I think he is now the design lead for MuseScore.

1: https://www.youtube.com/watch?v=4hZxo96x48A

I've always liked 1/7 as well, and I never realized that thing about doubling from 7 giving 14, 28, 56. People are always impressed when I can rattle off the digits of x/7.

By the way, I appreciate your use of U+0305 combining overline. Did you enter those manually or do you have some neat way of doing it?