The problem is I am currently running the stock AI on my laptop, and it does not have the ability to run at a good enough efficiency.
So I need to move it over to the desktop.
The laptop has already been upgraded but it still lacks compared to the desktop.
So I load up the satellite servers on my laptop to monitor the activity. I have the highest admin level logins so I don't need to hack in anymore.
Just sign in like any other staff member.
So with the satellite activity up on my laptop and the stock AI successfully moved over to the desktop.
I'm ready to begin the simulation.
Just set up the virtual desktop, run the stock training program and make sure it is on an hour delay.
Now pull up the AI user interface and connect it to the fake account on the training model.
Then set the parameters for the AI to run on.
Click start.
I have to now pull up another AI interface to run on the real home screen to analyze the data that the other AI produces.
Start.
Time to wait.
…
1 hour later.
I decided to update the simulation AI with the data the other AI analyzed from it.
For the hour, the simulation AI performed worse than I expected. It only had a success rate of around 10%. Which means it lost money 90% of the time.
I thought a good number to aim for would have been 30 - 40%. Closer to 50 - 50.
But instead it was drastically different.
Let's try another hour to see how it performs with new data.
…
Another hour later.
The AI performed a little better at 11% but that could just be a coincidence.
After updating the AI with the new data. I decided to just let it run all day to really get a good amount of data.
…
At the end of the day.
The AI has actually increased its performance to 35%. That is without the analyzed data from the other AI.
That should make a difference since it analyzes the mistakes it made. So it should stop making them or at the very least reduce them.
This is still not at a satisfactory percentage for real application so I will just have to leave it running over night.
One thing I am surprised about is how little data it is creating. I guess that is because I am only focusing on a few stocks.
Once I move to a more comprehensive data set I bet it will generate a lot more.
With the end of the day it brought another length of time that would solely be used for grinding out experience.
…
At least that was what I thought.
The next day I awoke to a text message from my uncle, Erik.
"Would you like to hit up the bike trails tomorrow?"
It was a message I had been expecting for quite some time. I had expected it to come sooner as the hot summer was coming to a close. But instead it came in autumn as the days were drawing shorter and growing colder.
I really don't think we will get too many rides in before the season changes and it gets too cold to ride.
What my uncle was referring to was mountain biking. It was something we started doing when I was 7 years old. Now I am 25, and we still go riding every year.
I do look forward to it but, the message always brings the anxiety of having plans to do something.
I don't know if other people feel the same?
But, when I have plans I dread having to do them, but once I'm there I always have fun and look forward to the next time.
I respond with the auto generated, "Sure, what time?" message that my text app suggests.
"Be at my house around 1?"
"Ok"
A quick text back and forth with no other conversation.
A very normal interaction.
We don't really text each other unless it is to set up plans, which mainly involve going mountain biking.
Welp, now that I have plans for tomorrow, today is going to feel infinitely shorter.
Time to manage the AI for another day of data mining.
Looking at the progress from last night. The AI made some significant strides forward. The gain to loss ratio had reached 55% in the favor of gain.
I finally hit a positive value. It had a 5% more likely chance to gain money than lose it. I think. I don't have a major in statistics. So if I get corrected it is by a guy who can't help but nit pick.
Anyway, I think it is about time I disappear from this website's servers.
So I load up the hacking AI and begin hacking into the website.
Once I'm in I erase my IP and go ghost. Now that my AI is about to go sicko mode I don't want them to be able to track its progress.
Now I can add the analyzed data, then start the simulation again.
…
The first hour of running the AI proved to be fruitful as the data from the analyzed AI was added to the simulated one.
Which bumped its operating efficiency of 55% up to a solid 62%. Quite a large jump, but that is what I would expect from an entire night of analysis.
I expect that after another entire day of data collection I will see that the effect will start to plateau. The easiest part is always to the half way point. Everything after is a slow grind, with each percent after the previous being just that much harder to obtain.
I do have to wonder if achieving 100% is even possible. It is not like the AI can see the future, all it can do is make predictions based on the data it has available.
So, I wonder where the line is drawn?
90%?
80%?
Maybe even less?
I guess I will find out.