Next task, get ChatGPT to be funny. If you ask it for a comedy bit / standup routine / roast about a topic, it'll always end with some lame paragraph that hedges everything, and directly telling it to cut that out doesn't seem to work either. :D
I would consider using prompt-writing under the right circumstances, but I have not yet found a good use case in my personal business life. I do think there's a lot of time saving value here, for what it's worth - I just haven't been able to make a convincing use case where I have to do a lot of writing, but it only kind of matters what the writing says or how it says it.
A few years ago, I read a report (whitepaper may be too formal) from Bridgewater, Ray Dalio's investing firm, on conditions in something like 50 different countries in various stages of development. It was perfect for this because the same info was in each report, just rewritten slightly to correct for the different regions. That's like the ideal use case for this.
I think I follow only about half of that! But yeah, the prompts themselves aren't as important as the idea behind the approaches. It's just a bunch of ways to nudge ChatGPT & co. out of the usual AI speak.
I guess I'm saying that even a well-prompted AI is still gonna be a pretty lousy writer, at least in terms of the types of measures I use for writing.... but there are really good cases where it doesn't need to be a good writer in order to do what you want. I would not use AI for an update to my website, for instance (not yet, anyway), but I would probably use it for an internal document that isn't public facing, like something only employees and managers can see.
Yeah I personally haven't gotten to a place where I can use AI output "as is" for any of my writing, either.
But for someone who wants to use AI to give outlet to written texts they otherwise wouldn't (like I can use Midjourney to make images I'd never be able to draw myself), I think it's nice to have a few tools in their arsenal to nudge something less generic out of LLMs.
Yes, absolutely! You're pointing folks in the right direction. Gradually, I expect more and more use cases for AI writing with decent prompts, as the areas where we've been reluctant to experiment gradually begin to disappear over the next year or two.
These are great tips Daniel, thanks. I've saved this and will be putting it to use.
Can AI content be improved? ..... Yes.
Does AI content suck?.... No.
The endlessly repeated memorized mantra of "AI content sucks" is a function of the Substack community, a very specific group of people trying to make a living as writers. That is, the LEAST OBJECTIVE group of people on the Internet, when it comes to the subject of AI.
There are a million different contexts in which words are used to communicate online. AI in it's current form is suitable for some contexts, and not suitable for other contexts. Thus, talking about AI content as if it were a single thing is silly.
Your article above illustrates this beautifully. If a person uses ChatGPT in an ignorant manner (as I have often done) then the results will be modest. If the same person take the time to learn about this new technology, the results can improve considerably, and thus be suitable for a wider range of contexts.
The range of contexts in which AI is suitable is very likely to continually expand over time. Thus, endlessly chanting "AI sucks" is a kind of a stupid way for Substackers to prepare themselves for what seems to be an inevitable future.
Your article is great Daniel, a very useful addition. But it would have been a bit better if you weren't also feeding the Substack Myth Machine.
I think in most cases all these GenAI chatbots are not perfect without human input, but humans are not known for being perfect either. I think most of us who have spent time working with these tools and learning about them know that you can't chuck them a crappy prompt and get a magical response.
I feel like I've learned a lot from people like Chantal Smith at Exponential View, books by Ethan Mollick and Reid Hoffman, and so many more writers and creators in this space, and I feel like one of the consistent themes is around the way we work *with* these tools.
I love this set of words from Ethan Mollick, especially the cyborg part:
"As I have discussed before, the most valuable way to start use AI for work is to become a Centaur or Cyborg … Centaur work has a clear line between person and machine, like the clear line between the human torso and horse body of the mythical centaur.
On the other hand, Cyborgs blend machine and person, integrating the two deeply. Cyborgs don’t just delegate tasks; they intertwine their efforts with AI, moving back and forth over the Jagged Frontier. Bits of tasks get handed to the AI, such as initiating a sentence for the AI to complete, so that Cyborgs find themselves working in tandem with the AI."
Oh yeah, I also enjoyed Ethan's "Centaurs & Cyborgs" post!
I agree that prompts matter, but I'm also cautious of making prompt engineering sound like inexplicable and precise science. I encourage an exploratory mindset in anyone starting out with large language models, AI images, etc.
That's why I like the "minimum viable prompt" approach - it gets people quickly dipping their toes into the water, then lets them built up their experience over time.
Oh, I'm on the same page - there's no voodoo magic needed to be good at prompting. Having said that, I still enjoy learning how to step up my game from people smarter than me who have spent more time going deeper and then shared tips.
Next task, get ChatGPT to be funny. If you ask it for a comedy bit / standup routine / roast about a topic, it'll always end with some lame paragraph that hedges everything, and directly telling it to cut that out doesn't seem to work either. :D
I actually did a Sunday Showdown where Pi and ChatGPT told jokes (https://www.whytryai.com/i/142722992/sunday-showdown-issue-chatgpt-vs-pi-which-one-tells-the-best-jokes)
But yeah, LLMs aren't particularly great at humor, although there are occasional moments of chuckleworthy stuff.
I would consider using prompt-writing under the right circumstances, but I have not yet found a good use case in my personal business life. I do think there's a lot of time saving value here, for what it's worth - I just haven't been able to make a convincing use case where I have to do a lot of writing, but it only kind of matters what the writing says or how it says it.
A few years ago, I read a report (whitepaper may be too formal) from Bridgewater, Ray Dalio's investing firm, on conditions in something like 50 different countries in various stages of development. It was perfect for this because the same info was in each report, just rewritten slightly to correct for the different regions. That's like the ideal use case for this.
I think I follow only about half of that! But yeah, the prompts themselves aren't as important as the idea behind the approaches. It's just a bunch of ways to nudge ChatGPT & co. out of the usual AI speak.
I guess I'm saying that even a well-prompted AI is still gonna be a pretty lousy writer, at least in terms of the types of measures I use for writing.... but there are really good cases where it doesn't need to be a good writer in order to do what you want. I would not use AI for an update to my website, for instance (not yet, anyway), but I would probably use it for an internal document that isn't public facing, like something only employees and managers can see.
Ah, now I get it!
Yeah I personally haven't gotten to a place where I can use AI output "as is" for any of my writing, either.
But for someone who wants to use AI to give outlet to written texts they otherwise wouldn't (like I can use Midjourney to make images I'd never be able to draw myself), I think it's nice to have a few tools in their arsenal to nudge something less generic out of LLMs.
Yes, absolutely! You're pointing folks in the right direction. Gradually, I expect more and more use cases for AI writing with decent prompts, as the areas where we've been reluctant to experiment gradually begin to disappear over the next year or two.
These are great tips Daniel, thanks. I've saved this and will be putting it to use.
Can AI content be improved? ..... Yes.
Does AI content suck?.... No.
The endlessly repeated memorized mantra of "AI content sucks" is a function of the Substack community, a very specific group of people trying to make a living as writers. That is, the LEAST OBJECTIVE group of people on the Internet, when it comes to the subject of AI.
There are a million different contexts in which words are used to communicate online. AI in it's current form is suitable for some contexts, and not suitable for other contexts. Thus, talking about AI content as if it were a single thing is silly.
Your article above illustrates this beautifully. If a person uses ChatGPT in an ignorant manner (as I have often done) then the results will be modest. If the same person take the time to learn about this new technology, the results can improve considerably, and thus be suitable for a wider range of contexts.
The range of contexts in which AI is suitable is very likely to continually expand over time. Thus, endlessly chanting "AI sucks" is a kind of a stupid way for Substackers to prepare themselves for what seems to be an inevitable future.
Your article is great Daniel, a very useful addition. But it would have been a bit better if you weren't also feeding the Substack Myth Machine.
I'm happy you find the tips useful!
Great post Daniel, especially the examples. It's all about our prompts, which leads me back to the theme of copilot mode.
Thanks man! What are you referring to by the "copilot mode" in this context?
I think in most cases all these GenAI chatbots are not perfect without human input, but humans are not known for being perfect either. I think most of us who have spent time working with these tools and learning about them know that you can't chuck them a crappy prompt and get a magical response.
I feel like I've learned a lot from people like Chantal Smith at Exponential View, books by Ethan Mollick and Reid Hoffman, and so many more writers and creators in this space, and I feel like one of the consistent themes is around the way we work *with* these tools.
I love this set of words from Ethan Mollick, especially the cyborg part:
"As I have discussed before, the most valuable way to start use AI for work is to become a Centaur or Cyborg … Centaur work has a clear line between person and machine, like the clear line between the human torso and horse body of the mythical centaur.
On the other hand, Cyborgs blend machine and person, integrating the two deeply. Cyborgs don’t just delegate tasks; they intertwine their efforts with AI, moving back and forth over the Jagged Frontier. Bits of tasks get handed to the AI, such as initiating a sentence for the AI to complete, so that Cyborgs find themselves working in tandem with the AI."
Oh yeah, I also enjoyed Ethan's "Centaurs & Cyborgs" post!
I agree that prompts matter, but I'm also cautious of making prompt engineering sound like inexplicable and precise science. I encourage an exploratory mindset in anyone starting out with large language models, AI images, etc.
That's why I like the "minimum viable prompt" approach - it gets people quickly dipping their toes into the water, then lets them built up their experience over time.
Oh, I'm on the same page - there's no voodoo magic needed to be good at prompting. Having said that, I still enjoy learning how to step up my game from people smarter than me who have spent more time going deeper and then shared tips.
100%. Speaking of, have you seen Anthropic's new Prompt Library? A very useful reference: https://docs.anthropic.com/claude/prompt-library
Yep, I like a lot about Anthropic. Have you seen this table from Exponential View, the last one under the Finally section?
https://www.exponentialview.co/p/promptpack-conversation-engineering