LLMs are all wrong with the computer world

Monday, December 23, 2024

( Permalink )

For decades corporations have done everything in their power to make computers even worse. Software back then was faster, leaner than it is today. Hardware performance has improved by leaps and bounds, but somehow the software is slower. Yes, the programs we run (cough, web applications, cough) are more complex, but the increase in complexity is less than the improvement in computation.

The software used to be better written too. If sending patches to clients is a risk in itself, releasing barely working software is akin to shooting yourself in the foot. Now bug expected. Again – I’m comparing simpler software to what we have today, but the increase in complexity is less than the increase in budgets and team size.

The software was also:

• more open
• buy
• does not require internet
• not as streamlined
• not designed by psychologists to be addictive
• is not designed to be a constant stream of income
• rapid development
• possibly created by teenagers in their bedrooms
• simple (as in have a limited use case)
• small
• running on different hardware architectures (yes, there used to be more than 3 systems)

I have often been accused of romanticizing the past, but I am far from that. The software of the golden age had many problems, it was far from perfect. I’m critical of the path we’re on – because it’s not meant to be better softwarebut in better return on investment. We wish we had better working computers, but here we are. Everything is barely moving but the dollars are pouring in. Some call it enshittification, others distribution. All I know is it won’t stop. It is not enough.

This led me to ChatGPT, LLMs and all that crap. I have to admit, the way it works is pretty cool. How much vector math can trick us into thinking we’re dealing with someone? It’s very nice! I love that it surprises you, but if you look at it critically, it’s a terrible peace of software. It is slow, expensive, and non-deterministic. When I release a production code with a few times to work or not, I need to fix it. If the same software gives different returns by designI need to rewrite it.

I am absolutely convinced that LLMs are not a path forward at best, and a huge step backwards at worst. Many people look at this technology with fearful eyes, but there is no chance that LLM will be better than them.

But, like the examples above, it’s better not to always win. I fear that too much money has been invested in this dead-end technology for it to fail. The entire computer world is in the hands of a few large companies, and each of them is completely risking their future on it. Google, Microsoft, Amazon, NVIDIA, AMD, Apple – by the end of 2024 there will be no major computer companies that are not “AI first”. Will they let their investments go to waste?

A Google search was futile; Windows is a clown show; Facebook/Twitter is untouched by humans. Although they own BILLION that dollar. They can throw piles of dollars into the data center furnace and wait. The fact that they also benefit from dollars does not make it difficult.

I know people who can no longer work without asking ChatGPT and they lived in better times. But I also know teenagers who don’t have fast computers at all; for whom a slow web app is the only status-quo they experience. For them the current crap software is simple software. I’m really afraid that future generations will never know deterministic software. Companies can continue to add “AI” features (since AI is now a synonym for GenAI), and raise its children. Can children notice that it is harmful? Will they ever know?

Will they give a rat’s ass that GenAI has only two goals: first, be the abstraction of all computer interactions; second – to make people obsolete? I think the first is the goal, while the other is what GenAI salespeople are selling to investors. Microsoft and Google want to be in the middle of everything we do without computers; that has been their strategy for a long time. This is a self-motivating nightmare: the worse the vision, the greater the flow of money, which drives the creation of computers that are useless. We’ve lost so many alternatives – Google is trying to fix itself with GenAI after destroying itself with ads, after destroying all the competition. The web is now beautiful. Microsoft Windows has eaten all other x86 OSes except Unix – and we’ve had a lot of good ones. Apple iOS or Google Android is required – but this was not the case a few years ago.

Progress is not linear and better don’t always win. In most cases anything that gives more returns to investors. And no other technology has more investor risk attached to it than LLMs. AI has many usecases, but they have been discarded to make way for GenAI. This steamroller can destroy computers, the economy, and the environment. But nothing can be done about it, because none of us are billionaires. All we can do is dance how it pleases them.

Except for what GREAT will happen. With infinite pockets, giants can only be burn the money – literally.

I will keep this site free of althole, and I will personally run the software without LLM. I don’t use copilot, I do dozens of chats on GPT. I will try to keep the LLM topics here to a minimum. This is my life and this is where I have control. This is the only rebellion I can do.

In unrelated news – SystemD is now trying to convince people that having file-based logs is obsolete. And in a few short years we have Linux “administrators” who don’t know how to use grep(1).