The End of PAA (people also ask) automated sites is here, seems like google manually push them down

Captain bd said:
As if everyone knows how to train their own model?
Click to expand...

You're out of luck then :-)

Alma said:
GPT3 is the largest and most accurate model (davinci) and it cost them between $4.6m to $12m to train it. How can anyone here possibly build anything better? Not even talking about the salary the geniuses get. They didn't receive $1B investment for no reason.
Click to expand...

Nope, not really.

GPT3 is just one of the largest by parameters. Opt-66b is open source and is 1/3 the size. Opt-175b is available but you have to apply to get it. YaLM-100b is open source.

But, it's a misunderstanding of machine learning to say any model is the "most accurate". GPT3 is just the best for general tasks where you don't fine tune. gpt-neox-24b will outperform it when fine tuned on a task. I'm using a combination of open source models for question answering and the content is at least on par with GPT3 in terms of writing and accuracy wise is more factually accurate.

A larger base number of training parameters doesn't mean the fine tuned model will be better. This also is a misunderstanding in machine learning.

Fiducia said:
You are not creating a base model from scratch. @splishsplash is referring to training a model ON TOP of a base model. Simple rewriting tasks do not require a huge model like Davinci/NeoX. Instructions on how to do so can be found on the network in various places, there are also vendors offering access to GPT-J (and below) at monthly flat fees.
Click to expand...

Exactly. Creating your own base models could be advantageous at an advanced level, but you don't need to train gpt-j, gpt-neox, opt-66b, any of the bert models. You just load them and fine tune them on a data set then use them.

nadhull said:
Maybe, and hopefully, @splishsplash can direct us to some resources so we can do it properly (and cheaply).
Click to expand...

https://huggingface.co/course/chapter1/1
And to learn the required math and machine learning, a good place to start is https://brilliant.org/. It helps to have a good foundation in calculus, vector calculus and linear algebra.

This is a good resource too: https://machinelearningmastery.com/start-here/

Sartre said:
how do you build your own model? Do you mean fine-tuning something like GPT-J or NeoX?

I'm not saying it's not possible, just pretty hard without a PhD or hiring someone who know a lot about ML and these guys tend to charge quite a bit of money. Plus the GPU time.
Click to expand...

GPUs are cheap. It's about $3/hr for an A100 80GB

You don't need a Ph.D. That won't help you at all. A Ph.D is a research degree. You don't learn anything. University is slow as fuck. You spend 4 years dicking around.

You just need to learn calculus, vector calculus and linear algebra to understand machine learning.
 
Advertise on BHW
splishsplash said:
GPUs are cheap. It's about $3/hr for an A100 80GB

You don't need a Ph.D. That won't help you at all. A Ph.D is a research degree. You don't learn anything. University is slow as fuck. You spend 4 years dicking around.

You just need to learn calculus, vector calculus and linear algebra to understand machine learning.
Click to expand...

So you're saying you created your own model? Do you have a Ph.D?

My university wasn't slow by any means. We had cutting-edge technology and learned A LOT. Our tech was so advanced that companies like Sony and Warner Bros were renting out our robotics/audio/motion capture labs. My university had the fastest supercomputer in the country. If you wanted to learn, you had crazy opportunities.
 
Last edited:
Y
splishsplash said:
You're out of luck then :)



Nope, not really.

GPT3 is just one of the largest by parameters. Opt-66b is open source and is 1/3 the size. Opt-175b is available but you have to apply to get it. YaLM-100b is open source.

But, it's a misunderstanding of machine learning to say any model is the "most accurate". GPT3 is just the best for general tasks where you don't fine tune. gpt-neox-24b will outperform it when fine tuned on a task. I'm using a combination of open source models for question answering and the content is at least on par with GPT3 in terms of writing and accuracy wise is more factually accurate.

A larger base number of training parameters doesn't mean the fine tuned model will be better. This also is a misunderstanding in machine learning.



Exactly. Creating your own base models could be advantageous at an advanced level, but you don't need to train gpt-j, gpt-neox, opt-66b, any of the bert models. You just load them and fine tune them on a data set then use them.



https://huggingface.co/course/chapter1/1
And to learn the required math and machine learning, a good place to start is https://brilliant.org/. It helps to have a good foundation in calculus, vector calculus and linear algebra.

This is a good resource too: https://machinelearningmastery.com/start-here/



GPUs are cheap. It's about $3/hr for an A100 80GB

You don't need a Ph.D. That won't help you at all. A Ph.D is a research degree. You don't learn anything. University is slow as fuck. You spend 4 years dicking around.

You just need to learn calculus, vector calculus and linear algebra to understand machine learning.
Click to expand...
Yes i know algebra, math. But i am a civil engineer. Not computer Scientist. I know a bit html css but machine learning is next level shit for me
 
Last edited:
Sartre said:
So you're saying you created your own model? Do you have a Ph.D?

My university wasn't slow by any means. We had cutting-edge technology and learned A LOT. Our tech was so advanced that companies like Sony and Warner Bros were renting out our robotics/audio/motion capture labs. My university had the fastest supercomputer in the country. If you wanted to learn, you had crazy opportunities.
Click to expand...

When I say it's slow as fuck, the way you learn is dated and not good at all.

I don't have a Ph.D, no.

You don't need to create your own model. There are 1000's of models you can choose from. A model is only a file you load. It's re-usable after it's trained.

Only gpt3 is closed source.

You could setup your own opt-66b, right now. It's not easy of course, but you don't need 10 years machine learning experience and a Ph.D.

https://github.com/facebookresearch/metaseq/issues/146 talks about it. Someone there got it running on 8xA100 80GB's, so opt-66b would probably need 3.

That's about $9/hr, or $6.5k/mo to run continuously, which is NOTHING to have your own opt-66b model. You could create billions of words of content with that.

The problem with university is it just teaches you a bit of everything, and it does it slowly. It's a terrible way to learn.

Look at how Elon Musk learns.. He doesn't just go and blindly study or degrees.

He has a problem and he learns what he needs to solve that problem.

This is how the brain is designed to learn. Not just reading through books in order to pass exams and get an 'A'.

For example, you want to understand first of all, RNN's. Recurrent Neural Networks.

So you start reading about it. You won't understand much. Let's say you read https://www.simplilearn.com/tutorials/deep-learning-tutorial/rnn

So you get down to a bit that's talking about a "sigmoid function". So now go read https://machinelearningmastery.com/a-gentle-introduction-to-sigmoid-function/

if your math is really basic, then start with high school math and learn the core stuff, but you don't need to study ALL of university math. You only need to learn about certain things. Calculus, vectors, matrices, vector space, dot product, cross product, linear algebra. And you don't need to be able to fully understand it or write proofs. You just need to know enough to be able to wrap your head around what's going on and start to tweak things.

It's a far more efficient way to learn. University teaching is about being slow, deliberate and understanding everything in the material.

Real learning is about skimming and taking the bits you need, and moving on without understanding.

Read something, and understand it 2/10, then read something else and you now understand that other thing 3/10. It's an iterative process. Eventually your 2/10, turns to a 3/10, to a 4/10 and so on. University isn't like that. You have to master each part of the material and pass an exam.

If you learn some basic calculus you won't be able to pass an exam and get an A, but you can read a paper about a machine learning model and maybe grasp 10% more than than you could before. That's progress. That 10% increase in understanding could be enough for you to get a working solution to a problem.
 
splishsplash said:
When I say it's slow as fuck, the way you learn is dated and not good at all.

I don't have a Ph.D, no.

You don't need to create your own model. There are 1000's of models you can choose from. A model is only a file you load. It's re-usable after it's trained.

Only gpt3 is closed source.

You could setup your own opt-66b, right now. It's not easy of course, but you don't need 10 years machine learning experience and a Ph.D.

https://github.com/facebookresearch/metaseq/issues/146 talks about it. Someone there got it running on 8xA100 80GB's, so opt-66b would probably need 3.

That's about $9/hr, or $6.5k/mo to run continuously, which is NOTHING to have your own opt-66b model. You could create billions of words of content with that.

The problem with university is it just teaches you a bit of everything, and it does it slowly. It's a terrible way to learn.

Look at how Elon Musk learns.. He doesn't just go and blindly study or degrees.

He has a problem and he learns what he needs to solve that problem.

This is how the brain is designed to learn. Not just reading through books in order to pass exams and get an 'A'.

For example, you want to understand first of all, RNN's. Recurrent Neural Networks.

So you start reading about it. You won't understand much. Let's say you read https://www.simplilearn.com/tutorials/deep-learning-tutorial/rnn

So you get down to a bit that's talking about a "sigmoid function". So now go read https://machinelearningmastery.com/a-gentle-introduction-to-sigmoid-function/

if your math is really basic, then start with high school math and learn the core stuff, but you don't need to study ALL of university math. You only need to learn about certain things. Calculus, vectors, matrices, vector space, dot product, cross product, linear algebra. And you don't need to be able to fully understand it or write proofs. You just need to know enough to be able to wrap your head around what's going on and start to tweak things.

It's a far more efficient way to learn. University teaching is about being slow, deliberate and understanding everything in the material.

Real learning is about skimming and taking the bits you need, and moving on without understanding.

Read something, and understand it 2/10, then read something else and you now understand that other thing 3/10. It's an iterative process. Eventually your 2/10, turns to a 3/10, to a 4/10 and so on. University isn't like that. You have to master each part of the material and pass an exam.

If you learn some basic calculus you won't be able to pass an exam and get an A, but you can read a paper about a machine learning model and maybe grasp 10% more than than you could before. That's progress. That 10% increase in understanding could be enough for you to get a working solution to a problem.
Click to expand...
I agree with most of what you said here, except the Elon Musk stuff. He just has people who learn things for him. That's a bit different. He only learns about concepts and doesn't actually implement them. AFAIK.

On the other hand, I stand by what I said; my Ph.D. program was fantastic and wasn't taking a lot of my time either. I was both working on my own stuff on the side and learning/researching what I wanted at school. At the Ph.D. level, exams are a formality. You're already, more or less, an expert in your field and specialize. I didn't have the mindset to make millions in my 20s anyway. I doubt many people do, but if they have, I agree with taking the path that you described.

I know what a model is. Just wanted to clarify if you actually did it. I actually built a simple model, but in the end got access to a more advanced model and fine-tuned it for my app :)

Cheers!
 
So... while you guys are at it, did anyone noticed this recently among their paa/ai/paraphrased websites?
1662983589683.png
 
Sartre said:
I agree with most of what you said here, except the Elon Musk stuff. He just has people who learn things for him. That's a bit different. He only learns about concepts and doesn't actually implement them. AFAIK.

On the other hand, I stand by what I said; my Ph.D. program was fantastic and wasn't taking a lot of my time either. I was both working on my own stuff on the side and learning/researching what I wanted at school. At the Ph.D. level, exams are a formality. You're already, more or less, an expert in your field and specialize. I didn't have the mindset to make millions in my 20s anyway. I doubt many people do, but if they have, I agree with taking the path that you described.

I know what a model is. Just wanted to clarify if you actually did it. I actually built a simple model, but in the end got access to a more advanced model and fine-tuned it for my app :)

Cheers!
Click to expand...

Elon Musk is a genius. He has genius level intelligence. Proper genius level intelligence. Most people don't recognize it because it's so far above the norm.

Research him more. He actually understands the physics and engineering behind Space X.

Yes, I know what a model is. That's pretty basic.
 
I say dont loose hope. Try something new. Your stats are damn good.
 
IM Rider said:
If you got all questions in your subheadings it's a big big red flag.
Convert questions to simple headings.
What is x
To
x
And
Do not scrap content thats duplicate content. No matter how good you spin it.
Use an ai writer instead.
Click to expand...
Oh lord, I am manually writing articles to answer those questions. Guess I am doing it wrong.
 
Advertise on BHW
You must log in or register to reply here.
Sign up now!

Main Menu

Home Main Forum List Black Hat SEO White Hat SEO BHW Newbie Guide Blogging Black Hat Tools Social Media Downloads

Marketplace

Account Selling Content / Copywriting EBooks / Methods Hosting Misc Proxies For Sale SEO - Link building SEO - Packages Social Media Social Media - Panels Web Design More

Making Money

Affiliate Programs Hire a Freelancer IM Journeys Making Money Pay Per Click Site Flipping Want to Buy

BlackHat World

Dispute Resolution Forum Suggestions & Feedback Introductions Lounge
Back
Top