AI systems are stochastic parrots.
-
AI systems are stochastic parrots. They mix and match what they have learned by reading what we have produced over the decades.
Some people have told me that AI systems suggest solutions similar, or identical, to those I was already suggesting in some posts on my blogs back in 2020, 2021, 2022... before the AI era. And yet, every now and then, someone writes to me saying that one of my blog posts from back then is "clearly AI generated because AI suggests the same method".
This phenomenon is becoming fairly recurring, and I know at least two people who have stopped contributing publicly to Open Source and education because they got tired of being accused of producing slop.
If we want to stay away from all this, if we want to demonstrate that AI is harmful and not a solution, shouting at anything that "might look AI generated, and often isn't" is not how we solve the problem. Let's produce good content. Let's explain the long term cost of all this. Let's do useful things. Let's show what the human mind is capable of doing great things.
Because otherwise, we risk the opposite: only AI systems and vibe coders will remain, while everyone else, disgusted, walks away for good from this wonderful world of ours.
If we want a world with less AI slop, let’s not make it unbearable for humans to create.
-
AI systems are stochastic parrots. They mix and match what they have learned by reading what we have produced over the decades.
Some people have told me that AI systems suggest solutions similar, or identical, to those I was already suggesting in some posts on my blogs back in 2020, 2021, 2022... before the AI era. And yet, every now and then, someone writes to me saying that one of my blog posts from back then is "clearly AI generated because AI suggests the same method".
This phenomenon is becoming fairly recurring, and I know at least two people who have stopped contributing publicly to Open Source and education because they got tired of being accused of producing slop.
If we want to stay away from all this, if we want to demonstrate that AI is harmful and not a solution, shouting at anything that "might look AI generated, and often isn't" is not how we solve the problem. Let's produce good content. Let's explain the long term cost of all this. Let's do useful things. Let's show what the human mind is capable of doing great things.
Because otherwise, we risk the opposite: only AI systems and vibe coders will remain, while everyone else, disgusted, walks away for good from this wonderful world of ours.
If we want a world with less AI slop, let’s not make it unbearable for humans to create.
@stefano One part of reducing the harm is to figure out ways to filter out at least part of the GenAI data collector bots from web traffic. That is constantly getting more difficult, as the amount of bots grows, but at least a small part of those follow robots.txt type of rules/guidelines.
I just hope there would be better examples of carefully curated robots.txt files that have been validated to reduce the amount of bot traffic (from larger commercial bots).
People with static websites (hosted on various SaaS/PaaS providers) have very little other options for filtering out bot traffic, as they can't add extra network traffic filters in between the websites and visitors. That is kind of a challenge, as a lot of hobbyists are with very limited resources to protect their original writing from GenAI bots.
-
@stefano One part of reducing the harm is to figure out ways to filter out at least part of the GenAI data collector bots from web traffic. That is constantly getting more difficult, as the amount of bots grows, but at least a small part of those follow robots.txt type of rules/guidelines.
I just hope there would be better examples of carefully curated robots.txt files that have been validated to reduce the amount of bot traffic (from larger commercial bots).
People with static websites (hosted on various SaaS/PaaS providers) have very little other options for filtering out bot traffic, as they can't add extra network traffic filters in between the websites and visitors. That is kind of a challenge, as a lot of hobbyists are with very limited resources to protect their original writing from GenAI bots.
@autiomaa the problem I'm facing is that those botnets (AI or not - I have very good reasons to say they're not all AI but different bad actors) use residential proxies and it's impossible to block all of them.
I don't like the PoW solutions, even if I'm using Anubis for some specific situations, because adding pollution and load to reduce load is...counterintuitive.
-
@autiomaa the problem I'm facing is that those botnets (AI or not - I have very good reasons to say they're not all AI but different bad actors) use residential proxies and it's impossible to block all of them.
I don't like the PoW solutions, even if I'm using Anubis for some specific situations, because adding pollution and load to reduce load is...counterintuitive.
@stefano Yeah, residential proxies are unsolvable problem, but at least some of the traditional search engines used to follow robots.txt rules. But even those are increasingly not following them anymore, as for example Google keeps indexing pages with noindex tags on the rules.
(I get monthly emails from Google telling that they indexed my no-index page that I have forbidden them to index in the robots.txt rules.)
-
AI systems are stochastic parrots. They mix and match what they have learned by reading what we have produced over the decades.
Some people have told me that AI systems suggest solutions similar, or identical, to those I was already suggesting in some posts on my blogs back in 2020, 2021, 2022... before the AI era. And yet, every now and then, someone writes to me saying that one of my blog posts from back then is "clearly AI generated because AI suggests the same method".
This phenomenon is becoming fairly recurring, and I know at least two people who have stopped contributing publicly to Open Source and education because they got tired of being accused of producing slop.
If we want to stay away from all this, if we want to demonstrate that AI is harmful and not a solution, shouting at anything that "might look AI generated, and often isn't" is not how we solve the problem. Let's produce good content. Let's explain the long term cost of all this. Let's do useful things. Let's show what the human mind is capable of doing great things.
Because otherwise, we risk the opposite: only AI systems and vibe coders will remain, while everyone else, disgusted, walks away for good from this wonderful world of ours.
If we want a world with less AI slop, let’s not make it unbearable for humans to create.
-
@05dc5187506e870cd19d2454597d7a91fffa5111b3c75ae371e75a3e4270781e @autiomaa I don't like the PoW solutions, even if I'm using Anubis for some specific situations, because adding pollution and load to reduce load is...counterintuitive.
-
AI systems are stochastic parrots. They mix and match what they have learned by reading what we have produced over the decades.
Some people have told me that AI systems suggest solutions similar, or identical, to those I was already suggesting in some posts on my blogs back in 2020, 2021, 2022... before the AI era. And yet, every now and then, someone writes to me saying that one of my blog posts from back then is "clearly AI generated because AI suggests the same method".
This phenomenon is becoming fairly recurring, and I know at least two people who have stopped contributing publicly to Open Source and education because they got tired of being accused of producing slop.
If we want to stay away from all this, if we want to demonstrate that AI is harmful and not a solution, shouting at anything that "might look AI generated, and often isn't" is not how we solve the problem. Let's produce good content. Let's explain the long term cost of all this. Let's do useful things. Let's show what the human mind is capable of doing great things.
Because otherwise, we risk the opposite: only AI systems and vibe coders will remain, while everyone else, disgusted, walks away for good from this wonderful world of ours.
If we want a world with less AI slop, let’s not make it unbearable for humans to create.
@stefano this is a really good set of ideas. There's some harsh reality to go with them. We need to pay money for this to happen.
In earth systems science - all the funding went to stochastic parroting. 4 years of job hunting now shows how taking a pro- first principles (ipso no LLMs etc) hurts career prospects.
In the internet? well, nobody wants to support human made writing and work. Kind words don't pay rent.
My own cup is empty, paying money is how to fill it.
-
@05dc5187506e870cd19d2454597d7a91fffa5111b3c75ae371e75a3e4270781e @autiomaa I don't like the PoW solutions, even if I'm using Anubis for some specific situations, because adding pollution and load to reduce load is...counterintuitive.
@stefano @05dc5187506e870cd19d2454597d7a91fffa5111b3c75ae371e75a3e4270781e And those compute-based gates only slow down regular users, bots don't care about wasted CPU, since they largely run on a stolen home hardware (via IoT botnets).
-
@stefano this is a really good set of ideas. There's some harsh reality to go with them. We need to pay money for this to happen.
In earth systems science - all the funding went to stochastic parroting. 4 years of job hunting now shows how taking a pro- first principles (ipso no LLMs etc) hurts career prospects.
In the internet? well, nobody wants to support human made writing and work. Kind words don't pay rent.
My own cup is empty, paying money is how to fill it.
@adamsteer You're right, and that's part of the problem. There are things that overwhelm us, whether we like it or not, and simply ignoring or denying them is often not the solution.
-
J jwcph@helvede.net shared this topic
