I did buy the AI hype. It's marketing bullshit.
The more I understand this technology, the more I'm resolute on this stance.
A waste of money
When looking at SPY, XAU, BTC against the dollar, what do you think did the best?
When looking at just these comparisons, gold won by a huge margin. (check my analysis here)
The top ten companies in the SPY represent 30% of the economy. These companies rely on AI investment, to the amount of 13+ trillion dollars.
Given this, why hasn't AI transformed the SPY? Why is it doing less than gold dug from the dirt?
Digging the truth with AI
First of all, LLMs aren't intelligent. They are predictive based on data. So the very description is misleading.
The technology works like this (put simply):
- An LLM (large language model) typically breaks a sentence down into chunks
- Those chunks turn into numbers
- An algorithm predicts the next sequence of numbers given all the numbers it has from the data it has
- It converts those numbers back to a sentence = output
This costs tokens.
This isn't intelligence. It's prediction.
Predictions based on data sourced under serious scrutiny. Accusations of theft and stolen work often surround this data.
The scaling problem
The hype for this technology drives investment.
Massive data centers in the hope this algorithm gets more precise with better compute and larger data sets.
This process and the use of tokens is not a good financial model. LLM companies are losing SERIOUS MONEY.
Maybe a more stable architectural base means cheaper token use for all, but nobody knows if the product will improve. Or if it even can.
The human problem
Like I said, I have fallen for this technology at the start - it's intentionally addictive.
Its overuse is causing AI psychosis and cognitive decline. I covered this in a wider sense in my previous post on AI, so I won't write on this again.
The other problem is greed. When you use it, it defaults to being very agreeable, not necessarily providing you with useful information. It's designed to hook you.
You can prompt that away, sure, but ask yourself this: Is it better than the rubber duck method?
Rubber ducking is the idea of talking to an inanimate thing (a duck) to solve problems.
It's a crazy take, but it's worth thinking about.
Despite my negative take on LLMs, it has its uses!
What is it good for?
Code compiler
It's great for writing code as LLMs can fill in the gaps. The trick is to learn what it does otherwise you fall in cognitive decline.
It's similar to how Python takes advantage of other languages and memory allocation. LLMs can do the heavy lifting.
I will say though, that if you're doing things not within the model's dataset(s), it will find it difficult to predict what the following code should be.
Hence the new data centers and so on.
Computational stuff
Protein mapping comes to mind. The old way was growing protein and blasting them with X-rays with lots of math involved to figure the shapes of the beams bouncing off.
AlphaFold was fed protein shapes that humans mapped and did said calculations very quickly.
Deep Learning Super Sampling (DLSS) boosts gaming performance and image sharpness. This stuff is really cool.
Language correction
Think Grammarly on steroids.
What does this mean for society and the future
Jobs that supposedly will be replaced by AI won't be.
There may be an initial explosion of layoffs as people rely on this technology to replace developers, but when the code breaks, those developers will be required again.
Developers will be fired from the hype that they can be replaced, but it won't fix the 70/30 code problem. They will be needed back when companies realise they don't understand the code.
When it comes to art, they can't keep stealing this data. Programs are being built to counter its theft.
The economy
This is what scares me.
With so much invested in this technology, if it's not realised, I don't quite know what that means.
- This technology will be realised
- The actual use of LLMs will be put to its actual benefit
- De-escalation of its marketing.
This conclusion comes from my data analysis and use of this technology. Not financial advice, but it's easy to conclude this technology is not financially viable.
If I'm wrong
My blogpost is like saying the SPY will never drop to 0.
If I am wrong, then it won't matter - I will be fucked. So will you.
Either no jobs will exist, or I will live a life of servitude like everyone else.
The question is whether the technology is owned by humans or LLMs.