I'm traveling for the week so I'll take the opportunity to do another AMA. It's been a while since the last public one (4 months to be precise).
This is a good moment for you to ask all those questions you shy away from asking me because they have nothing to do with a specific article.
You know, the usual:
Writing
How I manage/grow/nurture TAB
The future of the newsletter
“How can I get 1,000 paid subs?”
And, of course, anything-AI
Just so you know my opinion, I think these AMA sessions are often underused.
There are always interesting insights you can get from someone who's doing what you want to do or who knows what you want to know that you couldn't easily get otherwise.
From the standpoint of our current development in AI tech, I’d be very interested to hear if there are any fiction books you recommend that paint most likely potential futures
It's hard to decide what "likely potential futures" means. We really don't know where we're going. But if I were to bet, the older science fiction books have passed the test of time for a reason (not necessarily for being accurate in their predictions, though).
Alberto, I'm going to take you up on your suggestions. I'm assuming that you included these prompts because you have thought about the issue and want to share the answers. I hope that's the case. My questions are:
Haha not really, I haven't thought about it much yet. I'm working on a piece in the first question and the second question is perhaps too broad... Those were suggestions for topics! I could write a piece on the second question as well but I'm unsure interest is broad enough. My most active readers are writers as well, but most of my subscribers aren't. It's a dilemma for me. Once I publish the first one I'll decide depending on the reception.
So, you're just going to leave me hanging, then? I was so looking forward to reading two paragraphs and then walking away with 1000 new paid subscribers. I'm teasing.
I do hope that you'll write about both of these topics because I really am interested in your insight and I can. tell you from my personal experiences with you that there is some degree of craft or deliberation in how you nurture TAB.
I was a free subscriber for like 18 months? Somewhere around there. I appreciated your newsletter and would always recommend it for one reason: the length of your free previews. Even if you were going to get cut off by a paywall, there was still enough information before that cut off to justify reading it anyway. Being able to read the entire articles and see how you react to comments justifies the paid subscription. One thing I appreciate about how you respond to questions is that you never make people feel dumb for not knowing the answer or for being unfamiliar with the concept. There's a lot of value in that. The other thing I really appreciate is that you frequently include sources for additional learning for those who are curious enough to venture off on their own.
Thanks a lot John, really appreciate your words! 🙏🏻
I want to write more about those topics because I feel there's a lot of interest and I'm in a better and more stable position myself to talk about them. At the same time, I don't want to participate much in making Substack a new Medium, with those kinds of articles. I'm divided. I'll figure it out eventually!
I'm interested in finding a group blog of exclusively AI content. Humans behind the scenes only. Have you seen anything like this on Substack? If group blog is a bridge too far, how about any blog with exclusively AI content?
If there is nothing like this on Substack, or very few such blogs, is this an opportunity worth exploring in your opinion? A chance to be unique?
As you know better than I, there are a good number of blogs that talk about AI. But where are the blogs doing AI?
In other news, I just got ChatGPT Plus, and like so many others, am finding it pretty amazing. Dalle blows my little mind!
I hope you're going on a vacation to some exotic destination where they have real life babes, and not just digital friends! :-)
Hey Phil, I haven't seen anything 100% AI-written though I'm sure there is. Most newsletters have no traction so it's very hard to know for sure. I'd guess something fully AI-written would hardly get any.
I understand your point but there's still important value that a human can provide by writing that AI can't. The best you can do is find content farms to see how well they're doing.
There might be more people exploring AI use as a copilot or partner, but I think it's more common to use the tools for anything but writing. Like search, ideation, grammar, research, etc.
Hmm... I'm not sure why there are so many newsletters are discussing Gen AI if nobody is supposed to use it to publish. (??)
AI can provide value that humans can't too. For example, my current Substack blog wouldn't exist without AI, because doing it all myself would be way too much work for way too little return. Whether my current blog provides any value is of course another question.
That last one is the important question that answers why not that many people are using AI to write. AI (also generative AI) has many uses that don't involve writing articles. I don't think it's incompatible at all to write about AI and not use it for this specific purpose. For instance, I use it for the images, as you know, but not for the writing.
Ok, fair points, thanks. I'm wondering a number of things, such as....
Of those who express the view that AI writing doesn't currently provide value, how much time have they put in to learning AI writing? I'm just getting started myself, so while my results so far are suitable for my site, they're not going to win any awards, agreed. I think what may be happening is that a lot of people get this far, don't like what they see, so they quit, and then declare AI writing useless.
But then I see people like the Nerdy Novelist, who is serious about both AI and fiction. https://www.youtube.com/@TheNerdyNovelist. And I think to myself, these are the kind of people who are pioneering the future. From his work I see the AI writing realm is already a lot more sophisticated than anything I ever read about on Substack, and it's only going to get better from here.
Also, I'm coming to the conclusion that Substack is not really an ideal place to discuss AI writing due to the understandable built in bias from human writers who see AI as a threat. It seems more important to dismiss AI writing on Substack than to learn it. Anyway, given that AI is coming like it or not, it interests me to see what can be done with it.
Finally, much of my view is formed by the experience of spending months on Substack devoting myself to serious human writing on topics of great importance, and then watching it be completely ignored. Way too much work, for way too little reward. It's not clear to me why I should keep investing in that process.
"From his work, I see the AI writing realm is already a lot more sophisticated than anything I ever read about on Substack."
Do you mean his AI writing is better than anything on here written by anyone? If that's the case then I suggest you read more things on here! There are amazing writers on here who clearly invalidate your claim (I admit it's a matter of opinion but still).
That said, I've read things co-written with AI that are really good. I'm not denying its value as a copilot of writing. I'm denying its value as a tool that can autonomously write anything worthwhile.
I agree that it's hard to talk with writers about this! Of course they (we) get defensive. I like to think I'm rather objective but I can't be fully objective. I nevertheless acknowledge some writers are giving AI a very good use. It's all about learning how to integrate it with your workflow, not about how to replace the entire workflow with AI. That's just lazy.
Most people won't get to the point where they feel AI is useful *for anything* that's just how tech adoption works and why there's an important advantage for those who learn to make it work. Using AI (somewhere) to write is the same thing if you do it well (leaving ethics considerations aside).
Finally, about your own writing, I'm sad to hear that. Just know that your story is the most common story. It's really hard to make it big on these platforms, irregardless of your writing quality or the importance of the topics you touch on. Substack, like Medium, follows an extreme power law. Very few succeed although many more try just as hard.
Thank you! And wish you the best luck both with your human writing and the AI writing as well.
Hi again, sorry, no, I didn't mean that Nerdy Novelist's writing was better than anything on Substack. I meant that his channel introduced me to all kinds of AI writing tools and techniques I've never seen discussed on Substack. AI writing is a bigger world than I had previously understood.
Anyway, yesterday I upgraded to ChatGPT Plus, and am really liking it. Dalle is beyond incredible. As I start my journey in to a more serious exploration of AI, here's that beginning.
My story idea. The text is 99%+ straight out of ChatGPT 3.5. Images from Dalle. I haven't found the button where you get a Pulitzer Prize yet, still working on that.... :-)
I look forward to this one. One of my goals for this year is to write more consistently and really hone it as a skill. Witnessing the growth of your newsletter and your development as a writer is super-encouraging. Thank you for sharing.
Well, the persistently flawed narrative out there is that The Public must sacrifice IP observation, data privacy and other legal protectionism effects for humans for the pure sake of "innovation". OR Is there anything other than privacy roadkill for the innovators who don't care who is necessarily harmed by their products and innovation? The current standard is to deliberately make invasive pervasive ubiquitously dominant AI the governance standard ahead of the ability to legally govern it. That is the going rate of character of the business side innovation we have now. They're ready to nuke privacy right now - no looking back- because they have cannibalized people's personal information to their detriment for years. Just want to know if you seen any correction in the "hearts and minds" of the people making generative AI on this.
Many of the people I know use ChatGPT for search. As do I. Indeed, if I had to name a "killer app" for the technology that's what I would say. What in your view are the pros and cons of using the technology for that purpose? Anything interesting on the horizon?
I don't do it for one simple reason: I can't be sure I'm getting the info correctly. In many cases, one wrong data point is very problematic. If it had clear and direct references to where it's getting the info from, it'd be better. But it's hard to trust something that confabulates before saying "I don't know." Perhaps Perplexity or Bing are better options for search.
The pros are clear: if you have an "intelligent database of all knowledge on the internet," who wouldn't use it? It could be a super search engine.
For now, we're in the early adopters camp and the use is highly dependent on the goal of the search.
I've had great success with Perplexity for search-related tasks. ChatGPT does *ok* at search, however, you have to manually verify the relevance of links returned, as well as ensure that it's not merely hallucinating links which don't exist. You can also use Bing w/its version of GPT4 for search.
They had a coupon at the new year that gave you 2 months of their "pro" access for free. You can still use the GPT-4 model with their service if you prefer and I have found that it's really pretty dang good. It will ask a couple clarifying questions to help guide the search and cites all of the sources well.
I was skeptical and wouldn't have tried it without the free trial. That said, over the course of this month I've stopped using google entirely and now use ChatGPT more for when I want to have a conversation vs a web-search/answer.
I use Midjourney to create AI generative images. I would like to use AI to create story scenarios around the images - e.g., start with an image of a man and a woman are sitting at a table at a sidewalk cafe. I would like an AI application that would evaluate the image and give possible interpretations such as: 1) they are brother and sister discussing family business; 2) they are undercover police watching somebody; 3) they just met and are starting to get romantically involved. So two questions: 1) is there an AI application that does this now; and 2) if not, is there a corpus of fiction texts - novels, short stories, advertisements, etc - that could be used as training data? Any suggestions or pointers will be appreciated.
I don't know of any tool that does this, but I don't think it'd be that hard to create a GPT that did it. You really only need Dalle and ChatGPT and ask ChatGPT to play a game of guessing the scene from an image.
I don't know if ChatGPT can do this. It wouldn't surprise me.
It does however work great in the opposite direction. I started with having ChatGPT Plus write a story, and now I'm illustrating the story with the included Dalle. Dalle really kicks ass, imho.
I'll take a paragraph out of the story that describes some scene, and then tell Dalle to create an image to match the description. So far, for me at least, it nails it every time.
Hi Alberto, love your Substack for a number of reasons. One of which you are an AI expert of Latine background, and I work for a national (U.S.) Latine nonprofit that is trying to wrap their minds around how to operationalize AI. Not easy! On another matter, have you read Bostrom’s “Superintelligence”? What is your take on Yudkowsky’s (and others) fast AI takeoff/“AI ruin” scenario after we achieve AGI? Thanks!
With the exponential rise of AI as the new buzzword across all industries, there is a LOT of activity out there about "AI Trainings" of all sorts. my 2c is, while for those of us with a maker spirit the best way to learn is to start experimenting / trial and error, for the "feint of heart" there is a lot of emotional and functional barriers to adoption, so professional trainings are likely to play a role in ramping up adoption.
What do you think about the training offerings popping up out there? What do you think adds value for whom, and more broadly speaking, what do you think will be the main factors for accelerating adoption beyond the "trial and error" phase?
I'm pretty wary of courses which claim to teach you how to use ChatGPT or other AI tools. For me the best way to learn, is by doing. Meaning, sitting in front of my computer and using the tool. I find that a lot of training courses are way too superficial and high-level to provide much depth. But I may be an outlier here--a lot of people seem to like these kinds of things.
I cannot praise your content enough. Your writing is invaluable for keeping up to date on developments in the tech scene. Particularly handy for the underground commute where I can't access the internet.
I've recently moved to London to pursue a career in tech/automation ethics and policy. I completed my bachelor's in politics at the end of last year. I've spent the last few years (and currently) bartending while studying, so I've been looking for a more office admin type job to transition to the more corporate space. I'll have to wait three years before I can take a master's degree at local rates, so I'm desperate to gain some practical, career aligned experience in the meantime; hopefully more than just a generic office job.
Do you (or anyone in the forum) have any advice for those of us trying to break into this industry?
What sort of job do you want to do? You mention "automation ethics and policy". To me that sounds like you want to work at a think tank or similar organization. One suggestion would be to write a description of the things you want to do with your job, and feed it into ChatGPT. Ask ChatGPT: "given this description, what kinds of job titles should I be searching for?" Then, with your list of job titles start searching for employers.
From the standpoint of our current development in AI tech, I’d be very interested to hear if there are any fiction books you recommend that paint most likely potential futures
It's hard to decide what "likely potential futures" means. We really don't know where we're going. But if I were to bet, the older science fiction books have passed the test of time for a reason (not necessarily for being accurate in their predictions, though).
Alberto, I'm going to take you up on your suggestions. I'm assuming that you included these prompts because you have thought about the issue and want to share the answers. I hope that's the case. My questions are:
1. "How can I get 1,000 paid subs?"
2. How do you manage/grow/nurture TAB?
Haha not really, I haven't thought about it much yet. I'm working on a piece in the first question and the second question is perhaps too broad... Those were suggestions for topics! I could write a piece on the second question as well but I'm unsure interest is broad enough. My most active readers are writers as well, but most of my subscribers aren't. It's a dilemma for me. Once I publish the first one I'll decide depending on the reception.
So, you're just going to leave me hanging, then? I was so looking forward to reading two paragraphs and then walking away with 1000 new paid subscribers. I'm teasing.
I do hope that you'll write about both of these topics because I really am interested in your insight and I can. tell you from my personal experiences with you that there is some degree of craft or deliberation in how you nurture TAB.
I was a free subscriber for like 18 months? Somewhere around there. I appreciated your newsletter and would always recommend it for one reason: the length of your free previews. Even if you were going to get cut off by a paywall, there was still enough information before that cut off to justify reading it anyway. Being able to read the entire articles and see how you react to comments justifies the paid subscription. One thing I appreciate about how you respond to questions is that you never make people feel dumb for not knowing the answer or for being unfamiliar with the concept. There's a lot of value in that. The other thing I really appreciate is that you frequently include sources for additional learning for those who are curious enough to venture off on their own.
Thanks a lot John, really appreciate your words! 🙏🏻
I want to write more about those topics because I feel there's a lot of interest and I'm in a better and more stable position myself to talk about them. At the same time, I don't want to participate much in making Substack a new Medium, with those kinds of articles. I'm divided. I'll figure it out eventually!
Ok, here ya go Alberto....
I'm interested in finding a group blog of exclusively AI content. Humans behind the scenes only. Have you seen anything like this on Substack? If group blog is a bridge too far, how about any blog with exclusively AI content?
If there is nothing like this on Substack, or very few such blogs, is this an opportunity worth exploring in your opinion? A chance to be unique?
As you know better than I, there are a good number of blogs that talk about AI. But where are the blogs doing AI?
In other news, I just got ChatGPT Plus, and like so many others, am finding it pretty amazing. Dalle blows my little mind!
I hope you're going on a vacation to some exotic destination where they have real life babes, and not just digital friends! :-)
Hey Phil, I haven't seen anything 100% AI-written though I'm sure there is. Most newsletters have no traction so it's very hard to know for sure. I'd guess something fully AI-written would hardly get any.
I understand your point but there's still important value that a human can provide by writing that AI can't. The best you can do is find content farms to see how well they're doing.
There might be more people exploring AI use as a copilot or partner, but I think it's more common to use the tools for anything but writing. Like search, ideation, grammar, research, etc.
Hmm... I'm not sure why there are so many newsletters are discussing Gen AI if nobody is supposed to use it to publish. (??)
AI can provide value that humans can't too. For example, my current Substack blog wouldn't exist without AI, because doing it all myself would be way too much work for way too little return. Whether my current blog provides any value is of course another question.
That last one is the important question that answers why not that many people are using AI to write. AI (also generative AI) has many uses that don't involve writing articles. I don't think it's incompatible at all to write about AI and not use it for this specific purpose. For instance, I use it for the images, as you know, but not for the writing.
Ok, fair points, thanks. I'm wondering a number of things, such as....
Of those who express the view that AI writing doesn't currently provide value, how much time have they put in to learning AI writing? I'm just getting started myself, so while my results so far are suitable for my site, they're not going to win any awards, agreed. I think what may be happening is that a lot of people get this far, don't like what they see, so they quit, and then declare AI writing useless.
But then I see people like the Nerdy Novelist, who is serious about both AI and fiction. https://www.youtube.com/@TheNerdyNovelist. And I think to myself, these are the kind of people who are pioneering the future. From his work I see the AI writing realm is already a lot more sophisticated than anything I ever read about on Substack, and it's only going to get better from here.
Also, I'm coming to the conclusion that Substack is not really an ideal place to discuss AI writing due to the understandable built in bias from human writers who see AI as a threat. It seems more important to dismiss AI writing on Substack than to learn it. Anyway, given that AI is coming like it or not, it interests me to see what can be done with it.
Finally, much of my view is formed by the experience of spending months on Substack devoting myself to serious human writing on topics of great importance, and then watching it be completely ignored. Way too much work, for way too little reward. It's not clear to me why I should keep investing in that process.
Enjoy your trip!
"From his work, I see the AI writing realm is already a lot more sophisticated than anything I ever read about on Substack."
Do you mean his AI writing is better than anything on here written by anyone? If that's the case then I suggest you read more things on here! There are amazing writers on here who clearly invalidate your claim (I admit it's a matter of opinion but still).
That said, I've read things co-written with AI that are really good. I'm not denying its value as a copilot of writing. I'm denying its value as a tool that can autonomously write anything worthwhile.
I agree that it's hard to talk with writers about this! Of course they (we) get defensive. I like to think I'm rather objective but I can't be fully objective. I nevertheless acknowledge some writers are giving AI a very good use. It's all about learning how to integrate it with your workflow, not about how to replace the entire workflow with AI. That's just lazy.
Most people won't get to the point where they feel AI is useful *for anything* that's just how tech adoption works and why there's an important advantage for those who learn to make it work. Using AI (somewhere) to write is the same thing if you do it well (leaving ethics considerations aside).
Finally, about your own writing, I'm sad to hear that. Just know that your story is the most common story. It's really hard to make it big on these platforms, irregardless of your writing quality or the importance of the topics you touch on. Substack, like Medium, follows an extreme power law. Very few succeed although many more try just as hard.
Thank you! And wish you the best luck both with your human writing and the AI writing as well.
Hi again, sorry, no, I didn't mean that Nerdy Novelist's writing was better than anything on Substack. I meant that his channel introduced me to all kinds of AI writing tools and techniques I've never seen discussed on Substack. AI writing is a bigger world than I had previously understood.
Anyway, yesterday I upgraded to ChatGPT Plus, and am really liking it. Dalle is beyond incredible. As I start my journey in to a more serious exploration of AI, here's that beginning.
https://hippytoons.com/p/hippie-haven-clinic-nurse-nancy-restores
My story idea. The text is 99%+ straight out of ChatGPT 3.5. Images from Dalle. I haven't found the button where you get a Pulitzer Prize yet, still working on that.... :-)
You can see some examples of 98%-100% AI written fiction at Astral Fibers here on substack
Thanks Space Cadet, taking a look.
“How can I get 1,000 paid subs?” how???
I'm working on a piece on this. I want to do it very well so it's truly worth reading. Stay tuned!
I look forward to this one. One of my goals for this year is to write more consistently and really hone it as a skill. Witnessing the growth of your newsletter and your development as a writer is super-encouraging. Thank you for sharing.
What role do you think data protection has at the "speed of innovation" in global AI? Biden is finally catching up to spamoflage and other problems with amok ML modeling collecting and referring targets for foreign interference. https://www.bloomberg.com/news/articles/2024-01-23/biden-seeks-to-stop-countries-from-exploiting-americans-data-for-espionage
Hey Sheila, what do you mean by "role ... at the speed of innovation"?
Well, the persistently flawed narrative out there is that The Public must sacrifice IP observation, data privacy and other legal protectionism effects for humans for the pure sake of "innovation". OR Is there anything other than privacy roadkill for the innovators who don't care who is necessarily harmed by their products and innovation? The current standard is to deliberately make invasive pervasive ubiquitously dominant AI the governance standard ahead of the ability to legally govern it. That is the going rate of character of the business side innovation we have now. They're ready to nuke privacy right now - no looking back- because they have cannibalized people's personal information to their detriment for years. Just want to know if you seen any correction in the "hearts and minds" of the people making generative AI on this.
Many of the people I know use ChatGPT for search. As do I. Indeed, if I had to name a "killer app" for the technology that's what I would say. What in your view are the pros and cons of using the technology for that purpose? Anything interesting on the horizon?
Good question Fred.
I don't do it for one simple reason: I can't be sure I'm getting the info correctly. In many cases, one wrong data point is very problematic. If it had clear and direct references to where it's getting the info from, it'd be better. But it's hard to trust something that confabulates before saying "I don't know." Perhaps Perplexity or Bing are better options for search.
The pros are clear: if you have an "intelligent database of all knowledge on the internet," who wouldn't use it? It could be a super search engine.
For now, we're in the early adopters camp and the use is highly dependent on the goal of the search.
I've had great success with Perplexity for search-related tasks. ChatGPT does *ok* at search, however, you have to manually verify the relevance of links returned, as well as ensure that it's not merely hallucinating links which don't exist. You can also use Bing w/its version of GPT4 for search.
Plus one from me on Perplexity.AI.
They had a coupon at the new year that gave you 2 months of their "pro" access for free. You can still use the GPT-4 model with their service if you prefer and I have found that it's really pretty dang good. It will ask a couple clarifying questions to help guide the search and cites all of the sources well.
I was skeptical and wouldn't have tried it without the free trial. That said, over the course of this month I've stopped using google entirely and now use ChatGPT more for when I want to have a conversation vs a web-search/answer.
I use Midjourney to create AI generative images. I would like to use AI to create story scenarios around the images - e.g., start with an image of a man and a woman are sitting at a table at a sidewalk cafe. I would like an AI application that would evaluate the image and give possible interpretations such as: 1) they are brother and sister discussing family business; 2) they are undercover police watching somebody; 3) they just met and are starting to get romantically involved. So two questions: 1) is there an AI application that does this now; and 2) if not, is there a corpus of fiction texts - novels, short stories, advertisements, etc - that could be used as training data? Any suggestions or pointers will be appreciated.
I don't know of any tool that does this, but I don't think it'd be that hard to create a GPT that did it. You really only need Dalle and ChatGPT and ask ChatGPT to play a game of guessing the scene from an image.
I don't know if ChatGPT can do this. It wouldn't surprise me.
It does however work great in the opposite direction. I started with having ChatGPT Plus write a story, and now I'm illustrating the story with the included Dalle. Dalle really kicks ass, imho.
I'll take a paragraph out of the story that describes some scene, and then tell Dalle to create an image to match the description. So far, for me at least, it nails it every time.
Hi Alberto, love your Substack for a number of reasons. One of which you are an AI expert of Latine background, and I work for a national (U.S.) Latine nonprofit that is trying to wrap their minds around how to operationalize AI. Not easy! On another matter, have you read Bostrom’s “Superintelligence”? What is your take on Yudkowsky’s (and others) fast AI takeoff/“AI ruin” scenario after we achieve AGI? Thanks!
Hi Alberto,
With the exponential rise of AI as the new buzzword across all industries, there is a LOT of activity out there about "AI Trainings" of all sorts. my 2c is, while for those of us with a maker spirit the best way to learn is to start experimenting / trial and error, for the "feint of heart" there is a lot of emotional and functional barriers to adoption, so professional trainings are likely to play a role in ramping up adoption.
What do you think about the training offerings popping up out there? What do you think adds value for whom, and more broadly speaking, what do you think will be the main factors for accelerating adoption beyond the "trial and error" phase?
I'm pretty wary of courses which claim to teach you how to use ChatGPT or other AI tools. For me the best way to learn, is by doing. Meaning, sitting in front of my computer and using the tool. I find that a lot of training courses are way too superficial and high-level to provide much depth. But I may be an outlier here--a lot of people seem to like these kinds of things.
This is exactly the essence of my question to Alberto: what makes a good enabler for adoption for those who are too [word of choice] to try and err.
Hi Alberto,
I cannot praise your content enough. Your writing is invaluable for keeping up to date on developments in the tech scene. Particularly handy for the underground commute where I can't access the internet.
I've recently moved to London to pursue a career in tech/automation ethics and policy. I completed my bachelor's in politics at the end of last year. I've spent the last few years (and currently) bartending while studying, so I've been looking for a more office admin type job to transition to the more corporate space. I'll have to wait three years before I can take a master's degree at local rates, so I'm desperate to gain some practical, career aligned experience in the meantime; hopefully more than just a generic office job.
Do you (or anyone in the forum) have any advice for those of us trying to break into this industry?
What sort of job do you want to do? You mention "automation ethics and policy". To me that sounds like you want to work at a think tank or similar organization. One suggestion would be to write a description of the things you want to do with your job, and feed it into ChatGPT. Ask ChatGPT: "given this description, what kinds of job titles should I be searching for?" Then, with your list of job titles start searching for employers.