Neural network strategy

TensorFlow has a stock price prediction model available on its homepage. Apparently, it is about 60% accurate. The challenge is the changing externalities (new releases over time from the company, government, competition, interest rate changes, tax changes and so on) that impact price can't be predicted.
I would really like to connect with someone who could amend this model for a slightly different approach that I believe will be better aligned with how AI solutions work and the markets for trading.

How is it possible to get a statistically reliable outcome from disorder? If betting against x event is profitable then we have to assume x happens more than it doesn't which implies there is already an order but there isn't. Am I missing something here?

I believe you have to categorize the data into ordered and unordered then look for patterns in the ordered as well as signs for start and end to the ordered. Your strategy should focus on trading the ordered and standing aside on the disorder.

You have to analyze the code, think for yourself what you are trying to accomplish and what you are actually doing. You can analyze the data many ways but you have to start somewhere and go from there.

Thanks for the reply. We have analyzed the data and have a model that is profitable based on the data. The next detail is to code the rules and come up with introductory descriptions of the patterns for the AI to operate with, the goal is that over time the AI should be able to refine the patterns.

Hi,
as for predicting future prices, I have developed an indicator based on data provided by a broker, with inputs including OHLCV data and candlestick time data. I presented the results on YouTube during my live broadcast (just to demonstrate my work). I discovered that neural networks (NN) are suitable for estimating price levels where the price is likely to reach with a higher probability. This works particularly well for larger price movements above 300 pips, with a 72% reliability (tested on data from three years ago). I realized that training on large amounts of data from many years ago is inefficient, and it's much more effective to train the data continuously. This led me to develop an indicator that learns from live data every millisecond. This approach was the only way I achieved the desired result of at least 60% prediction accuracy. I'm not sure if it's possible to provide a link to the YouTube videos demonstrating the functioning of the NN on live data. If it's allowed and you're interested in taking a look, please let me know, and I will provide a link to the YouTube.
The key is to ensure that data prediction systems are constantly familiarized with the ongoing changes in the market. The fact that you train on the data you trade with gives it real meaning. The inputs for the neural networks (NN) should be clean data based on your actual trading activities. Inputs consisting of various indicators essentially represent a relativized perspective and, in my opinion, are not suitable for predictions (this is just my opinion). Trading currency pairs, for example, using NN inputs such as MA200, RSI, ASX, does not provide a pure input; it is already a relativized input. A neural network can essentially generate these indicators within its black box based on (for example, 500 OHLCV candles + time). It may sound unbelievable, but NN doesn't need to receive data distorted by some function; the creation of functions is the essence of NN. This is purely my empirical opinion, which I have formed over years of work, primarily in predicting currency pairs.