AI appears to have peaked

Ludic Mataora recently blew up with a piece on AI. It's not exactly clean speech, but the vulgarity is entertaining. But there is one part of his rant that I will expound on:

When I was out on holiday in Fiji, at the last resort breakfast, a waitress brought me a form which asked me if I'd like to sign up for a membership. It was totally free and would come with free stuff. Everyone in the restaurant is signing immediately. I glance over the terms of service, and it reserves the right to use any data I give them to train AI models, and that they reserved the right to share those models with an unspecified number of companies in their conglomerate. [1]

The AI companies are now scouring hotels in Fiji for training data. This is a problem, because AI needs data. There is no such thing as 'generative' AI. Just like machine learning before it, the AI is all about refactoring and collating existing data, created by something or someone else. A few years ago, I was involved with using machine learning algorithms to understand the language of DNA. We had the same thing, where we needed clean readouts of a lot of DNA to feed the model. Any algorithm is only as good as the quality and quantity of the input data.

And the AIs are out of data. At the very least, they are out of human data, which is why they are getting creative. Allegedly, they occasionally fail to ask the owners of said data. This assumes that their product contains actual AI algorithms to begin with. [2] [3]

Now, this does not mean that new features will not hit the markets here and there. We will certainly a fair few AI-integrated products hit the shelves for a while, as every industry figures out if AI can help them move product. We may also see some technological leaps in a few years. But these things take time, and the breathless interest of the masses seldom contributes more than money. After all, the current AIs started their groundwork in 1953. [4]

As an aside, this means that it is a good time investment to figure out if and how to use the current level of AI tech, because that level will not be surpassed anytime soon. Much Love!

[1] https://ludic.mataroa.blog/blog/i-will-fucking-piledrive-you-if-you-mention-ai-again/

[2] https://apnews.com/article/openai-chatgpt-scarlett-johansson-voice-her-532c849ccae3ca9e9325dacfe88e0436

[3] https://www.youtube.com/watch?v=zLvFc_24vSM

[4] https://infogalactic.com/info/Machine_learning#Overview

The Two-Platform Rule

The Kurgan recently wrote a post about the Substack platform [1], in which he points out one issue that many communications platforms have. And it's not just communications platforms, it's also an issue with things like cars, computers, even the power grid. The problem is that these things can get taken down/out without their users, like you or me, having any say in the matter.

These outages and crashes do not even have to be the result of enemy action. Sheer incompetence is often enough to make things really fragile [2]. Most systems are neither stable nor secure, and they are getting worse. The two-platform rule exists for things that you cannot go without for a while. Here is said rule:

If you need something, get it twice. 

For example, I have two computers and two phones. If one of either fails, I can use the other one to try and figure out a fix. If the fix takes a while, I have most of the functionality on the backup. For financial reasons, none of these four devices are state-of-the-art, but they all get the job done. Most importantly, all the key functions are covered redundantly.

Back to the Kurgan's example: if you need access to a specific content creator / geographically distant individual, connect with them on two platforms, preferably more. This way, any cancellation or just plain server outage cannot sever the lines of communication.

All of this applies twice as much to actual necessities, such as food, energy, and defense. It is better to have two mediocre guns than a really good one. Some kind of electrical generator, or a heater, and a bunch of fuel are also an extremely good investment. For food, it is prudent to have at least some of it in a different part of the house, so that a kitchen fire cannot destroy all of the supply.

In general, the backup only makes sense if it is reasonably well protected from things that can happen to the primary thing. Case in point, my second computer and phone are not plugged in when I don't use them, should there be a power surge. There may be surge protections in the building and/or power strip, but I don't know, so I'm not about to find out.

Those are all the necessities that I can think of on the fly, but there are certainly more. People who cannot walk to any grocery store, for instance, should have two cars. If you can think of more, leave a comment. Much Love!

[1] https://www.gfilotto.com/substack-suspect

[2] https://www.youtube.com/watch?v=P2gJdbb3Pec

Why all your tools are fragile garbage

Behold, my recently expired kitchen knife. Cause of death was the first quarter of a fairly boring soft cheese that I had for dinner. You read that right: soft cheese. Today, I will narrate the life story of this knife, and many like it.

Our tragic tale of consumerist usury starts on the drawing board. An engineer of some description was asked to come up with a cheaper and lighter kitchen knife. Now, this engineer was not the top of his class. If he had been, he would be designing fighter jets or something, not household tools. This guy is a ways below the spatial reasoning prodigy that his job title used to be associated with. Furthermore, he's anything but happy to be working this job. Regardless of his professed faith, he is highly unlikely to try being 'faithful in little things'. In short, he's neither competent nor motivated.

So here he is, asking to cut weight and costs out of the previous design, in a way that prototyping won't notice. Since prototyping and design testing is going to be done by people of comparable skill, they won't notice too much. In mass production, plastic is cheaper than metal every time. Therefore, he makes the blade socket as short as he can get away with. The cheapest production material of all is air, so the plastic grip is hollowed out. Maybe the CAD has a button to automatically insert some ribs into the hollow, that will make the FEA guys happy, while saving weight. [1] [2]

Here is where the problems start to show up: neither CAD nor FEA are particularly handy, because so few of my generation are. Especially the nerdier guys, who constitute the bulk of engineering students, grew up in the gaming chair, not in dad's garage. Ergo, their structural requirements and tests are fairly inaccurate. To save weight and cost, they always err on the side of too litttle material. After all, the worst that could happen is that the knife breaks sooner, giving them an easy sale in six months time.

A third way to cut costs is to use lower grades of material. The stuff will break sooner, but it's cheaper, lighter, and probably easier to work with. Now we have an under-strength design of inferior materials, so let's go to the actual manufacturing of our short-lived knife.

As we already saw, nobody's trying to spend any money whatsoever on this thing. So the cheapest manufacturer gets hired. You get three guesses where this manufacturer is and how much he cares about training his workforce, but you'll only need one. These guys do not care to make the end product one iota better than their QC forces them to. Here's an audio of their QC guy looking at bad production runs. I have heard stories about QC getting pressured to pass a certain amount of product, to make their toxic manager look good. The effect of that is the same as curves and affirmative action in academics. [3]

Finally, we get to the last culprit of this tragic knife, the customer. You and I share some of the blame. See, we're also not as handy as our forefathers. So we use the underspecced cheap knife in ways that it should not be. For instance, the tragedy up there was not a cheese knife. And when it did not want to cut, I just pressed down even harder.

This is one of the ways in which nothing works anymore. Looking on the bright side, consider the benefits of teaching your children the universally attainable virtues of thoroughness and humility. If they don't think of their work as beneath them and pay attention to it, they will outperform a lot of people.

[1] Computer Aided Design: drawing stuff on a computer.
[2] Finite Element Analysis: a way to stress test a design on the computer, without having to make a prototype and then smash it up.
[3] Quality Control: the people who are supposed to stop bad product from hitting the market.