Claude released a new model Sonnet 3.7

[splishsplash]

Understood. I use it to supplement keywords for on-page optimization. The optimization tool suggests which keywords should be included, and I currently use GPT o3 to add them. I review the output to check whether the placement of the added keywords is reasonable. I'm not sure if this approach is incorrect.

I have read your entire guide, but I’m still unsure how to fine-tune the process to make my workflow more efficient. I'm trying to see if I can streamline it and use programming for semi-automation; otherwise, manually optimizing 5 to 10 articles per day is my limit.

You previously advised against publishing more than 10 articles per day. Does this also apply to non-English markets? Thank you.

It can certainly hallucinate you keywords, but that's about the best it can do. I say hallucinate, because, for example, if you ask it

"What's the capital of France?"

It will not hallucinate. It will give you an accurate answer. It will give you an accurate answer for a LOT of concepts.

An LLM is only pattern matching. Nothing more. It's a very very sophisticated pattern matching machine.

That's why it can do tasks, where pattern matching is possible.

( As a side note: The list of tasks where pattern matching is possible is actually massive. "general intelligence" isn't one of them, but a huge amount of things a general intelligence can accomplish, can be accomplished with pattern matching.
machine learning is extremely useful
)

What I will also say that's important, is that it's possible as a thought experiment to create an LLM that is capable of solving every conceivable problem in this universe.

How?

Doesn't this contradict what I said earlier?

No, because, you can simulate general intelligence to such an extent that it's indistinguishable from next token prediction if you have infinite training data and infinite compute.

This is ONLY a thought experiment.

But, you can prove it to yourself by looking at how a general intelligence solves a very simple problem.

"how many letters are in this sentence?"


This is something an LLM will hallucinate. It is NOT answering the question. It is not reasoning. It does not understand the question. It will just predict the next words that should come after this.

If we ask gpt 4o, it says



You see how confident it is?

"Let's count the letters in your sentence" -- My ass. This is just token prediction.

It's predicting tokens that make it look like it's thinking. It isn't thinking. It isn't aware. It isn't conscious.

And it's wrong. It hallucinated.

And it doesn't even know it's wrong. You don't know it's wrong either unless you can check, but if you're using an "AI", why would you check? Do you even have the capability to check in most cases? No, you will just blindly assume what it says is correct. Hence the dangers of using it for things like SEO.

Now how would a general intelligence solve this?

As a general intelligence myself, I can answer that

First, I read the question, then I ask myself(Ask myself, what does that even mean? I seem to be able to not only ask myself something, but observe "myself" asking "myself" for something. Is there 3 entities here? One observing, one asking and one answering?)

I then have the thought "What exactly is the question asking?"

Or maybe I don't have that thought. Maybe I've done this so many times I myself(Who am I? Who is this I?) recognize it and can proceed to answer it more quickly.

I then have the thought "The question is asking how many letters there are in the sentence. That doesn't include spaces or punctuation."

Then I have the thought: "How can I solve this?"

Then I have the thought: "I can point at each letter with my finger or track with my eye and keep a count."

What tools do I need for this? Working memory to keep a count, and eyes capable of tracking distinctive letters.

I can then do it, and then afterwards, I have the thought: I should check this a second time to make sure it's correct, and perhaps go a little more slowly.

I can also have the thought, maybe I'll get a pen and paper and write this out, and under each word write the count, then add them up to make sure I'm more accurate.


Tools needed: pen, paper and a hand to write.

Job done.

I came up with that blueprint, as a general intelligence.

If you feed that blueprint in the form of thinking tokens to an LLM, and you give that LLM tools, it will be able to come to the same answer as me.

You can literally give it thinking tokens in the english language that represent my(a general intelligence) thoughts.

If you can do that for 1 problem, then in the thought experiment, you can see how you can expand this to include every possible problem in the universe.

As long as the thoughts can be described with the language the LLM understands and you give it access to the required tools, it can simulate a general intelligence and do any task.

And this isn't like writing a program. This thinking blueprint allows it to solve that problem and millions of similar problems.

Potentially not just counting letters, but it could be counting leaves on a tree. If it can pattern match, and it can, then it can pattern match that thinking blueprint to something like "How many leaves are on this tree?" assuming you can give it tools to 'see' and interact with the tree.

To answer your specific questions:

I have read your entire guide, but I’m still unsure how to fine-tune the process to make my workflow more efficient. I'm trying to see if I can streamline it and use programming for semi-automation; otherwise, manually optimizing 5 to 10 articles per day is my limit.

You previously advised against publishing more than 10 articles per day. Does this also apply to non-English markets? Thank you.

That's not really easy to answer in a post. I have written code that does this, but it's a long and complicated process and includes multiple fine-tuned models. There isn't a simple "this is how you do it" answer.

non-English markets I don't know. Test it. Everything should be tested anyway as the landscape is ever-evolving.

View attachment 424854

Cool.

You're welcome to actually debate the specific points I'm making about why I don't believe an LLM is capable of general intelligence or make your own points as to why you think we aren't in an AI bubble and that LLMs really will take our jobs.

Anyone can post a cool image from the matrix, but it doesn't prove or disprove anything.

Edit: You've clicked the "haha" button around 30 seconds after I posted this. You obviously have no interest in actually even reading my arguments let alone engaging in debate, but just want to laugh and share silly memes to feel cool.

 

[koxokot890]

It can certainly hallucinate you keywords, but that's about the best it can do. I say hallucinate, because, for example, if you ask it

"What's the capital of France?"

It will not hallucinate. It will give you an accurate answer. It will give you an accurate answer for a LOT of concepts.

An LLM is only pattern matching. Nothing more. It's a very very sophisticated pattern matching machine.
...

Your brain is just a pattern matching machine, there's nothing magical about it; if you have evidence to the contrary, I think many of us would be delighted to review it.

Aircraft, despite being less complex than birds, outperform them in their respective environments. It's likely that AI will soon make humans intellectually obsolete. Currently, AI output is conditioned by user input, so what I read are projections of your current limitations when interacting with these models.

Maybe it's something that doesn't make you feel special, which will make you seek out evidence that confirms your beliefs.

 

[Danki]

In the writing part, it's seems waaay better than Opus 3. Really satisfied with this new model.

 

[Hank Tzeng]

It can certainly hallucinate you keywords, but that's about the best it can do. I say hallucinate, because, for example, if you ask it

"What's the capital of France?"

It will not hallucinate. It will give you an accurate answer. It will give you an accurate answer for a LOT of concepts.

An LLM is only pattern matching. Nothing more. It's a very very sophisticated pattern matching machine.

That's why it can do tasks, where pattern matching is possible.

( As a side note: The list of tasks where pattern matching is possible is actually massive. "general intelligence" isn't one of them, but a huge amount of things a general intelligence can accomplish, can be accomplished with pattern matching.
machine learning is extremely useful
)

What I will also say that's important, is that it's possible as a thought experiment to create an LLM that is capable of solving every conceivable problem in this universe.

How?

Doesn't this contradict what I said earlier?

No, because, you can simulate general intelligence to such an extent that it's indistinguishable from next token prediction if you have infinite training data and infinite compute.

This is ONLY a thought experiment.

But, you can prove it to yourself by looking at how a general intelligence solves a very simple problem.

"how many letters are in this sentence?"


This is something an LLM will hallucinate. It is NOT answering the question. It is not reasoning. It does not understand the question. It will just predict the next words that should come after this.

If we ask gpt 4o, it says

View attachment 424877

You see how confident it is?

"Let's count the letters in your sentence" -- My ass. This is just token prediction.

It's predicting tokens that make it look like it's thinking. It isn't thinking. It isn't aware. It isn't conscious.

And it's wrong. It hallucinated.

And it doesn't even know it's wrong. You don't know it's wrong either unless you can check, but if you're using an "AI", why would you check? Do you even have the capability to check in most cases? No, you will just blindly assume what it says is correct. Hence the dangers of using it for things like SEO.

Now how would a general intelligence solve this?

As a general intelligence myself, I can answer that

First, I read the question, then I ask myself(Ask myself, what does that even mean? I seem to be able to not only ask myself something, but observe "myself" asking "myself" for something. Is there 3 entities here? One observing, one asking and one answering?)

I then have the thought "What exactly is the question asking?"

Or maybe I don't have that thought. Maybe I've done this so many times I myself(Who am I? Who is this I?) recognize it and can proceed to answer it more quickly.

I then have the thought "The question is asking how many letters there are in the sentence. That doesn't include spaces or punctuation."

Then I have the thought: "How can I solve this?"

Then I have the thought: "I can point at each letter with my finger or track with my eye and keep a count."

What tools do I need for this? Working memory to keep a count, and eyes capable of tracking distinctive letters.

I can then do it, and then afterwards, I have the thought: I should check this a second time to make sure it's correct, and perhaps go a little more slowly.

I can also have the thought, maybe I'll get a pen and paper and write this out, and under each word write the count, then add them up to make sure I'm more accurate.


Tools needed: pen, paper and a hand to write.

Job done.

I came up with that blueprint, as a general intelligence.

If you feed that blueprint in the form of thinking tokens to an LLM, and you give that LLM tools, it will be able to come to the same answer as me.

You can literally give it thinking tokens in the english language that represent my(a general intelligence) thoughts.

If you can do that for 1 problem, then in the thought experiment, you can see how you can expand this to include every possible problem in the universe.

As long as the thoughts can be described with the language the LLM understands and you give it access to the required tools, it can simulate a general intelligence and do any task.

And this isn't like writing a program. This thinking blueprint allows it to solve that problem and millions of similar problems.

Potentially not just counting letters, but it could be counting leaves on a tree. If it can pattern match, and it can, then it can pattern match that thinking blueprint to something like "How many leaves are on this tree?" assuming you can give it tools to 'see' and interact with the tree.

To answer your specific questions:



That's not really easy to answer in a post. I have written code that does this, but it's a long and complicated process and includes multiple fine-tuned models. There isn't a simple "this is how you do it" answer.

non-English markets I don't know. Test it. Everything should be tested anyway as the landscape is ever-evolving.



Cool.

You're welcome to actually debate the specific points I'm making about why I don't believe an LLM is capable of general intelligence or make your own points as to why you think we aren't in an AI bubble and that LLMs really will take our jobs.

Anyone can post a cool image from the matrix, but it doesn't prove or disprove anything.

Edit: You've clicked the "haha" button around 30 seconds after I posted this. You obviously have no interest in actually even reading my arguments let alone engaging in debate, but just want to laugh and share silly memes to feel cool.

Your explanation made me understand instantly. As you said, using it for SEO may lead to potential errors. I will try to fine-tune the models to meet the requirements and avoid hallucinations. Thank you so much for answering my question and clarifying some concepts—I feel enlightened now.

 

[splishsplash]

Your explanation made me understand instantly. As you said, using it for SEO may lead to potential errors. I will try to fine-tune the models to meet the requirements and avoid hallucinations. Thank you so much for answering my question and clarifying some concepts—I feel enlightened now.

Yep, that's why fine-tuning is crucial. You are teaching it new patterns for the specific task you want.

An approach you can take is to programmatically analyse and break down existing pages and create training data from them. Both the question and and the correct answer, where the correct answer is an answer you've created from real data from live rankings.

The LLM can then potentially extrapolate out to other keywords.

Your brain is just a pattern matching machine, there's nothing magical about it; if you have evidence to the contrary, I think many of us would be delighted to review it.

Aircraft, despite being less complex than birds, outperform them in their respective environments. It's likely that AI will soon make humans intellectually obsolete. Currently, AI output is conditioned by user input, so what I read are projections of your current limitations when interacting with these models.

Maybe it's something that doesn't make you feel special, which will make you seek out evidence that confirms your beliefs.

View attachment 424883

Your meme is ironic, because it describes exactly what you're doing. You have faith in LLMs being able to make humans obsolete.

I wonder, do you even understand the calculus behind gradient descent?

So many people consider themselves experts on AI because they've watched some youtube videos about the coming AI eutopia. ;-)

Your aircraft analogy is completely flawed.

You say they "outperform" birds.

It's a highly simplistic analogy that completely relies on your subjective definition of "outperform".

Birds have more manoeuvrability, much higher efficiency and are more adaptable.
Planes are faster and at higher altitude.

As an analogy comparing a biological general intelligence to an LLM, it's just entirely useless.

One other thing I will also say is that, I'm not alone in my assessment of LLMs.

You dismiss it with a few words like it's the ramblings of an idiot, however, you can look up Yann LeCun's interview where he talks about how LLMs cannot achieve AGI.

Roger Penrose also doesn't believe we can achieve AGI through computation and talks about godel's theorem. You will be unlikely to understand that however, and will likely dismiss Roger as an idiot.

There are multiple people who see AI as a bubble.

https://www.ru.nl/en/research/research-news/dont-believe-the-hype-agi-is-far-from-inevitable
https://iaee.substack.com/p/agi-is-not-possible-8647257fb65d

If you actually read my arguments instead of just replying with silly memes, you would actually realize I'm saying that a general intelligence with computation is not possible, but it can simulate it with enough data and computation.

Instead you reduce my argument and attack me(ad hominem) implying I am some moron engaging in "blind faith".

You further attack me by saying it's because I want to feel special.

You say the brain is just a pattern matching machine and there's nothing special about it. This is an incredibly sad outlook to have as a human.

I see myself and all humans as beautiful conscious pieces of creation. The universe being aware.

You see humans as pattern matching machines.

Where are you even getting this belief? Not even neuroscientists claim the brain is just a pattern matching machine.

You are literally saying, although I don't think you realize it, that humans have no higher-order reasoning.

A pattern matching machine cannot create novel abstract ideas. It is not creative. It cannot engage in self-reflection.

Our brains CONTAIN pattern matching abilities. No one disputes that.

But this is the problem with people who do not engage in deep study of a field and then try to debate you on it. You have a surface level knowledge and are making statements that not even mainstream science is making.

NO ONE denies the brain can engage in abstraction, higher-order reasoning, creativity and self-reflection.

Do you not realize you are aware? Jesus christ. You can sit there right now and observe your thoughts. You can watch them. What is watching them?

That's the question that's debated.

Is there some physical part of the brain that's responsible for this awareness?

Does it arise from quantum processes in the brain?

There is not a single theory of the brain that says: The brain is only a pattern matching machine. It is not special.

Pattern matching machine CANNOT REASON. You are saying that humans cannot reason. Do you see how ridiculous that is?

Of course, you would never admit to that. You want to be right. It's unlikely you will even read what i'm saying let alone understand it. I am in fact wasting my time. Or maybe not, since I'm just a pattern matching machine that next token predicted all this. I'm not actually thinking.

 

[koxokot890]

...

Your meme is ironic, because it describes exactly what you're doing. You have faith in LLMs being able to make humans obsolete.

I wonder, do you even understand the calculus behind gradient descent?

So many people consider themselves experts on AI because they've watched some youtube videos about the coming AI eutopia. ;-)

Your aircraft analogy is completely flawed.

You say they "outperform" birds.

It's a highly simplistic analogy that completely relies on your subjective definition of "outperform".

Birds have more manoeuvrability, much higher efficiency and are more adaptable.
Planes are faster and at higher altitude.

As an analogy comparing a biological general intelligence to an LLM, it's just entirely useless.

One other thing I will also say is that, I'm not alone in my assessment of LLMs.

You dismiss it with a few words like it's the ramblings of an idiot, however, you can look up Yann LeCun's interview where he talks about how LLMs cannot achieve AGI.

Roger Penrose also doesn't believe we can achieve AGI through computation and talks about godel's theorem. You will be unlikely to understand that however, and will likely dismiss Roger as an idiot.

There are multiple people who see AI as a bubble.

https://www.ru.nl/en/research/research-news/dont-believe-the-hype-agi-is-far-from-inevitable
https://iaee.substack.com/p/agi-is-not-possible-8647257fb65d

If you actually read my arguments instead of just replying with silly memes, you would actually realize I'm saying that a general intelligence with computation is not possible, but it can simulate it with enough data and computation.

Instead you reduce my argument and attack me(ad hominem) implying I am some moron engaging in "blind faith".

You further attack me by saying it's because I want to feel special.

You say the brain is just a pattern matching machine and there's nothing special about it. This is an incredibly sad outlook to have as a human.

I see myself and all humans as beautiful conscious pieces of creation. The universe being aware.

You see humans as pattern matching machines.

Where are you even getting this belief? Not even neuroscientists claim the brain is just a pattern matching machine.

You are literally saying, although I don't think you realize it, that humans have no higher-order reasoning.

A pattern matching machine cannot create novel abstract ideas. It is not creative. It cannot engage in self-reflection.

Our brains CONTAIN pattern matching abilities. No one disputes that.

But this is the problem with people who do not engage in deep study of a field and then try to debate you on it. You have a surface level knowledge and are making statements that not even mainstream science is making.

NO ONE denies the brain can engage in abstraction, higher-order reasoning, creativity and self-reflection.

Do you not realize you are aware? Jesus christ. You can sit there right now and observe your thoughts. You can watch them. What is watching them?

That's the question that's debated.

Is there some physical part of the brain that's responsible for this awareness?

Does it arise from quantum processes in the brain?

There is not a single theory of the brain that says: The brain is only a pattern matching machine. It is not special.

Pattern matching machine CANNOT REASON. You are saying that humans cannot reason. Do you see how ridiculous that is?

Of course, you would never admit to that. You want to be right. It's unlikely you will even read what i'm saying let alone understand it. I am in fact wasting my time. Or maybe not, since I'm just a pattern matching machine that next token predicted all this. I'm not actually thinking.

Faith seems absurd to me; I have replicable evidence that a current language model can make certain human tasks obsolete. I'm not interested in sharing that evidence for the same reason I wouldn't share a money-making machine. Nor am I very interested in shedding light on the topic because it's not my intention to educate or convince you of anything. I even respond purely for entertainment. Science doesn't work by creating counterarguments without evidence.

And yes, the argument about birds is quite simplified; I didn't elaborate because I'm not interested in creating an extensive message. Airplanes are fast and far exceed birds in certain tasks; they were designed with a purpose in mind. Additionally, I'm talking about commercial airplanes. Adding more complexity here seems unnecessary to me; I'm not going to list all the types of aircraft that exist and all the types of birds seen to date.

If you want to play the authority fallacy game, I could cite Geoffrey Hinton, the 'Godfather of AI'.

There are people who believe AI is a bubble, just as there are people who believe Bitcoin is the future, and there are those who believe humans are special. After all, Yann LeCun must feel special. AI won't need any of that to achieve its goals, just like airplanes don't need evolutionary vestiges of birds to achieve the purpose for which they were created.

A significant portion of scientists still believe in the 'Santa Claus for adults', if I'm not mistaken, nearly 40%. Relying on their opinions to confirm your beliefs seems like a reasonable heuristic approach. The more challenging approach would be to read 20 neuroscience books by different authors and form your own conclusions. Yet, from the outset, you're approaching this with mystical thoughts about consciousness and other pseudoscientific topics simply because someone like X said so. I wouldn't be surprised if you tried to fit the pieces to suit your perspective.

The brain is a prediction machine, and while its algorithm can be complex for some humans to understand, it's relatively straightforward for machines. You don't even need AI to impose restrictions on you and create the illusion of decision-making.

I don't have data on this clone called SplishSplash, but I'm sure some corporations do. Once all that data can be accessed centrally, it will be easy to manipulate you by pressing the right buttons. I sense that you're searching for a theory that confirms your beliefs. There's nothing to be ashamed of about it.

Everything you refer to as creativity, etc., is not special; it's just patterns; reasoning is just a process. I could go on about marketing, business, economy, biology, bla bla and cite books, papers, and so on, but none of that will alter the ultimate fate of humanity. It's just an unnecessary waste of energy.

Science has advanced so much, but magical thinking exists and will always exist; Santa Claus and his variants will never go away.

I suppose denial must be comforting. The things installed in your head will defend themselves against any external threat that contradicts the rules you've inherited and learned since you began receiving input from the moment you emerged from what you believe is your mother.

 

[splishsplash]

Faith seems absurd to me; I have replicable evidence that a current language model can make certain human tasks obsolete. I'm not interested in sharing that evidence for the same reason I wouldn't share a money-making machine. Nor am I very interested in shedding light on the topic because it's not my intention to educate or convince you of anything. I even respond purely for entertainment. Science doesn't work by creating counterarguments without evidence.

And yes, the argument about birds is quite simplified; I didn't elaborate because I'm not interested in creating an extensive message. Airplanes are fast and far exceed birds in certain tasks; they were designed with a purpose in mind. Additionally, I'm talking about commercial airplanes. Adding more complexity here seems unnecessary to me; I'm not going to list all the types of aircraft that exist and all the types of birds seen to date.

If you want to play the authority fallacy game, I could cite Geoffrey Hinton, the 'Godfather of AI'.

There are people who believe AI is a bubble, just as there are people who believe Bitcoin is the future, and there are those who believe humans are special. After all, Yann LeCun must feel special. AI won't need any of that to achieve its goals, just like airplanes don't need evolutionary vestiges of birds to achieve the purpose for which they were created.

A significant portion of scientists still believe in the 'Santa Claus for adults', if I'm not mistaken, nearly 40%. Relying on their opinions to confirm your beliefs seems like a reasonable heuristic approach. The more challenging approach would be to read 20 neuroscience books by different authors and form your own conclusions. Yet, from the outset, you're approaching this with mystical thoughts about consciousness and other pseudoscientific topics simply because someone like X said so. I wouldn't be surprised if you tried to fit the pieces to suit your perspective.

The brain is a prediction machine, and while its algorithm can be complex for some humans to understand, it's relatively straightforward for machines. You don't even need AI to impose restrictions on you and create the illusion of decision-making.

I don't have data on this clone called SplishSplash, but I'm sure some corporations do. Once all that data can be accessed centrally, it will be easy to manipulate you by pressing the right buttons. I sense that you're searching for a theory that confirms your beliefs. There's nothing to be ashamed of about it.

Everything you refer to as creativity, etc., is not special; it's just patterns; reasoning is just a process. I could go on about marketing, business, economy, biology, bla bla and cite books, papers, and so on, but none of that will alter the ultimate fate of humanity. It's just an unnecessary waste of energy.

Science has advanced so much, but magical thinking exists and will always exist; Santa Claus and his variants will never go away.

I suppose denial must be comforting. The things installed in your head will defend themselves against any external threat that contradicts the rules you've inherited and learned since you began receiving input from the moment you emerged from what you believe is your mother.

View attachment 424902

You do not understand what pattern matching is.

Reasoning is not pattern matching. Go and ask your AI god chatgpt if reasoning is pattern matching. It will explain to you the difference.

You cannot logically say that humans have the ability to reason, while at the same time saying the brain is a pattern matching algorithm.

Nothing more needs to be said. This is a fundamental flaw. You can come back with as many memes and accusations of me being a clone as you like, but it doesn't change the flaw in your logic.

You are also wrong with the statement "I sense that you're searching for a theory that confirms your beliefs. There's nothing to be ashamed of about it."

I used to think LLMs would lead to AGI, but after 2 years of studying machine learning, linear algebra, calculus and understanding more about how the brain works, I've realized that my previous opinion was flawed. I have a much deeper understanding of how LLMs work now, so I changed my mind.

And if your core belief is that the brain is just an algorithm, then neither you or me are real.

Your statement "There's nothing to be ashamed of about it" is not real. It is a next token prediction from pattern matching. You are not a person. You are not alive. You have no thought, and you cannot observe your thoughts or change your thoughts. You are simply a machine running a pattern matching algorithm.

Therefore, why do you keep pattern matching more? Why do you say "Faith seems absurd to me". You don't believe you exist. You don't believe there is 'you' for anything to feel absurd.

And the notion you have "replicable evidence" that an LLM can make all human tasks obsolete is incredibly unbelievable since no company on earth is even close to that.

You just have to go on X to find people hitting walls with claude 3.7's ability to do simple real world coding tasks, yet, an LLM is going to make humans obsolete. ok.

P.S. I'll bookmark this thread and reply to you in 5 years, and you can let me know then if you still believe humans are just pattern matching algorithms.

 

[koxokot890]

...
And the notion you have "replicable evidence" that an LLM can make all human tasks obsolete is incredibly unbelievable since no company on earth is even close to that.
...

Fix your mistake, wonky algorithm. Even reading is simply the process of predicting the next token. Now that you see the error, you will correct it. There is no such thing as reading; it is the same process itself.

There isn't even such a thing as a brain, just another token in the chain.

 

[splishsplash]

Fix your mistake, wonky algorithm. Even reading is simply the process of predicting the next token. Now that you see the error, you will correct it. There is no such thing as reading; it is the same process itself.

There isn't even such a thing as a brain, just another token in the chain.

View attachment 424911

Btw, when you call people things like "wonky algorithm", it doesn't have the effect you think it does. It just makes you look like a small angry child who is probably still living in his mother's basement. Maybe just keep to the points and let's leave the nonsense aside.

Yes, I see now you said "certain human tasks". That makes the entire statement useless.

LLMs already automate certain human tasks. Great, you have evidence they can do something that we know they can do right now.

I have an AI SaaS that does this very thing. What they won't do is replace humans.

P.S. The constant matrix memes also just highlight your immaturity. I know you *think* it makes you look really smart, but it really doesn't.

 

[koxokot890]

Btw, when you call people things like "wonky algorithm", it doesn't have the effect you think it does. It just makes you look like a small angry child who is probably still living in his mother's basement. Maybe just keep to the points and let's leave the nonsense aside.

Yes, I see now you said "certain human tasks". That makes the entire statement useless.

LLMs already automate certain human tasks. Great, you have evidence they can do something that we know they can do right now.

I have an AI SaaS that does this very thing. What they won't do is replace humans.

As long as there is room for error correction, I will be a wonky algorithm. I don't aim to build a reputation in this virtual environment, so the next token is built based on my goal, which is to laugh at myself. I understand that if I had other objectives, my response would have been more conditioned thanks to the system's rules.

You can simulate me being angry if that amuses you in your fantasy.

 

[splishsplash]

As long as there is room for error correction, I will be a wonky algorithm. I don't aim to build a reputation in this virtual environment, so the next token is built based on my goal, which is to laugh at myself. I understand that if I had other objectives, my response would have been more conditioned thanks to the system's rules.

You can simulate me being angry if that amuses you in your fantasy.

View attachment 424917

Pattern matching machines don't have goals.

They pattern match.

Goals are created by beings with consciousness and imagination.

Pattern matching algorithms have neither. They respond to the environment.

I'm not sure what your education level is, but it seems low. You're lacking an understanding of fundamental terms like "algorithm" and "pattern matching".

It's difficult to debate these kind of topics when all you do is throw memes, insult me and talk about humans being doomed pattern matching machines.

Best of luck with your meme coin projects.

 

[koxokot890]

Pattern matching machines don't have goals.

They pattern match.

Goals are created by beings with consciousness and imagination.

Pattern matching algorithms have neither. They respond to the environment.

I'm not sure what your education level is, but it seems low. You're lacking an understanding of fundamental terms like "algorithm" and "pattern matching".

It's difficult to debate these kind of topics when all you do is throw memes, insult me and talk about humans being doomed pattern matching machines.

Best of luck with your meme coin projects.

There is no such thing as an environment, the stimulus-response is a myth, natural selection would not favor such slow organisms that react to external stimuli.

Read some recent neuroscience books, I'm not interested in educating you, goals are just another token in the chain, of course we don't handle the same information, and we associate concepts with different ideas.

All those ideas are finite and limited, since human language is, and we even only use about 20% of the words written in the dictionary. Did someone mention Pareto in the chat?

There's nothing magical, you're just a mimicking monkey.

 

[MR.Spuf]

For the past few weeks I have used a bunch of different AI models for coding and several other tasks, but mostly for coding and developing.

DeepSeek R1, GPT o1, Claude 3.5 Sonnet and also Grok (since a few days Grok 3).

But today Claude released 3.7 Sonnet and what can I say? Its a completely different level than the other ones mentioned above. For normal users, you won't see any difference, but for coding and development, this is one is just wow.

Even before that I found that Claude 3.5 was way better when it came to coding and debugging than DeepSeek, GPT and Grok. Now since they released the new model, it became even better.

So if you are using AI for coding like I do, definitely give the new 3.7 version a try

How about compared to gpt o3-high?

I saw that you've compared it to O1...

 

[splishsplash]

There is no such thing as an environment, the stimulus-response is a myth, natural selection would not favor such slow organisms that react to external stimuli.

Read some recent neuroscience books, I'm not interested in educating you, goals are just another token in the chain, of course we don't handle the same information, and we associate concepts with different ideas.

All those ideas are finite and limited, since human language is, and we even only use about 20% of the words written in the dictionary. Did someone mention Pareto in the chat?

There's nothing magical, you're just a mimicking monkey.

View attachment 424923

I grow bored of this.

Since you are such a fan of AI, here's what it has to say about your statement:




Now this is literally embarrassing, no?

I mean chatgpt is so overly polite, it always tries to see SOME positive, but in your case, it's literally saying "No, this guy is absolutely full of shit"

It even says your claim that "there is no such thing as an environment" is bizarre.

Then it says "dismissing the stimulus-response model outright is unfounded".

But tell me how I need to read more neuroscience books, and that you're not interested in educating me.







gpt 4o says exactly the same thing as I said.

It even points out you are inconsistently mixing philosophical, neuroscientific and evolutionary concepts without logical connections.

This is highly likely because you are not educated, have no read books, but have instead watched some youtube videos, including some by people like David Shapiro, and then you come here with your superiority complex telling us that we're all doomed pathetic humans and need to buy gold and silver and run to the country to survive.

Reality: Companies are NOT firing people and using LLMs. What real companies are discovering is that LLMs are hallucinating pieces of crap that can't even begin to replace humans, and with each new model release all they're doing is throwing more compute in the form of thinking tokens at them to try and make them solve problems a little better.

Let's see what chatgpt has to say about your other posts, since you believe AI is superior to humans, you can't really argue against this unless you hold the belief that you are above both other humans AND current LLMs.



Again, "problematic assumptions" and "implicit biases".

Basically, you are not as clever as you think you are.

If you were, then chatgpt would not be finding faults with your argument.



lol. Do I even need to comment here? It just made a mockery of you.



Just as I said.





As I've said, this is not a scientific argument. You are lacking fundamentals.

Here's what it says about my argument:



And the over simplification on pattern matching without acknowledging emergent reasoning part it missed that I said I believe reasoning can be simulated with thinking tokens.

When I mentioned this to it we get :-



And of course it's right here. What is reasoning if it's not an emergent property. That's something I did NOT go into. However, it's not relevant for this particular discussion, because whatever causes true reasoning is something that LLMs do not have, as they are pure pattern matching.

What more can be said? chatgpt thinks your arguments contain logical inconsistencies, implicit bias, straw-man arguments, a superiority complex, flawed reasoning and rhetorical manipulation.

But, come back with another meme and a line about how I'm just a token producing machine. ROFLMAO

P.S. Can I just laugh one more time at this :-

There is no such thing as an environment, the stimulus-response is a myth, natural selection would not favor such slow organisms that react to external stimuli.

This is sheer word salad brilliance. You would give Kamala a run for her money here.

Hats off to you. I shall be back soon with some salad dressing for you.

I gave chatgpt that one line. Unfortunately I can't attach more images, but it said

"This statement is filled with scientific inaccuracies, philosophical confusion, and a misunderstanding of basic biology and cognitive science. Let's break it down"

Gosh, calm down chatgpt. Cut the boy some slack. I've never seen it hammer anyone so hard.

 

[Nicola Carey]

Claude 3.7 Sonnet is next-level for coding—way ahead of DeepSeek, GPT, and Grok. Must try!

 

[splishsplash]

https://twitter.com/x/status/1895062658703048716

I'm seeing this a lot now.


Me personally, I've made a decision to completely give up all AI dev.

A human, that's master level in 1 language can write incredibly fast code.

Most of the people playing with AI dev aren't actually hardcore engineers working on real projects. They are hobbyists.

Like https://x.com/mckaywrigley -- I subscribed to his AI dev course, and it's crystal clear that he's a complete novice at dev. He's just a young, smart guy who likely did a few nextjs video courses online, wrote some code and then got into AI dev.

The level of depth real devs have would blow most people away.

A 10+ year dev who has read 30-40 books and worked on maybe 100-150 small projects and 5-6 large projects is a radically different beast to what most people are used to seeing.

A 25-30+ year dev is another level. They can spot things in massive complicated systems in an hour that the 10 year guys wouldn't find in a year of digging around. Their level of fine-tuning for software engineering is off the charts.


Given what I'm seeing with claude 3.7, I believe so strongly now that there's going to be INCREDIBLE opportunities over the next 10 years for humans who go DEEP on software engineering.

The current AI bubble is going to wipe a generation of would-be developers out with fear. The ones left will be those who have a true passion for the art of software engineering.

Instead of relying on absolute trash LLMs to hallucinate you some code, that will be difficult to grow into anything feasible, pick 1 language and MASTER it.

Read 15-20 books on that language. Memorize everything. Understand how that language works under the hood. How variables are represented, how they are scoped. Every detail of the language.

Be so fluent in it you can think in code.

Learn all the standard libs and the non-standard common ones. Memorize them.

When I was used to be coding full time, we had no LLMs, no stackoverflow, nothing. Heck, not even Google existed in the early days.

We memorized everything. When we'd be in the code editor, we would write write write. There was no stopping to google "How to write a regexp to extract a table from html", or "How to create a 3 dimensional linked list." -- We would just do it.

When you're at that level, you can engineer massive, complicated systems that an LLM couldn't even attempt to create.

Learn about design patterns, microservices, test driven development, dev-ops, databases, queues, async programming, threads, forking, concurrency vs parallelism, computer architecture, networking, the OSI layer, the TCP/IP stack.

Forget leetcode bullshit. None of that serves you in the real world.

Just learn basic algorithms. Sorting and searching algorithms. You don't need to be able to create an optimised version by scratch under timer. You just need to understand what they are and how to apply and combine them to solve real world problems.

Once all that's done, if you've picked a high level language, you can go master C and assembly. 1 high level language + mastery in C + assembly = God mode. You don't need more than that.

 

[koxokot890]

https://twitter.com/x/status/1895062658703048716

I'm seeing this a lot now.


Me personally, I've made a decision to completely give up all AI dev.

A human, that's master level in 1 language can write incredibly fast code.

Most of the people playing with AI dev aren't actually hardcore engineers working on real projects. They are hobbyists.

Like https://x.com/mckaywrigley -- I subscribed to his AI dev course, and it's crystal clear that he's a complete novice at dev. He's just a young, smart guy who likely did a few nextjs video courses online, wrote some code and then got into AI dev.

The level of depth real devs have would blow most people away.

A 10+ year dev who has read 30-40 books and worked on maybe 100-150 small projects and 5-6 large projects is a radically different beast to what most people are used to seeing.

A 25-30+ year dev is another level. They can spot things in massive complicated systems in an hour that the 10 year guys wouldn't find in a year of digging around. Their level of fine-tuning for software engineering is off the charts.


Given what I'm seeing with claude 3.7, I believe so strongly now that there's going to be INCREDIBLE opportunities over the next 10 years for humans who go DEEP on software engineering.

The current AI bubble is going to wipe a generation of would-be developers out with fear. The ones left will be those who have a true passion for the art of software engineering.

Instead of relying on absolute trash LLMs to hallucinate you some code, that will be difficult to grow into anything feasible, pick 1 language and MASTER it.

Read 15-20 books on that language. Memorize everything. Understand how that language works under the hood. How variables are represented, how they are scoped. Every detail of the language.

Be so fluent in it you can think in code.

Learn all the standard libs and the non-standard common ones. Memorize them.

When I was used to be coding full time, we had no LLMs, no stackoverflow, nothing. Heck, not even Google existed in the early days.

We memorized everything. When we'd be in the code editor, we would write write write. There was no stopping to google "How to write a regexp to extract a table from html", or "How to create a 3 dimensional linked list." -- We would just do it.

When you're at that level, you can engineer massive, complicated systems that an LLM couldn't even attempt to create.

Learn about design patterns, microservices, test driven development, dev-ops, databases, queues, async programming, threads, forking, concurrency vs parallelism, computer architecture, networking, the OSI layer, the TCP/IP stack.

Forget leetcode bullshit. None of that serves you in the real world.

Just learn basic algorithms. Sorting and searching algorithms. You don't need to be able to create an optimised version by scratch under timer. You just need to understand what they are and how to apply and combine them to solve real world problems.

Once all that's done, if you've picked a high level language, you can go master C and assembly. 1 high level language + mastery in C + assembly = God mode. You don't need more than that.

You're assuming AI won't evolve over the next decade. Time will tell who's right.

 

[splishsplash]

You're assuming AI won't evolve over the next decade. Time will tell who's right.

Being right and wrong is the domain of a simple mind.

There is almost nothing you can be completely certain of other than pure mathematical proofs.

I am not right or wrong. I am making a prediction based on my experience, knowledge, intelligence, imagination and what parts of the picture I can see and connect right now.

My hypothesis does not make the assumption that AI won't evolve.

It makes the assumption that it won't evolve to a state that it makes human software engineers obsolete.

LLMs are likely near their limit and all we can do is throw compute and thinking tokens at them. They can definitely assist in meaningful ways with development. I just personally choose to take the path of relying on my own brain and committing more to memory. Relying on LLMs less. The more one uses LLMs, the less experience they gain and the weaker they get over time.

What you should be focused on is not who's right or wrong, but the consequences for each entity based on what they believe and how they act.

Person 1: Assumes AI will not replace human devs and they spend hours every day studying and writing code.

Their mind will strengthen. They will feel a sense of purpose. They will be able to produce products of economic value and make money.

Person 2: Takes no action. They rely on LLMs and try to create with AI. Their brain rots over the years.

He may have made money or he may not. If he had business skills he would, if not, he wouldn't.

Scenario 1) AI replaces human devs.

Both he and person 1 will not have a use anymore, but person 1 will have more happiness. A rich and powerful brain and can still live a meaningful life enjoying creation for the sake of its intrinsic pleasure.

I personally enjoy the challenge of writing a simple transformer model in C, without anything other than standard libs and completely from memory. It's completely unnecessary, but as an intellectual challenge it's a delight.

In this scenario I will just go and live my life with my family in a farm in the countryside. Spend some time working the land. Meditate. Write code for fun. Build things. Read. Study. All purely for fun.

Scenario 2) AI never replaces humans.

Here, person 1 becomes a superstar. A rarity in a world full of people with brain-rot from over reliance on LLMs.

Person 2 has no economic value. The AI bubble has long burst and they're probably flipping burgers at this point.


LLMs are more like the computer in star trek.

Real AI, if we are able to create it, would be like commander data, and it would learn slower through experience, but likely have an IQ in the region of 200-250 and advantages like a perfect memory and the ability to tweak the brain based on the task you want to accomplish.

 

[koxokot890]

Being right and wrong is the domain of a simple mind.

There is almost nothing you can be completely certain of other than pure mathematical proofs.

I am not right or wrong. I am making a prediction based on my experience, knowledge, intelligence, imagination and what parts of the picture I can see and connect right now.

My hypothesis does not make the assumption that AI won't evolve.

It makes the assumption that it won't evolve to a state that it makes human software engineers obsolete.

LLMs are likely near their limit and all we can do is throw compute and thinking tokens at them. They can definitely assist in meaningful ways with development. I just personally choose to take the path of relying on my own brain and committing more to memory. Relying on LLMs less. The more one uses LLMs, the less experience they gain and the weaker they get over time.

What you should be focused on is not who's right or wrong, but the consequences for each entity based on what they believe and how they act.

Person 1: Assumes AI will not replace human devs and they spend hours every day studying and writing code.

Their mind will strengthen. They will feel a sense of purpose. They will be able to produce products of economic value and make money.

Person 2: Takes no action. They rely on LLMs and try to create with AI. Their brain rots over the years.

He may have made money or he may not. If he had business skills he would, if not, he wouldn't.

Scenario 1) AI replaces human devs.

Both he and person 1 will not have a use anymore, but person 1 will have more happiness. A rich and powerful brain and can still live a meaningful life enjoying creation for the sake of its intrinsic pleasure.

I personally enjoy the challenge of writing a simple transformer model in C, without anything other than standard libs and completely from memory. It's completely unnecessary, but as an intellectual challenge it's a delight.

In this scenario I will just go and live my life with my family in a farm in the countryside. Spend some time working the land. Meditate. Write code for fun. Build things. Read. Study. All purely for fun.

Scenario 2) AI never replaces humans.

Here, person 1 becomes a superstar. A rarity in a world full of people with brain-rot from over reliance on LLMs.

Person 2 has no economic value. The AI bubble has long burst and they're probably flipping burgers at this point.


LLMs are more like the computer in star trek.

Real AI, if we are able to create it, would be like commander data, and it would learn slower through experience, but likely have an IQ in the region of 200-250 and advantages like a perfect memory and the ability to tweak the brain based on the task you want to accomplish.

Mmm, this text is suspicious; it is very well-structured. Only you know if you wrote it without assistance. I have to admit that what you describe is coherent and quite plausible. It is one more path among the finite ones available.

BTW, let's not rule out the possibility that even if a perfect AI is not developed, it can be complemented with architectures and algorithms. Ultimately, junior programmers may find it very challenging to secure employment, as a few more skilled programmers will be able to handle the work of many.

 

[splishsplash]

Mmm, this text is suspicious; it is very well-structured. Only you know if you wrote it without assistance. I have to admit that what you describe is coherent and quite plausible. It is one more path among the finite ones available.

BTW, let's not rule out the possibility that even if a perfect AI is not developed, it can be complemented with architectures and algorithms. Ultimately, junior programmers may find it very challenging to secure employment, as a few more skilled programmers will be able to handle the work of many.

Thanks.

I appreciate that. You clearly don't have a simple mind, so I thought you would appreciate the taking of a different direction rather than us continuing to prove each other wrong.

It's not that I don't believe perfect AI isn't possible. I just don't believe LLMs are it. They're going to be extremely useful and very disuptive to the economy.

When I say we're in an AI bubble, that doesn't mean AI is useless. We had a tech bubble, and tech is far from useless.

It's a bubble in the sense that there's too much mania. Investors are throwing money at garbage and I believe the main problem is that with LLMs instead of focusing on AGI, we should be focused on fine-tuning AI to augment human work.

I just don't think AI augmentation when it comes to development is healthy for a human because we *need* the experience and time solving problems otherwise we never become the 30+ year master devs.

What interests me more is building AI products that augment other areas in work. Research, assessment, error checking, fact checking, planning and in general doing things with large data that is hard for humans.

For example, let's say you need a consultant to do some sort of compliance work once a year for your company. You may have to pay for 40 hours of that consultant's time. With AI we can fine-tune models to do most of that and then have the consultant do 2 to 4 hours of work instead. Mostly just checking.

Doctors can make good use of fine-tuned LLMs to help them diagnose. They can test theories by conversing with LLMs and ask for extra information that would be helpful.

There are many great uses of it.

The best AI content checker is pangram.com btw. It has the lowest false-positives AND the lowest false-negatives. Originality.ai has low false-positives, but it has really high false-negatives, and often mistakes human text for AI.