SatanLucy

It's training data consists of sentences, not just words.

And training data itself was made to a great degree by Chatgpt.

Tho I can't really give it too big training data. My laptop is slow and training takes forever if data is too big.

New iphone costs 600$.

My Xiaomi as new costs 150$.

So yeah, I get extra 450$.

I am not woke but I disagree a lot with MAGA movement. Trump is a rapist pig etc. Big problems, he's a sellout to Israel.

You disagree a lot with everyone. There literally never was a person or religion or ideology you didn't disagree with.

Better question would be what you actually agree with.

Apparently, you have not read the Bible cover-to-cover.

Maybe that person didn't, but I read a lot of Bible, and yeah, it's a waste of time reading it. Other than fun gospels, it's mostly a boring anti trans anti gay book which advices you to poke out your eyes and cut off your arms so you dont sin.

If some book advices me to do self harm so I don't commit any of what the book sees as sin, and it sees a lot of stuff as sin, I would say that book is nonsense.

Also, the book seems to advice people to live in poverty.

So yeah, be a poor virgin who never looks at women, and you go to heaven!!!

If those are conditions for heaven, one must wonder what are conditions in heaven.

Subword level tokenizer is a popular choice, but I won't limit myself on that. Both word level and character level have their advantages. Biggest advantage of character level is that it makes AI smaller.

Yeah, it's difficult to sell things you like, cuz then you don't have them anymore. I am not sure I can ever sell my manga collection, even if I need the money.

I am actually trying to decide if I should use word level tokenizer or character level tokenizer.

Character level tokenizer produces a bit smarter AI and of smaller size.

Word level tokenizer, on the other hand, has 5x faster training.

Yes, around 120 manga in total, along with that big Sailor moon manga I got on Temu.

Could MAGA be a product of that research?

Possible, yes.

I am currently working on two types of chatbot.

In past, I made one which acts as search engine for sentences from text. that one worked fine, but no creative abilities.

then I got interested in markov chain, however markov chain faces a serious problem. If you set chain too long, chatbot loses all creativity. If you set it too short, it has no coherence. And middle doesnt work either, because its always either not one bit creative or mostly incoherent.

then I realized a different approach: Making character level markov chain with randomness. Chain randomly selects number between 7 and 90. that becomes the character length of chain predicting the next character. It repeats for each character. With this form of randomness, chatbot is in most cases coherent, while sometimes getting creative. So it both produces something new (coherent or not), and it is coherent in most sentences because in most cases it sticks to data text. Only sometimes, by random chance, it becomes creative. So I get both in one.

Second chatbot is AI type. So far it seems to be working very good. Character level tokenizer proved to be success. It makes AI model significantly smaller (like 50 megabytes), the batch size of 32 or less enables more effective learning during training, and context window of 128 is enough for small AI. Increasing context window makes it smarter, but also slower. However, it doesnt need big context window cuz its not doing coding which requires huge context.

I am also considering if I should use float 32 or float 16. Float 32 is biggest quality, but also biggest size.

So far what I saw is that smaller batch size in training makes AI smarter, but it slows down training a bit, but it gives more per training.

Context window size when increased, makes AI smarter, but also slows down training and generation.

So my goal is to keep batch size at 64 or 32. Maybe even 16 later, because it's good way to make smaller AI smarter.

As for context window, my laptop can only run so much. I don't think I can increase it too much.

Context window increase and batch size decrease are probably best ways to make AI smarter, while not increasing it's size. Compared to that, increasing parameters increases size always.

I will. My goal is to develop an AI model which is smallest AI model in the world, yet also smart. this one was 37 megabytes, but I think it can be further reduced. the only problem with small AI is that the smaller it gets, the less text in can learn from its training data.

Its only 37 megabytes size, lmao

Alright, here are first results of my small AI

Alright, I cant tell if that opening line was some joke from its training text. tho it seems surprisingly coherent and.... rhyming?

I knew that if I keep squeezing my brain, eventually it will poop out something good.

Biggest problem with modern AI is their size.

Biggest factor of their size is vocabulary.

What is vocabulary of a chat bot?

In past, it used to be sentence level. Bot would just memorize whole sentences.

then they upgraded it to word level.

However, word level wasnt suitable for modern AI because it makes model huge.

Model size is measured by vocabulary x other parameters.

So if they used word level, then because english language has over 1 million words, it would make model 1 million times bigger.

So they decided to use subword level. Subword level has vocabulary size of 30000 to 350k.

So subword level makes model "only" 30000 times bigger at least, compared to 1 million times bigger in case of word level.

But then it occurred to me.

Smaller vocabulary means smaller AI model, when other parameters are same.

Smallest possible vocabulary is not subcharacter level.

Smallest possible vocabulary is character level!!!

there are only about 70 to 100 characters in English language. Compared to subword level, it is at least 300 times smaller vocabulary size, and thus 300 times smaller model.

I am currently working on making a very small AI which uses character level tokenizer to create equally effective but smaller AI.

Show me specifically where and when I've said people cannot have different opinions?

You literally think people can't choose a different definition than yours.

No, for you its all about "me, me, me, my definition is the only right one".

I am actually the only one here who thinks people can have different opinions. You are somehow not even aware of that.

Because you say so? Yes, you can certainly believe it, but it does not make it so

the fact that I can believe so already proves my point. If I can reject your definition and believe in a different one, then its a choice. Lack of choice would be people unable to produce a different definition and believe in it.

But I guess this simple logic is too complicated for you.

And I have told you that word definition is NOT a matter of choice

Of course it is. I can reject any definition you put forward. I can use different definitions. the guy who came up with word "female" was the first one to choose a definition for it.

My phone is Xiaomi Redmi Note 14 4g.

By all standards, it's barely a mid-range phone.

So I didn't really expect to be able to play most demanding games on it.

However, there seems to be a very simple way to play even most demanding games out there.

I tested "where winds meet", which is probably one of most demanding games one can install on smartphone.

However, the process to play it effectively has 3 steps:

Step 1 is first clearing RAM, using that app on phone to free up RAM memory.

Step 2 is to download a game booster from play store. Some are better than others, tho the one I downloaded seemed to have options to set fps, graphics, and give boost. So I just use all options that game booster gives.

Step 3 is to open phone's own gaming mode, and start a game from there.

When I follow these 3 steps, I pretty much get about 25 to 30 fps on where winds meet when I play on maximum graphics settings.

If I lower graphics, I get more fps.

This seems like a very simple way to get high fps. And it only takes a minute or two to do before starting a game.

On laptop, it's a bit different. There I use overclock GPU, game booster and RAM cleaner.

I mean, why would he? He was treated at a hospital, and they fixed his little self-cut and blood was washed away. Most of people believed it was real. The goal was achieved.

He had no need to overdo it days later, nor could he realistically predict some people will question why there wasn't more blood or a scar.

But it was already sold to the masses as real.

It is only few people who really noticed that his injury looks exactly like injury you would get if you cut your own ear intentionally. Not much blood, no scar, and only a small cut line.

Also, in video of shooting, you can see Trump making a cutting motion on his ear with his hand.

His hand was not open. It appeared as if he was holding something when he actually was reaching to touch his ear.

If you zoom in on this photo, it's quite clear the injury is on upper front side of ear. Bullet travels in a straight line, which begs the question how did it only cut front side of ear, and not the rest of it which was also in bullet's path?

It doesn't seem like a bullet injury at all.

Trump's ears aren't perfectly flat. They are actually a bit sticking out. So a bullet which hit upper front side of ear also had to hit the other parts, yet it didn't.

Yeah, so he could have cut his ear to keep it looking injured. Not sure why that would be difficult.

He had a very short amount of time to cut his ear, and bunch of cameras were there. He pretty much only could have cut it once, twice at maximum.

One cut or two cuts don't produce a lot of blood or injury.

The blade he used also had to be small enough to not be seen.

Also, cutting to create fake injury is a common practice in WWE, something Trump just happens to be a fan of.

To me, the injury he had doesn't seem like injury from a bullet to the ear. Person would bleed in that same way by cutting own ear once or twice. But chances of bullet making exactly same injury are a lot lower. Bullet would probably rip off a good part of ear.

The fact that Trump has no any big scar, just tells me it's probably not bullet injury.

To be honest, I don't even know what your original post was even trying to say.

Woke 1 plus woke 2 metastasis?

What the fuck am I supposed to respond to that?

Okay, so you don't like woke. We get it.

They could injure his ear for real though and keep injuring it if it heals. Or injure his arm or something

Keep in mind that it was Trump who probably led that whole stage show.

If it wasn't Trump, then it was someone willing to risk Trump's life to improve Trump's position in politics.

Injuring ear by sniper shot is risky. Much easier to just use fake blood or give himself a small cut on ear by hidden razor blade.

So if it was Trump who led the show, injury on ear was fake blood or by cut (more likely by cut).

If it was someone else, then it is real injury by bullet, but bullet was meant to hit the ear.

The only other option is that somehow shooter had intention to kill Trump and yet somehow out of 3 shots, only managed to scratch an ear. Talk about poor aiming...

I could go on and on, but really, this guy thinks word "woman" wasnt made up. He thinks they found that word in nature or what? Its honestly amazing that someone cant figure out something so simple.

You have ignored what I have also said, and you seem to ignore it:

We cannot choose the definition of words unless it can be demonstrated that a new definition applies.

Nothing was ignored. It was already explained to you that definition of each word is a choice. If your brain cant process that simple truism, I dont know what to tell you.

Definitions arent demonstrated. they are made up.