AI - What is REAL AI and what is not.

[XDN]

@Keith Smith Can you pin this, it's an important read
 

Chat GPT has brought a general awareness to the buzzword AI. Recently, I’ve seen the acronym popping up in the baccarat world. So I wanted to write an article about how the acronym AI is being used incorrectly, and why it matters to you….and your wallet.

 

Is it the holy grail for baccarat – No

Can you buy a spreadsheet/app from a YouTube video claiming to use AI to beat baccarat? Very very VERY unlikely.

 

What is AI? - AI/ML is, at it’s core, is training a decision model on a set of data, building mathematical functions, weights and vectors into a decision model based on statistical probabilities. It “learns” to make decisions based on probabilities. This is slightly different than programming. In programming you explicitly tell the computer what to do. When an AI model is trained, it learns what to do. That is it’s power, it’s also why baccarat is currently not “beatable” by AI.

 

Deeper dive…but not too deep 😊

Probabilities: An easy way to think about it, if you train a model on shoes where Banker dominates Player 60% to 40%. The AI model WILL recommend Banker every time. The model is perfectly trained on the data it saw, it knows banker was 60% of the occurrences, and will recommend banker for every hand because that is the highest “probability” for the next hand in the data it has seen. It’s a simplistic example but that is the baseline.

 

So…lets give it more data, data that matches baccarat statistics. Ok so we generate a million shoes and retrain the model on data that is more in line with baccarat statistics. Almost 50/50. The AI model will basically recommend opposites…until the bitter end. Get a banker, AI will recommend player because it has “learned” from the data it saw that Player and Banker are 50/50. It’s fine if you are playing a zig zag shoe. You might win, but if you get a 10IAR player run, it will be recommending banker all the way down that run.

 

How does XDN know? Who does this guy think he is? I'm no expert on the scientific level but I took a deep dive into this stuff a few years ago, I trained my own models. The model that performed the best only recommended on the first bet, 1S1 or 1G2. This model hovered around that 50/50 mark. If it hit 7/10, you KNEW it was going to eventually lose 7/10. It was perfectly balanced, again because the AI models predicted exactly what the baccarat statistics said to predict. You can do this without an AI model, this is where XTB came from, join the forum to find out about XTB.

 

But But But Chat GPT!!!  - Yes ChatGPT is a huge advance in AI technology. It can find data from text that it has seen, it can generate text related to the question, its a vast improvement in plucking out TEXT data from the data it has seen.....….but it cannot “predict” the future. Lets go back to the baccarat problem….

1.       If we feed AI biased data/baccarat shoes…it will give us biased predictions.

2.       If we feed AI data that perfectly duplicates baccarat shoes (50/50) it will predict the opposite of the current hand.

There is no statistical probability for it to recommend outside of 50/50. If there is, the training data was not complete. So...how can that possibly predict future hands. It cannot.

 

So down to the crux of the article. Do not buy apps using the AI Buzzword claiming to beat baccarat. It is not possible at the current stage of the technology. There are guys out there calling their spreadsheets and card counting apps AI when it is no more than a pre-programmed set of rules and statistical forecasting methods that have been in use for 100 years. They may be helpful, but it's all known methodology. A "machine" has not "learned" to beat baccarat. The machine cannot make its own decisions, it can only "decide" based on the most probable decision it has calculated from past data it has seen. If a model finds a bias, due to an incomplete data set, because over time baccarat is 50/50. It may win on shoe A but will eventually destroyed on shoe B. If training data is 50/50 on a "true" baccarat stats dataset, AI will always recommend the opposite hand. It's just mathematical probabilities. The YouTube guys you see are just using the buzzword incorrectly at best...as a ploy to lighten your pocket book at the worst.

 

Why does XDN say this, who is this guy, why does he think he knows? It's not me,  lets look at some facts so you can make up your own mind.
 

Chat GPT – Fantastic tech…why? Here are some relevant stats…

https://news.microsoft.com/source/features/ai/openai-azure-supercomputer/

https://lambdalabs.com/blog/demystifying-gpt-3

--Was trained on 300 BILLION tokens (words) collected from crawling basically the entire English language internet.

--Has 175 MILLION training parameters/Variables in the program. That’s 175 MILLION ways the training can be

----“tweaked” to fine tune the model. Think of a parameter like this simplistic example… “Does the model predict better

----with lower case text or upper-case text.” 175 MILLION of those kinds of "tweaks" to train ChatGPT

-- Using 1,024x V100 GPUs, researchers calculated that OpenAI could have trained GPT-3 in as little as 34 DAYS.

-- It would take 355 years to train GPT-3 on a single NVIDIA Tesla V100 GPU. Those are top of the line right now and about$3500 EACH on Amazon.

---- https://www.amazon.com/PNY-TCSV100MPCIE-PB-Nvidia-Tesla-v100/dp/B076P84525

- - Estimated that it cost around $5M in compute time to train GPT-3

-- Microsoft invested  $ 10 BILLION dollars into Open AI to fund ChatGPT. This is HIGH DOLLAR STUFF!!

 

So…when you see guys on YouTube claiming to “Win Every Hand” with an AI spreadsheet or an card counting app that can "Predict with AI" that you can buy for $ 50 bucks…. Use your common sense. Is it possible this person had access to the data, hardware and financial resources to train a REAL AI model that is as effective as ChatGPT? Would they be making videos on YouTube and selling an AI Spreadsheet/App for $50 bucks? That NO ONE KNOWS ABOUT? If someone does crack blackjack/roulette etc.. with AI they will win the Nobel Prize. It’s not going to be some guy in his home office (like me).

 

It takes millions/billions of dollars of supercomputers with vast networks of GPUs to train Chat GPT models. Simply asking the model a question takes high dollar hardware. It’s EXPENSIVE both monetarily and computationally. To even get a recommendation, you either need a web service like ChatGPT has or a very VERY specific set of code and code packages on a development machine to load and query a trained model. It’s not like loading up your email program, you won't be getting "AI in a spreadsheet". The code to create/interpret these models breaks constantly as new changes are rolled out

 

What do your instincts tell you? Do you think a guy got all this in a spreadsheet or a card counting web app?

 

I’m sorry, I don’t want to seem like I’m slamming the guy, his app may be helpful. But when the YouTube video title says “I win every hand using my AI spreadsheet/app” then you see a negative progression used...Its NOT AI. If you watch the entire video, buried within is the negative progression play. NO, this is NOT AI. It may be some rather advanced math and statistics, and it may be a helpful program, but it is NOT AI. A TRUE AI would not need a negative progression. It would not lose. This is a marketing ploy using the buzzword of the week in an attempt to extract funds from your wallet.

 

Why am I writing all this, I just don’t want you to waste your money. The Ultimate Baccarat app doesn’t use the moniker AI because it is NOT AI. It’s a very flexible authoring of playing methods/rules that have been fine tuned here at BTC by guys with decades of experience. It's the best tool for finding statistical anomalies in baccarat today....BUT the player STILL must make the decision. This is what sets TUBA apart. It looks for potential exploitable anomalies in the game and presents them in a fast concise way so you can make the best decision possible...quickly and accurately.

What we do here at BTC is try to identify SHORT TERM ANOMALIES in the normal baccarat stats. Make a couple of units then look for the next anomaly because eventually that ANOMALY will succumb to the normal baccarat stats....that everything is 50/50. There is no "prediction."

My Pledge: I will NEVER use the moniker AI ... if it is not REAL AI.

AI/ML fascinates me to no end…it’s exciting! I plan to keep an eye on it over the years. Someday, someone might crack it. But if you ever see me use the moniker AI related to any of my software, rest assured it WILL be a REAL AI model and solution. Not just the buzzword of the week.

Edited February 21, 2023 by XDN

[mookie]

Thanks for writing this Eric.  It's not easy to explain things like this and you made it very easy to understand the concept.  We all appreciate you!

[Craftsman]

Thanks @XDN! This was a great public service announcement. Naturally, everyone is trying to duplicate the thinking and computational processes of the human brain!

This is the holy grail everyone is looking for. The human brain has an unbelievable number of neurons and synapses which all work together to enable us to think.

In fact, the process by which the human brain does actually think is known as fuzzy logic. That is the closest way we can approximate it in the computer world.

You are right XDN... You will never win Baccarat simply by looking at the raw data. You must be able to infer what will happen from what the shoe is implying will happen!

Which means everything is basically a set of probabilities.

The human brain is 'The Ultimate Computer'! Which also happens to be the name of a great Star Trek TOS episode!

All of the methodologies we use, and the 10 best bets, and all of the other little tricks and schemes we use to try and win at Baccarat, is just our brain using fuzzy logic.

Fuzzy logic is not simply true or false, light or dark, one or zero. It is a range of variables from one to the other; an infinite number of possibilities between 1 and 0.

This is what gives our brain the ability to 'see' approaches to playing Baccarat based on insights that our brain gets from analyzing unclear data.

 

There is a very interesting paper (PDF) from Cornell University concerning how the brain actually works. I'm going to attach it here in case anyone is interested.

The title is; Basic Neural Units of the Brain: Neurons, Synapses and Action Potential  it is very interesting reading! Even lightly skimming it is a wealth of information!

As a side note, I believe that scientists attempting to duplicate the human brain's capacity to think is one of the most dangerous things they can possibly do with their toys.

Scientists can be very stupid sometimes, and lately, they are getting very stupid! Mounting machine guns onto robotic dogs!? Making robots that actually think!?

In their insatiable quest to always try to outdo themselves, scientists seem to be always forgetting the dangers involved in what they're doing!

There is a famous quote from Jurassic Park... "Just because they can do something, doesn't mean they should"! That is so very true.

Remember what happened with Skynet!? Things didn't work out very well there, did they!?

Your brain is the most powerful computational device in the known universe! If anything can be used to win at Baccarat, it can! 🙂

You just have to be able to interpret the data correctly.

Basic Neural Units of the Brain - Neurons, Synapses and Action Potential (1906.01703).pdf