This post resonates a lot with me. All the AI content has also made me question my writing and what I do. It would be so easy to turn on the AI switch and have all my stuff automated. But then again, what does that mean for me? I'd definitely lose my own voice and writing skills. Also, I write to learn. Most of my books and posts have been about me learning new stuff and taking whoever wants to along the way. Using LLMs for writing would interfere with that process. I also came to the conclusion to use AI ONLY on the edges: write boilerplate code, proofread, flag facts I got wrong, brainstorm ideas (sometimes), but not for writing itself.
You are one of the few machine learning writers whose posts/books I instantly read whenever I have the chance! I appreciate that you don't offload your writing to AI, we would lose a great creator.
Honestly, I don’t think people need “more”. I would love to have 2-3 articles to read per week from all authors I subscribed to. If everyone generates 2-3 per week, I feel so inadequate (somehow I need to find time to read 20-30 articles per week instead) that I tend to just abandon the platform or reduce number of authors I follow (which is a pity).
To fuck it! (I needed to add swearing, so it is clear it is not written by AI). By doing less you will be doing more - each article will be much more valuable for us to read.
Beautiful perspective, Tivadar. Definitely, less is more nowadays. And that’s the path to take.
But on my end, I think “automating with AI” doesn’t mean letting it do all the thinking or just spitting out the weights of the model into your Substack.
In my particular case, I love drawing diagrams, building things, thinking about how to display them, creating systems, etc., but I don’t like putting words into prose. So I automated that because it bores me. Not saying everyone should do that, but highlighting that using AI is neither good nor evil. It just makes being greedy a lot easier, where you can lose yourself competing with others.
On my end, I spend hours thinking about what I want to say, laying it out in the right format, phrase by phrase, and using AI just for the styling step.
Your posts are extremely high quality indeed, because the core value (your knowledge about AI engineering and systems) is there.
I'm definitely not anti-AI: all of my animations nowadays are made by Codex, from detailed scripts given by me. Even though the skill ceiling is still high, the overall content pool is diluted because of the easy access to generative AI. So while posting more was a valid strategy a couple of years ago, now content creators should do the opposite IMO.
“…I make something that’s better than 99% of what’s out there.” That’s not a choice for the future, you already do this, so I’m very happy to follow your feeds and to buy your book.
If you are sincere, then I appreciated your choice of a different approach to content to be, as to say, a bit unfair, uncomfortable, unpolite: some times truths are crude. And maybe, sometimes violent words are the only way to ask for that. Or is only our deficiency to ask for our right to know? Because, first we should aknowledge to have been only used for the plans of somebody else in our life...Second: I was pleasant all my life and the concrete results were terrible: I'm persecuted even in my home; people hate me and try to destroy my dignity, trying to speak badly about me with other people... It's all so absurd! I'm very tired.
Only faith sustains me and gives me the strenght needed.
Welcome to Unpleasantville. Or...are you trying only to train your AI to simulate our imperfect human nature?
Anyway, I like your content videos and writings a lot...Thank you, sir. Ah, you're an excellent science and math teacher! Science really constitutes a comfort to me.
Very well said. There was a time I spent months on a mathematical problem. After getting the answer, I still go back and revisit the solution just to disprove the question or manipulate it. With AI it makes it almost impossible with the sugar rush. Now I'm trying to savour the AI output as well which, as you've said, takes time.
Life wasn't meant to be rushed, it was meant to be slowly savoured, same goes for technical writing, I'd rather wait for something truly original than be bombarded with redundant content.
“Stochastic parrots” is accurate and awesome. Thank you. Two notes about coding something quickly in Python and the “benefits” of vibe coding in general: the act of coding is (imho) the very best way to learn a subject and when it comes time to change your code (Python or otherwise) you will wish you knew how it worked. My 2c.
You are absolutely right, and it does need careful consideration.
Do What people (don't realise they) need, that LLMs can't do: build community in public. As we all ignore each other: instead of dialogue we reach for the perfect answer, immediately available to every question, we stop talking to each other, and we stop having community. The isolation cake covid gave us, now has a perfect icing on top courtesy of AI. We don't know why we are depressed, alone, isolated because life has never been this good. We're missing community and we need community builders. I think you've been doing this if you've got readers and followers. Keep going, just change gear.
Build community by inviting your audience into your output - start a dialogue. Find followers who want coaching, and post the recorded teaching sessions. Pick some engaged readers and facilitate 'Peer Seminars' - putting the output available to just the community. Run a series called Unpopular Opinion (totally borrowed from 'the changelog's' 'go time' podcast) and interview the opinion holder, challenging the audience to learn something. Take requests, and bring on audience guests. Line up difficult conversations, bring them back on later for reflection. Invite a newbie to sumarize what they know and ask you/ others to fill the gaps. Find two intermediates to teach each other something, then get the learner for each topic to present an extension to learnt topic in another article.
Secret sauce: Lift people up. Use your platform to showcase people so they can get ahead in their life-goals / career: They can reference the posts they were in, or the contribution they made, recruiters look back at your community for evidence. Members of your community will do great things and will point to where it started: The Palindrome. It's the long game but its also a snowball. And LLMs can't do it.
Don't try to out produce LLMs.
This response was carefully typed by hand. No AI was used in dictation, nor in generation, nor in editing. This is 100% bona fide human consideration and brain sweat, and real human life-hours were expent to produce it.
This post resonates a lot with me. All the AI content has also made me question my writing and what I do. It would be so easy to turn on the AI switch and have all my stuff automated. But then again, what does that mean for me? I'd definitely lose my own voice and writing skills. Also, I write to learn. Most of my books and posts have been about me learning new stuff and taking whoever wants to along the way. Using LLMs for writing would interfere with that process. I also came to the conclusion to use AI ONLY on the edges: write boilerplate code, proofread, flag facts I got wrong, brainstorm ideas (sometimes), but not for writing itself.
You are one of the few machine learning writers whose posts/books I instantly read whenever I have the chance! I appreciate that you don't offload your writing to AI, we would lose a great creator.
thank you, I feel the same and I appreciate your choice and direction ❤️
Honestly, I don’t think people need “more”. I would love to have 2-3 articles to read per week from all authors I subscribed to. If everyone generates 2-3 per week, I feel so inadequate (somehow I need to find time to read 20-30 articles per week instead) that I tend to just abandon the platform or reduce number of authors I follow (which is a pity).
To fuck it! (I needed to add swearing, so it is clear it is not written by AI). By doing less you will be doing more - each article will be much more valuable for us to read.
Beautiful perspective, Tivadar. Definitely, less is more nowadays. And that’s the path to take.
But on my end, I think “automating with AI” doesn’t mean letting it do all the thinking or just spitting out the weights of the model into your Substack.
In my particular case, I love drawing diagrams, building things, thinking about how to display them, creating systems, etc., but I don’t like putting words into prose. So I automated that because it bores me. Not saying everyone should do that, but highlighting that using AI is neither good nor evil. It just makes being greedy a lot easier, where you can lose yourself competing with others.
On my end, I spend hours thinking about what I want to say, laying it out in the right format, phrase by phrase, and using AI just for the styling step.
Your posts are extremely high quality indeed, because the core value (your knowledge about AI engineering and systems) is there.
I'm definitely not anti-AI: all of my animations nowadays are made by Codex, from detailed scripts given by me. Even though the skill ceiling is still high, the overall content pool is diluted because of the easy access to generative AI. So while posting more was a valid strategy a couple of years ago, now content creators should do the opposite IMO.
“…I make something that’s better than 99% of what’s out there.” That’s not a choice for the future, you already do this, so I’m very happy to follow your feeds and to buy your book.
Thanks for your kind words! There's always room to improve, and I always want to take things to the next level.
If you are sincere, then I appreciated your choice of a different approach to content to be, as to say, a bit unfair, uncomfortable, unpolite: some times truths are crude. And maybe, sometimes violent words are the only way to ask for that. Or is only our deficiency to ask for our right to know? Because, first we should aknowledge to have been only used for the plans of somebody else in our life...Second: I was pleasant all my life and the concrete results were terrible: I'm persecuted even in my home; people hate me and try to destroy my dignity, trying to speak badly about me with other people... It's all so absurd! I'm very tired.
Only faith sustains me and gives me the strenght needed.
Welcome to Unpleasantville. Or...are you trying only to train your AI to simulate our imperfect human nature?
Anyway, I like your content videos and writings a lot...Thank you, sir. Ah, you're an excellent science and math teacher! Science really constitutes a comfort to me.
What a refreshing take. Thanks for that
Very well said. There was a time I spent months on a mathematical problem. After getting the answer, I still go back and revisit the solution just to disprove the question or manipulate it. With AI it makes it almost impossible with the sugar rush. Now I'm trying to savour the AI output as well which, as you've said, takes time.
IMO this kind of tinkering with solutions is the foundation of science.
Life wasn't meant to be rushed, it was meant to be slowly savoured, same goes for technical writing, I'd rather wait for something truly original than be bombarded with redundant content.
yes, a good thing. All the important information is, always, in the footnotes.
“Stochastic parrots” is accurate and awesome. Thank you. Two notes about coding something quickly in Python and the “benefits” of vibe coding in general: the act of coding is (imho) the very best way to learn a subject and when it comes time to change your code (Python or otherwise) you will wish you knew how it worked. My 2c.
To credit where it's due: I didn't come up with the "stochastic parrot" terminology, it is probably originated from this paper: https://dl.acm.org/doi/10.1145/3442188.3445922
You are absolutely right, and it does need careful consideration.
Do What people (don't realise they) need, that LLMs can't do: build community in public. As we all ignore each other: instead of dialogue we reach for the perfect answer, immediately available to every question, we stop talking to each other, and we stop having community. The isolation cake covid gave us, now has a perfect icing on top courtesy of AI. We don't know why we are depressed, alone, isolated because life has never been this good. We're missing community and we need community builders. I think you've been doing this if you've got readers and followers. Keep going, just change gear.
Build community by inviting your audience into your output - start a dialogue. Find followers who want coaching, and post the recorded teaching sessions. Pick some engaged readers and facilitate 'Peer Seminars' - putting the output available to just the community. Run a series called Unpopular Opinion (totally borrowed from 'the changelog's' 'go time' podcast) and interview the opinion holder, challenging the audience to learn something. Take requests, and bring on audience guests. Line up difficult conversations, bring them back on later for reflection. Invite a newbie to sumarize what they know and ask you/ others to fill the gaps. Find two intermediates to teach each other something, then get the learner for each topic to present an extension to learnt topic in another article.
Secret sauce: Lift people up. Use your platform to showcase people so they can get ahead in their life-goals / career: They can reference the posts they were in, or the contribution they made, recruiters look back at your community for evidence. Members of your community will do great things and will point to where it started: The Palindrome. It's the long game but its also a snowball. And LLMs can't do it.
Don't try to out produce LLMs.
This response was carefully typed by hand. No AI was used in dictation, nor in generation, nor in editing. This is 100% bona fide human consideration and brain sweat, and real human life-hours were expent to produce it.
Thanks Amigo,
⚓ Semper Fortis!
......one day at a time.