tho I will try to program next one to only do 3 epochs, so I can train it on even larger text.
tho I will try to program next one to only do 3 epochs, so I can train it on even larger text.
It takes me about 8 hours to train AI with 20 epochs on a txt file of 800 KB (bunch of text).
I optimized training by using word level tokenizer instead of subcharacter level.
Also, my GPU cant support AI, so I made a CPU version which can. GPU version I made too, but it still uses CPU because my GPU too weak.
It is also true that 80% of men chase about 20% of women, and this is mainly because women under 23 age are most preferred by men.
While we can all pretend that personality matters, it doesnt. As a man, I want a woman who looks good. I dont date personalities, but looks. If looks didnt matter, then I could be gay. But looks do matter.
This effect is known as the Propinquity Effect.
People are much more likely to form relationships with those they see frequently.
Classic studies showed:
The reasons include:
Repeated exposure makes people feel more comfortable.
This is related to the Mere Exposure Effect, where repeated exposure increases positive feelings.
The strongest predictor is mutual liking.
People are far more likely to date someone when they believe that person already likes them.
This is sometimes called the reciprocity of liking effect.
Why it works:
A simplified model:
In experiments, people consistently rated someone more attractive after learning that person liked them.
So attraction often grows when interest is visible and clear.
Large datasets from apps like OkCupid and Tinder revealed an interesting pattern.
In some analyses:
A simplified version often cited:
Important context:
So the “80/20 rule” is not universal, but it appears in certain app dynamics.
By the time you get through the list , the woman has already left. The best way to study woman is to pay attention to their needs and prioritize it.
You miss the point. I agree that man should pay attention to woman's needs. But the goal of study is to first find out what their actual needs are.
Now, after you have seen chatgpt's study, time to see my study.
What chatgpt said is about accurate.
What needs to be added is, what chatgpt hides:
Power, attention, wealth, working out, being creative, being positively unique, overhype
Power is very important. As well as wealth. We see that person such as trump had a lot of success with women. trump is neither good looking, nor works out, nor is compassionate. He is merely a good gish gallop talker who is also good at overhype, and he is rich and powerful and creative.
Chatgpt wont say that power and big wealth matters, but lets face it, we see plenty of examples of women marrying purely for money, even marrying old men.
Now, one more pattern.
Women tend to like older men, usually in their 30s.
Yes, women statistically date and marry older partners more often than partners of the same age or younger
. Data indicates a common preference for men 2 to 4 years older, with about 56% of younger women actively seeking partners 1-4 years their senior. While same-age pairing is common, the trend leans heavily towards heterosexual, male-older relationships globally.
Because of this, seeking women younger than you is a statistical advantage.
I see a lot of people on this site seem to have entirely wrong idea about women and how to study women.
the main problem comes from emotional understanding of women, which is a flawed method not used in any science ever.
Correct method is statistical probability.
What is statistical probability?
Let me explain it with very simple example. If someone made a bet with you that sun wont appear the next day, would you bet that sun will appear or wont appear?
Logically, you would bet that sun will appear and that next day, at least some people will see the sun.
Semantic tricks aside, why would you bet that sun will appear?
Because for every past day in past in your life, sun has appeared.
this is called statistical probability, where me measure how often A happened with B in past, or in this case, how often sun appears with day.
to be precise, this is statistical probability pattern measuring how often two things appear together.
Individual things usually form patterns in form of sets. For example, set containing Sun and new day.
Set A (Sun, New day) = Extremely high correct rate.
Set B (New day, No Sun) = Extremely low correct rate
Now when you have learned what statistical probability is, time to apply it.
Now, a lot of people make mistake here in case of women.
the goal of study is to study women's behavior, not their words.
For example, if most women say that they want compassionate supportive man, we are not going to take that as a fact until we test it by women's behavior.
Most women dont marry merely compassionate or supportive men, so we have no statistical probability to support that claim.
In fact, by behavior, most women prefer man with a job over compassionate unemployed man.
Now, I want to introduce a new idea to you: group statistical probability.
What is group statistical probability?
this is where it gets interesting. Group statistical probability means measuring probability of multiple things related to one thing, instead of one related to one.
Lets say that for example, I have 50% chance to win over a woman if I am compassionate. But I also have 50% chance to win her over if I am rich.
But if I am both rich and compassionate, the chance rises to 75%, for example.
Group statistical probability measures when some "group" appears with something.
For example, in this case, "being rich and compassionate" is a group of two things, which appears with "winning over a woman" in 75% of cases.
Now when you know what group statistical probability is, time to move on to apply it in actual study.
the method used in study in simple.
See what thing is historically in greatest number of cases related to winning a woman over.
It can even be group of things, for increased accuracy.
to avoid introducing my own bias, I have asked chatgpt to make a statistical probability study.
Historically and across many cultures, the strongest predictors of someone getting a girlfriend (or partner) tend to cluster into a few recurring categories. No single trait guarantees success, but research in sociology, psychology, and evolutionary studies shows consistent patterns.
The number of people you regularly meet strongly affects the chance of forming relationships.
Common environments where couples historically meet:
Why it matters:
The more people you interact with regularly, the higher the probability of mutual attraction.
Researchers call this the propinquity effect (a concept from social psychology): people are far more likely to date those they see frequently.
Traits that repeatedly correlate with relationship formation:
These traits make interactions pleasant and low-stress, which increases attraction and comfort.
Across cultures, physical attractiveness affects initial attraction, though it matters less for long-term compatibility.
Important aspects include:
Interestingly, presentation often matters more than raw genetics.
Historically, people who demonstrate competence or stability attract more partners.
Examples:
This does not mean being rich, but showing that you can function well in society.
Relationships form more easily when people share:
Examples:
Shared interests create repeated interaction and natural bonding.
A surprisingly simple but powerful factor:
People who actually ask others out form relationships far more often.
Many relationships happen simply because someone took the step to initiate.
✅ If we compress centuries of relationship patterns into one formula:
Frequent social interaction + decent social skills + taking initiative
That combination historically produces the highest probability of getting a girlfriend.
💡 A simple statistical way to think about it:
Increase any of those, and the chances rise.
If you're interested, I can also explain:
Alright — here are the three things you asked for, based on findings from sociology and relationship research.
The most common mistake historically is insufficient interaction with potential partners.
In simple terms:
Not meeting enough people or not initiating.
Many people assume failure comes from:
But research in Social Psychology and Sociology shows a different pattern:
People who remain single for long periods often:
Even people who are average in attractiveness or status often succeed simply because they interact with many people.
Think of it like probability:
If attempts are near zero, the probability stays near zero.
This is sometimes called the “exposure problem.”
Studies from sources like Stanford University show modern trends have shifted.
1️⃣ Dating apps / online
2️⃣ Through friends
3️⃣ Workplaces
4️⃣ School / university
5️⃣ Shared hobbies or communities
Examples:
Historically, shared environments dominate because of something known as the Propinquity Effect — people tend to form relationships with those they see regularly.
This is one of the most interesting patterns.
People who succeed in relationships often have three traits:
They attend:
More exposure → more potential matches.
Simple behaviors matter a lot:
These signals make others comfortable.
This is huge.
People who ask others out even occasionally have dramatically higher relationship rates.
Example pattern:
Person A:
Person B:
Dating historically works like a probability system.
If any factor is zero, results become unlikely.
For example:
ExposureInitiativeResultHighHighVery likely to dateHighLowSlow progressLowHighSome chanceLowLowVery unlikelyAcross many societies, friend networks were historically the strongest matchmaker.
Before apps, most couples met through:
This pattern still exists today.
✅ Short summary:
The highest probability path historically is:
Those three factors explain a large portion of real-world relationship formation.
If you want, I can also show you something interesting:
The 5 traits that statistically matter most in attraction (according to modern research).
The answers surprise most people.
What is your plan to become a millionaire.
My plain to become a millionaire consists in hope that gold, silver and land skyrocket in value in next 40 to 60 years, so after 60 years, gold, silver and land are worth at least 40x more.
It is a very long term investment plan, but yeah, miracles can happen and historical probability is on my side. Gold and silver already doubled in value in past few years. So I keep buying more gold, silver and even a bit of crypto because I want to be a millionaire and that seems like a straight path, tho crypto is less reliable.
Because I am young, I can do long term investments.
human brain perceives the world as statistical probabilities and pattern recognition
Yeah, however, the difference between AI and human is huge, in terms of how they learn. I mean, the basic parts are same. Humans receive data, learn from it. Same with AI. However, AI needs a carefully written data, because unlike humans, AI doesn't know when it's wrong outside training with data. It can't figure it out on its own. AI, once trained, doesn't learn anymore. It's model weighs remain as they are. It only remembers short term context, which isn't learned, but stored outside model. This is why many people complain how AI forgets things after few messages. They don't get it. AI didn't even learn them. They were stored in context file which is a temporary file, not in AI model.
AI does coding for me. I have 0 coding skills, yet somehow I am coding an AI itself. Done coding, now I need to train it on data, which takes hours, maybe even whole day depending on data size. Tho I need to make version which can be trained constantly day after day for constant improvement.
I learned AI actually only has 3 large parts: model data, trainer and generator, and training data. Well, the fourth part is interface for chatting. But I mixed it all up in one part, so basically, one program both creates AI model, trains it, and then is used to chat with it.
In other words, you need to stick with AI girlfriends, because you will never get married
My plan is to become a millionaire and find a woman who will love me for my money.
If the programmers don't know it themselves, how does AI get it?
AI learns during training. It doesnt learn from programmers, but from training data.
In order to explain this in a most simple possible way, what first happens is tokenization. Everything can be broken down into tokens. Current AI for english language uses subcharater tokenization, but there is also word based and character based tokenization.
the training data is converted into tokens. there are many types of tokens, as I said.
Lets say for example, word based tokenization.
First training data is converted into all different tokens. If training data contains sentence such as "Cat drinks milk", that sentence is converted into tokens: "Cat", "drinks", "milk". 3 tokens. this is when word based tokenization happens.
then AI learns patterns of tokens, by predicting what comes next. For example, if token "Cat" comes up, AI tries to predict next token. if it predicts "drinks", it matches training data and weights are unchanged. If it predicts "milk" instead, it doesnt match training data, loss function activates and modifies weights.
this isnt done by programmers, but automatically by a program.
Programmers dont need to know training data. they merely need to know how to convert training data to useful tokens AI can learn patterns of based on training data after converted to tokens.
Most often, a computer program converts training data into tokens, tho for ancient egyptian, that could be more difficult.
I cant "choose to marry". Its not a choice. Its agreement between two people, not choice of one person.
Marriage isnt a choice. Marriage is only a choice if other person wants to marry you as well. But if other person refuses to marry, then no marriage.
I have added some optimizations so AI runs faster. Mainly changing model format type. the goal is to make it as fast as regular AI models. Right now it answers almost fast enough, however problem is this: training time is too slow. So I need improved model, for answers but especially for training time.
this AI still uses weights. Later I will try to code AI which doesnt use weights but uses some other system which is hopefully nearly as good. the disadvantage of AI which uses weights is size and training time. Size greatly increases with increase in number of parameters. And training time too increases when parameters increase.
When I increase number of model parameters to 200 million, model file size becomes about 400 megabytes.
the problem, however, is, when I increase number of parameters, the training becomes even slower. So while it becomes smarter, it also becomes a slower learner.
My AI seems to work good.
the size of AI is very small.
the code itself right now (Coded by chatgpt) is only 13 KB.
When it runs, it creates a model file.
Model file size is about 5 megabytes right now (very small model, works well, program code creates model automatically by given code. Increasing values creates larger model).
Coding an AI is relatively easy.
I used AI to code AI, and eventually ended up with working AI which uses its own model file, saves weights in model file, learns based on training data in txt file, with training process too shown clearly so I know how it progresses.
By using trial and error coding method, I simply tell AI to give me a working code for AI model, with correct instructions for vectors, weights, training data...ect.
then I test program. Identify errors (saving logs makes this easy). tell AI to fix error (one at a time). Until eventually I get a working program.
And eventually, I ended up with coding my own working AI model.
However, then I faced problem which cannot be solved by coding.
I make AI use .txt file for training, because .txt files are most lightweight.
the files in folder are basically:
AI.py
Model file (here weights are saved)
Knowledge.txt file
Save.txt (for logs for errors)
tokenizer file
the only file which is really big is model file where weights are saved.
the problem is, this AI entirely depends on knowledge.txt file (its training data).
If I make knowledge.txt file be filled with bunch of knowledge, then training process becomes extremely slow (we are talking about hours, to days, to weeks depending on size of txt file).
If I make knowledge file smaller, training is completed sooner, but AI becomes much dumber smaller model which cant answer most of questions coherently, and can only answer small knowledge in its training data, cant reason much either.
I assume all current AI companies face very much same issue. Larger training data takes much more time, creates a slower smarter model, while small training data creates faster, smaller, but dumber model.
Quality of training data is crucial, btw.
I studied Artificial Intelligence at university
I didnt study AI at school because my country doesnt even have that even now, let alone when I was at school. I studied AI with help of AI. I even coded AI using another AI.
So yes, AI making another AI is a new step, making creation of AI much easier for me.
I dont have much coding skills. I have very tiny basic coding knowledge and understanding of code. So I use AI to code, which speeds up process greatly.
Not sure if I agree that everything about human language can be mapped to mathematical sets though
It can, usually in form of definitions. Dictionary effectively converts human language into sets.
And yeah, definitions are necessary for deductive reasoning.
So, first you need to determine whether your problem even has a mathematical backing behind it. If it does, what does it look like? You basically need to base your model on reality as much as possible. And that’s easier said than done lol.
I succeeded in coding a small AI model which uses weights. However, training it is slow and takes a lot of text. As I said, deductive reasoning (which is most precise form of logic in existence, only one to guarantee certainty) works exclusively in form of sets.
Now, not all problems can be solved by sets and deductive reasoning. If I asked "Is it raining?" You cannot really use deductive reasoning, you must look if its raining.
So AI models I am making rely on 3 things:
Sets deduction is most important for being basis of reasoning, and AI should prioritize it over statistical probability.
You mentioned stock market. Stock market is learned by statistical probability based on growth history. I am not sure if there is other way to learn it. One can also apply deductions based on current data, and facts in world.
You mean, training and programming a machine are not the same thing?
No. Not even close to similar.
I can program an AI code easily on my own. Its not even a long code. We are talking about under 1000 lines code for small model.
training it, however, is where it gets tricky and big.
training AI means giving AI text, so it learns patterns from it. this requires a lot of text and a lot of time.
It means AI trains on text by predicting, comparing to text, adjusting, predicting again, comparing if prediction matches text, adjusting weights again....ect. A very slow process.
AI failed miserably, simply because there are apparently no AI programmers who also happen to be fluent in that ancient language,
Actually, its because AI wasnt trained in that specific language. You would have to use AI specifically designed for translations of that specific language.
Sort of like how chatgpt fails at chess, while Stockfish AI is world chess master.
I did some more digging, and I think Constitutional AI is more promising than I gave it credit for. When Claude had hidden goals, it still pursued the spirit of the constitution it was given.
Claude is also very good at coding, especially when it comes to more complex coding. It outperforms chatgpt for sure, even tho chatgpt is by far fastest coder.
China leads in AI
China leads in cheap technology.
The U.S. leads the world in AI investment, thanks to a robust ecosystem of startups, venture capital firms and the largest publicly traded tech companies. The U.S. accounted for $109 billion in corporate AI investments in 2024 alone, which is nearly as much as the rest of the world combined.
The USA is currently the No. 1 country in AI, thanks to foundation model breakthroughs, semiconductor dominance, enterprise AI maturity, and global research leadership.
China doesnt lead anywhere except in mass production. But due to population collapse, its questionable how long will China even be world factory of cheap goods.
The idea is you could look for weight combinations that pursue your goal and stumble upon weight combinations pursuing another goal, that appear to pursue your goal.
Yes, but I am saying extremely dangerous AI is unlikely, and more and more unlikely the more weights are adjusted for safety measures because that also changes "goal ai pursues".
I don't think it would be easy to check for hidden vulnerabilities or exploits though.
Checking code is easy, when you realize AI code isnt exactly long. its knowledge base which takes most of storage. In weights AI, weights take a lot of storage. Now, RAG AI doesnt use weights. It can still use vectors like today's AI, it can still do calculations like calculator to answer math questions, it can still retrieve knowledge effectively, and with good combination of vectors and sets logic, it could even be made able to reason in human language too in a way which is as good as current AI or even better if deduction guarantees precision.
Plus to get a RAG that is x intelligent, you need a black box AI that is at least x intelligent
Sort of. AI can make another AI. that is next step in evolution of AI. AI which reproduces itself or produces similar or even different AI. However, AI can also make AI which is better than it, in this specific case, AI with weights can make AI without weights.
the reason why I argue for working on AI without weights is because weights are unknown area, and we could really be making anything there, and also, we could be wasting a lot of resources on something which wont lead anywhere. Because, as I said before, reasoning using deduction sets is mathematically perfect. Reasoning built on training data is only as good as training data.
The black box AI just existing is dangerous because humans can be manipulated and computers can be hacked, and the technology existing means someone else may use it much less carefully.
What I simply think is that a different type of AI needs to be developed, because problem with current AI is that it is already too complicated. We have already reached a point where we cant understand it, and that is dangerous on its own.
We’ve succeeded in reducing the number and adding more safeguards.
the world is still developing nuclear weapons and means of delivery as fast as ever. What we reduced is number of old nuclear bombs. the new ICBMs are pretty much being developed on yearly basis, and those ICBMs are more dangerous than nuclear bombs themselves, because nuclear bomb is nothing without means of delivery.
But when it comes to AI, problem is, AI isnt something owned by only 9 countries's governments. AI is pretty much owned almost universally. I know a very advanced AI currently isnt owned by masses, but almost all advanced models are, and masses can even built their own AI.
And also, countries agreed to limit nuclear weapons because it was simply pointless to have more. they already have more than enough to wipe out planet 7 times.
Countries will never agree to limit AI, because any country which refuses to agree gains huge advantage over countries which agree, and this is true for all future researched helped by super AI.
Disclaimer: I did some more digging, and I think Constitutional AI is more promising than I gave it credit for. When Claude had hidden goals, it still pursued the spirit of the constitution it was given. So deceptive alignment may not be as hard a problem as I thought, at least based empirically on recent models. I don’t know how well this will generalize to smarter models though.
I agree with your point that there could be a super AI which is also evil in the future. In fact, I think it is inevitable. My argument was simply that we can make different types of AI which are easier to control, and also, linked AI models would introduce more control. And also, we are in a situation where if we dont advance AI, someone else eventually will, same with ICBMs. I think the more we advance AI, the more we will be able to understand it, and maybe good AI will even fight against bad AI and detect bad AI if it comes to worst.
Yes — but “better” and “bigger” improve AI in different ways, and one is usually more important than the other.
Let’s break it down clearly.
More data generally helps because:
Large models like OpenAI’s systems or Google DeepMind’s models are trained on massive datasets for this reason.
However…
If the data is:
Then making it bigger just means more noise.
More garbage ≠ smarter AI.
“Better” usually improves AI more efficiently than just “bigger.”
Better means:
High-quality curated datasets can outperform much larger low-quality ones.
For example:
Research from OpenAI shows performance improves smoothly with:
But the improvements follow diminishing returns.
After a point:
In practice, best results come from:
So the real formula is:
Large × Clean × Diverse × Well-trained
Not just “large.”
Since you’ve been building symbolic/deductive chatbots in Python:
For your system:
For symbolic systems, structure matters even more than size.
✔ Bigger training text helps.
✔ Better training text helps more per unit.
✔ The best systems use both.
If you want, I can also explain how this differs between:
Statistical probability is how AI learns.
AI starts as extremely dumb, unable to form one coherent sentence. then it learns from training data, it learns patterns which statistically repeat more, to calculate statistical probability of them.
Bad training data is usually biggest cause of failed AI. Feed AI garbage patterns, it learns garbage patterns.