AI is a bubble.
I used to be a believer and thought we were heading for an ASI.
I don't believe that now because I have a much deeper understanding of how they work and can see that I was wrong.
There is really no such thing as "artificial intelligence". It's machine learning.
LLMs are flawed in so many ways.
They are frozen black boxes. They are trained only on text and they use gradient descent to predict what comes next based on what they've seen.
In no world is this intelligence. They only simulate intelligent language with math.
They have no understanding of what they're doing. They're not conscious.
The human eye alone processes 10 million bits per second.
That's 315,360,000,000,000 bits by age 12.
315 trillion bits.
And that's just the eyes.
You also have touch, hearing, smell and taste.
touch gives us full interaction with the world and combines with the other senses so you have a 1+1+1 = 5
You touch, see, hear, smell, taste and manipulate.
This is why babies/toddlers put everything in their mouth. It's part of the pre-training.
On top of that, the human brain has memory.
It never stops learning.
It's LIVE, not frozen.
And we are CONSCIOUS. We are aware. We can observe our thoughts and change them.
Consciousness is not some simulated emergent property of the brain. Your consciousness/awareness comes from somewhere beyond the brain. It's likely consciousness is the inherent property of all existence and the rest emerges from it.
The whole "we're close to AGI" is hype. ASI is laughable. Absolutely laughable
LLMs are machine learning. They are prediction machines.
Even the code they produce will miss so much. With the right guidance it's useful and you can generate a lot of good code if you use them a certain way. Ie, not trying to one-shot an application, but having it build piece by piece and you, as a human, guiding it and making it aware of "gotchyas"
They can certainly be fine-tuned and improved to find some bugs and help with debugging, but it'll never be a general intelligence like a human. It can only be trained on past data in order to predict similar things in future data. This is NOT what we do. We are alive and aware and we observe and change thoughts based on those thoughts and the observable world.
SEO is the same. AI is not going to takeover. AI content can be good, but it can't add anything new. It's only consolidating and refactoring existing information.
You'll see over the coming year or so things will start to cool down as more and more investors realize we're in a bubble.
Each new model release is flat.
There's really not that much of a difference between gpt 3.5 and o1 pro. It's all hype and "benchmarks".
Most of the differences you see are just with reinforcement learning to produce more useful answers for humans.
o1 pro and reasoning models improved because they are "multi-pass" with thinking tokens. That makes them better at coding.
But all that's happening now is cost is going way up in order to get slight improves through more thinking tokens.
o3 is likely very good, but it's just computation. It doesn't reason. It can't reason. It can only statistically predict a correct answer by statistically predicting thinking tokens, which help it to predict a final answer.
