He didn't win it. It was won by a team of students / collaborators / mentees, who felt he deserved it. I can't disagree with them. Among the nicest people in the world.

I don't think anyone meant it in the sense of "You're a Nobel Prize Winner," so much as "We couldn't have done this without your mentorship, and you deserve to hold onto this." He certainly doesn't consider himself to be a Nobel Prize winner.

- His breaking up images into grids was a poor-man's convolution. Render each letter. Render the image. Dot product.

- His "contrast" setting didn't really work. It was meant to emulate a sharpen filter. Convolve with a kernel appropriate for letter size. He operated over the wrong dimensions (intensity, rather than X-Y)

- Dithering should be done with something like Floyd-Steinberg: You spill over errors to adjacent pixels.

Most of these problems have solutions, and in some cases, optimal ones. They were reinvented, perhaps cleverly, but not as well as those standard solutions.

Bonus:

- Handle above as a global optimization problem. Possible with 2026-era CPUs (and even more-so, GPUs).

- Unicode :)

It was actually a bit ironic that, at the time, you could burn through the whole high-speed quota in seconds or minutes, if you went to the wrong web page. Most carriers would stop or bill you an arm-and-a-leg after.

Free internet existed before paid internet, true, but mostly because people did things for other motives (like fun). Altavista was a tech demo for DEC. Good information was found on personal web pages, most often on .edu sites.

Banner ads existed, but they were confined to the sketchy corners of the Internet. Thing today's spam selling viagra. Anyone credible didn't want to be associated with them.

What Google figured out was:

1) Design. Discrete ad-words didn't make them look sketchy. This discovery came up by accident, but that's a longer story.

2) Targeting. Search terms let them know what to ads to show.

I can't overstate the impact of #2. Profits went up many-fold over prior ad models. This was Google's great -- and ultra-secret -- discovery. For many years, they were making $$$, while cultivating a public image of (probably) bleeding $$$ or (at best) making $. People were doing math on how much revenue Google was getting based on traditional web advertising models, while Google knew precisely what you were shopping for.

By the time people found out how much money Google's ad model was making, they had market lock-in.

To kick-start, given them machines with Plan9, ITS, or an OS based on LISP / Smalltalk / similar? Or just microcontrollers? Or replicate 1970-era university computing infrastructure (where everything was homebrew?)

Build out coursework to bootstrap from there? Perhaps scholarships for kids from the developing world?

Software is bloated in part because it's built in layers. People wrap things over, and over, and over. Stripping down layers is neigh-impossible later. Starting from scratch is easy.

Starting from scratch fails in practice because you don't get feature parity in time short enough for VC (or grant) funding cycles.

If we build a tech tree around 200MHz 32MB machines, except for things like ML and video, we'd have a tech tree which did everything existing machines do, only 10x more quickly in 0.1% of the memory. Machines back then were fine for word processing, spreadsheets, all the web apps I use on a daily basis (not as web apps), etc.

Need would drive people to rebuild those, but with a few less layers.

Parts of Africa and India are very much like that. I would guess other places too. I'd pick a hill station in India, or maybe some place higher up in sub-Saharan Africa (above the insects)

> What problem are we trying to solve that is not possible right now?

The point is more about identifying the problem, actually. An independent tech tree will have vastly different capabilities and limitations than the existing one.

Continuing the thought experiment -- to be much more abstract now -- if we placed an independent colony of humans on Venus 150 years ago, it's likely computing would be very different. If the transistor weren't invented, we might have optical, mechanical, or fluidic computation, or perhaps some extended version of vacuum tubes. Everything would be different.

Sharing technology back-and-forth a century later would be amazing.

Even when universities were more isolated, something like 1995-era MIT computing infrastructure was largely homebrew, with fascinating social dynamics around things like Zephyr, interesting distributed file systems (AFS), etc. The X Window System came out of it too, more-or-less, which in turn allowed for various types of work with remote access unlike those we have with the cloud.

And there were tech trees build around Lisp-based computers / operating systems, SmallTalk, and systems where literally everything was modifiable.

More conservatively, even the interacting Chinese and non-Chinese tech trees are somewhat different (WeChat, Alipay, etc. versus WhatsApp, Venmo, etc.)

You can't predict the future, and having two independent futures seems like a great way to have progress.

Plus, it prevents a monoculture. Perhaps that's the problem I'm trying to solve.

> Do we start from hardware at the CPU ?

For the actual thought experiment, too expensive. I'd probably offer monitors, keyboards, mice, and some kind of relatively simple, documented microcontroller to drive those. As well as things like ADCs, DACs, and similar.

Zero software, except what's needed to bootstrap.

"To make something really different, and not get drawn into the gravity well of existing solutions, you practically need an isolated monastic order of computer engineers."

As a thought experiment:

* Pick a place where cost-of-living is $200/month

* Set up a village which is very livable. Fresh air. Healthy food. Good schools. More-or-less for the cost that someone rich can sponsor without too much sweat.

* Drop a load of computers with little to no software, and little to no internet

* Try reinventing the computing universe from scratch.

Patience is the key. It'd take decades.

The cog-in-a-machine corporate culture is not fun. Tech culture is much healthier.

There's no upside to big electronics companies here.

The spend prices most of the developing world out -- an programmer earning $10k per year can't pay for a $200/month Claude Max subscription..

And it does better than $6k-$10k programmers in Africa, India, and Asia.

It's the mainframe era all over again, where access to computing is gated by $$$.

That's all noise.

Progress has been exponential in the generic. We made approximately the same progress in the past 100 years as the prior 1000 as the prior 30,000, as the prior million, and so on, all the way back to multicellular life evolving over 2 billion years or so.

There's a question of the exponent, though. Living through that exponential growth circa 50AD felt at best linear, if not flat.

$1-2T with no legal risk.

$300B assuming a rational and uncorrupt government, which should, at some point, kick them back to non-profit status, and convict people for fraud

Of course, too-big-to-fail means this won't happen.

https://duckduckgo.com/?t=ffab&q=gpu+lifespan&ia=web

You'll need to scroll past the ones talking about obsolescence versus failure. Toss in 'data center' if you like.

You'll see a range of numbers -- including some lower than I cited -- but it's all in that ballpark.

This seems improbable.

50-year-old technology works because 50 years ago, transistors were micron-scale.

Nanometer-scale nodes wear out much more quickly. Modern GPUs have a rated lifespan in the 3-7 year range, depending on usage.

One of my concerns is we're reaching a point where the loss of a fab due to a crisis -- war, natural disaster, etc. -- may cause systemic collapse. You can plot lifespan of chips versus time to bring a new fab online. Those lines are just around the crossing point; modern electronics would start to fail before we could produce more.

The denominator varies wildly based on whether or not the PR is made. If codex makes nonsense, I don't ask it to make a PR.

Poor data: If I make one, I either if I want to:

a) Merge it (success)

b) Modify it (sometimes success, sometimes not). In one case, Codex made the wrong changes in all the right places, but it was still easier to work from that by hand.

c) Pick ideas from it (partial success)

So simple merge rates don't say much.

https://lunary.ai/blog/openai-developer-role

There is a 3-level hierarchy:

System prompt > Developer prompt > User chat

You provide that middle level.