So a friend of mine does algorithmic trading on the stock market. He's tried time and again to use chatgpt or other AIs to help predict the direction of the stock market - along with massive amounts of input data, in addition to training data the LLMs were already provided with.

Guess what? Less than 5% accuracy in the prediction over a short timeframe (eg. if you ask it today, it's answer for tomorrow is inaccurate) except for a 'general direction'.

Useless for investing because a shoeshine boy could tell you the same news for less.

Now this is for a structured setup where there are sentiment forces and potentially predictable factors.

With chance-based games, I suspect this will be like using a coin flip to predict the output of another coin flip. Highly random. And because the information is not probabilistically actionable enough for the AI, I'm quite sure chatgpt will 'manufacture' fiction and give you a number / direction because you asked for it.

So that fictional pair of dice you told chatgpt about are about to get a fictional number. 🎲🎲

Or as we say in programming,