breput

My favorite von Neumann anecdote was a quote from Edward Teller: "von Neumann would carry on a conversation with my 3 year old son, and the two of them would talk as equals, and I sometimes wondered if he used the same principle when he talked to the rest of us."

It was too early, but also lingered in development for far too long. It also never had the proper server-side implementations, so it could feel like a solution in search of a problem.

That said, the design is far more advanced than the rolling disaster which is Passkey and it is much better suited for small Yubikey-type devices, where you could easily have unlimited site support. It also had intrinsic portability and advanced real-world security considerations, such as an attestation that instructed a server to disable weaker authentication methods, such as email or SMS (which is also a customer support disaster, but still...).

Ultimately, SQRL is an object lesson that the best technical design doesn't always win - it needs the right timing and robust community/corporate support.

Edit: Also, the (client) reference implementation was written in x86 assembly language for Windows. So I'd say the timing, support, and portability are all reasons for the lack of adoption.

I think you meant a quote attributed to Bill Gates:

"Well, Steve, I think there's more than one way of looking at it. I think it's more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it."

That's good information. I couldn't possibly even start to run even DeepSeek Flash on my system, but also if you're assuming multiple GPUs, that is going to affect the napkin math.

> We'll assume a 32B dense model, as they've have gotten quite good for production use and a B200 can comfortably serve them. This could be a Gemma, Qwen, DeepSeek, whatever.

That seems like a very consequential point to include halfway through the post. They aren't wrong that Qwen 3.6 26B or Gemma 4 31B are quite good, depending on the use case, but if we're doing napkin math, I'd want some more headroom in the assumptions.

They really ought to have Qwen parameterize their post's calculations and add sliders so a reader could play around with the values.

Edit: And since they especially mentioned DeekSeek (or whatever), as far as I know, none of their current generation of models is a dense model, and even the smallest of the mixture of experts (MoE) models is 284B parameters (13B activated). That will completely incinerate their napkin.

Interesting. It seems like a ToS violation would have been worthy of a warning and revoking the offending earnings, but nope, it was no mercy or review.

Several years ago, I ran a niche hobbyist website and incorporated Adsense (because why not?!?). The site featured a fuzzy search function since it referenced tens of thousands of named parts. The search result page would echo the (sanitized) search term followed by the matching results - along with recent search terms in the right sidebar.

One day, some spambot hit the site and started searching for terms like "mesothelioma". Adsense would see that page with "mesothelioma" in the sidebar, query for it, and served up the ambulance chaser's paid ads, even though there obviously were no matching results.

I didn't realize this was happening for several weeks since this low volume site was earning very little and I never even hit the minimum withdrawal limit. Suddenly I was earning $50 - $100 - per day. This lasted for a few weeks but before I could transfer the earnings, Google locked the AdSense account due to abuse. It might surprise you, but Google support was not helpful and after a series of reviews, they permanently shut down Adsense for this site.

Therefore, I also turned off Google Adsense for my websites.

As much as I respect the site and gladly financially support it, this is ultimately a failure on Ars Technica and its editors. If there are any.

If this were just some random blogger, then yes the blame is totally theirs. But this was published under the Ars Technica masthead and there should have been someone or something double checking the veracity of the contents.

That said, there are a number of Ars Technica contributors that are among the best in their fields: Eric Burger, Dan Goodin, Beth Mole, Stephen Clark, and Andrew Cunningham amongst many, so one f'up shouldn't really impugn the entire organization.

*Starting January 2027

It feels like GLM has either a bit of a fan club or maybe some paid supporters...