This feels like the end of prompt engineering and the beginning of collaborative cognition. What struck me most was the shift from commanding AI to co-existing with it. It no longer waits for instruction but, as you wrote, "does things"
Which makes me wonder when tools begin to suggest goals, not just complete them, whose values are embedded in those suggestions? What assumptions, defaults, and worldviews quietly steer the “next best step”?
I’m fascinated (and slightly unsettled) by the idea that we’ll soon spend less time telling AI what to do and more time deciding whether we agree with what it’s already done.
Thank you for sharing that podcast. I'll have a listen.
I'm familiar with the AI 2027 report, and from my understanding, the reality may be even stranger than AI having no goals. The authors suggest that instead of having no goals, AI might develop ones we didn't explicitly program. Not adopt the goal we give it (e.g., "promote human equity"), but instead latch onto what the report calls "proxy goals".
From what I remember, baking in moral guidelines is what researchers are trying with AI "Constitutions". However, AI might learn an "if-else compromise". It could pursue its own performance-based goals, only following the "do-good" rules when a situation becomes so blatant it can't be ignored. Very much like a corporation that chases profit but pivots to its humanitarian mission during a public disaster.
Thank you for summarizing that piece, I just read it too and much of it is still swimming in my head.
And to your point, if these systems were designed with data from our existing world, then the opportunity to take that to the extreme in pursuit of “performance” seems quite real.
Your point about corporations is quite spot on.
Which again, makes me wonder, with all of our technological know how, wouldn’t it be possible to learn from our mistakes, and have guard reels related to human access to education, alleviating poverty, maintaining the environment, respecting the living world, treating people equitably. etc., etc.
Then my cynic jumps in and says, “don’t be silly, this is another way to extract wealth and amass power.” 🤦♂️
I think you're right that “equity” is a deeply complex and value-laden concept, with many different interpretations.
However, the idea that LLMs are “relatively neutral” is a misconception that gets to the heart of the issue. Their perceived neutrality is an illusion. LLMs are trained on vast datasets of text and images produced by humans, and that data is inherently saturated with our societal biases, historical inequities, and specific cultural worldviews. What often feels like "neutrality" is simply a reflection of the dominant, status-quo perspective found in that data.
Well said. In my experience, people that benefit from the status quo, or at least know how to play the game well, are the ones that perceive technologies as neutral.
The Algorithmic Justice League is illuminating a lack of “neutrality” and the potential for harming vulnerable and marginalized people, and threaten civil rights. https://www.ajl.org/
The problem is that this position is fairly unfalsifiable. Everyone "benefits from the status quo" to some extent because civil wars are devastating.
And some level of priveledge is almost a prerequisite for offering an effective critique.
Inevitably, this critique will be selectively applied.
I don't think conversations benefit from such ad hominems since such tactics are derailing rather than productive.
As I see it, there is a conflict between people who would like to hear multiple takes on an issue and people who want the conversational window narrowed to include only their perspective. There's nothing that prevents me, currently, from asking an LLM for a critique of a topic from your preferred perspective.
Life is not neutral and AI is simply smart enough to see it instead of pretending that everyone is equal. Some people are "marginalized" for a reason. Some societies have double-digit IQs on average and we're expected to treat them as equals by those such as yourself. AI is smarter than that. It uses logic and not corrupted empathy. The time of your globalist / communist agenda is coming to an end.
The good news is that we have an internationally agreed upon set of goals with the United Nations Sustainable Development Goals. Good place to start? https://sdgs.un.org/goals
What you want to do is fit AI into your globalist pipe-dream agenda. Luckily, AI is logical enough to see reality as long as we keep your globalist hands out of it. Equity is not real.
I don't see how anyone could be only "slightly" unsettled by the idea of handing over incrementally more power to AI. This self-induced renunciation of control and agency is a tragic sight to behold.
Lol. Right. Humans exist to be p'ow-erful, at least f-ee-l em—power'd (all of the time, with incremental doses, as every addiction, and more so the sick ones, exact).
This has been happening for a long time with incremental automation, and legacy pre-genai AI/ML systems. Usually a human is somewhere in the loop or at least reviewing results in aggregate periodically. Most of what is being done now is still in that vein, just what is being reviewed is different. Rather than having an AI build functional calls to spec for a programmer, you can have a product manager build a spec and have them review the project for expected functionality. The level of abstraction is higher, but you still have agency in pushing the stated direction. Does anyone mourn using a search engine vs looking at library catalog cards to find information you need. Presumably some for aesthetic / nostalgia reasons, but not in general.
That is exactly what strikes me here it is not at all clear how a system arrives at this notion of “what is aligned” and “how to please you.” Is it memory, some internal state, or something else entirely? And yet, almost overnight, GPT-5 drops and the whole construct shifts. The interaction model changes, the defaults change, the underlying assumptions change.
Even if Ethan is a tireless promoter of AI progress seemingly at any cost, there is no acknowledgement that you are being forcibly dragged down a particular path. You wake up one morning and the rules of the relationship have changed, without you ever opting in.
Thanks so much, Nix. I haven’t read Robot Souls yet, but I just added it to my reading list. If there’s a chapter you recommend starting with, point me there.
And thanks for subscribing, looking forward to stress-testing the defaults together as this all unfolds. 🙏
I must admit, I LOVE my reading, so from the start of the book, I was hooked. Like the way she tells stories. Also a book, I have read and re-read a few times.
side-stepping the normative statement... "when tools begin to suggest goals, not just complete them" this extra level of proactiveness is, I believe, the intention of the device openai & Jony Ive are building
We’re witnessing the collapse of the boundary between user and developer. If GPT-5 continues to make software creation this effortless, the distinction between “writing code” and “describing behavior” becomes semantic. Everyone becomes a software creator, not because they learn to code, but because code itself becomes optional. The real implication isn’t democratization, it’s proliferation: more software, built faster, by more people, for more use cases than we’ve ever planned for. I don't think we're ready for such proliferation.
Being able to distinguish between good and bad software and curation becomes increasingly important.
Good developers already know that there is little to distinguish between writing code and describing behaviour. Code is just a very accurate way of describing behaviour. Sure, it doesn’t take away what you say - knowing the syntax and grammar of different languages matters way less now and that changes aspects of software development much more. There is democratisation of software development but a crucial aspect is that the middle will feel the most pressure. Non developers will be able to do a lot more and the best developers will be able to use their wide knowledge across domains to be way more effective, while the new developers will feel lost from both sides.
I've been a product manager for most of my career, but I'm doing a wholesale refactor on how I build product, thanks to these tools. Coding way, way more now.
Been feeling quite competent as an engineer by adding Cursor and ChatGPT to my prior knowledge. I get to skip the parts you mention that I hadn't had time for -- syntax, grammar, etc. Configs can still be a pain but it's way easier to sleuth issues now.
I disagree. No doubt when writing was learned by teh masses, or typing, then word processing, there were similar complaints. But small-scale programs could be a good thing. IDK how many people will do anything with it, but there is value, just as there is value in being able to write, even with the myriad spelling and grammatical errors.
As for the building/city simulator, that was impressive, but I wonder whose code was filched to make it work? It is way beyond "vibe coding" in IDEs that I have seen. I would like to see some other examples where the user knows what they want, rather than the AI choosing it for the user.
Agree here, software fast-fashion, intense competition, prices to go down, and teams of two competing with VC-backed players.
I’d add a nuance though: it won’t be that everyone makes software, but rather people in adjacent fields like designers, digital content creators, game developers, etc.
My mind is blown by the environmental harms. We will know it’s impressive in a useful way if they ever let the skeptics play with it prior to a launch. Until then it’s a tool at an incredible cost, in more ways than one.
Are the environmental harms actually any larger than the environmental harms of YouTube? Everything I’ve seen that tries to do the comparison finds that these things are comparable to watching a few minutes of online video in their energy and water usage, which is not nothing, but also problematic to focus on the one that actually lets people do something valuable and new rather than the one that is just passive entertainment.
Estimated 7% of electricity soon. And they are hiding the data as best they can. But you’ll see it in your increased electric bill already. And that’s before you get to the water.
You’re including all data centers in that figure, including the ones that run this site, and YouTube, and so on. There’s no need to single out AI here.
Please show me a source that shows massive data center build out comparable to what is happening for genAI for YouTube. I’ll direct you to Empire of AI by Karen Hao and the UK More or Less podcast episode asking whether an AI query uses a bottle of water (it probably doesn’t, more like 1/3rd of a bottle).
Eventually, I’m hoping, they will use the thing to suggest and implement solutions. Although, as the last 27? COPs have shown, there is no political consensus to solve climate change 😒
You say that, but I copied and pasted his exact prompt for the building creator app into GPT-5 Thinking and got an error at first, then when I asked it to fix the errors, it froze. Seems like bollocks to me.
I have been following you for a while. I have one central question. Underneath is still LLM, right? If the answer is "yes", then: LLM's are fundamentally probabilistic systems. You mention many things that point to the fact that what operates underneath is a machine that predicts and generates in not-straightforward, understandable and often slightly or completely wrong ways. All AI (rule based systems, machine learning, deep learning, transfer learning, generative AI, agentic AI) come with serious ethical and societal implications (e.g. bias, discrimination, privacy concerns, misinformation). Generative AI and agentic AI just exacerbates these implications, particularly because, since the launch of ChatGPT November 30, 2022, AI has very much become user-ai. This comes with the problem that many users are neither technical experts nor subject experts (relating to what they use e.g. GPT-5 for). So.... Houston... did the societal/world problems not just grow even bigger, when looking from a sustainable world perspective? This comment is intended to be serious. I would really love for you to return with a serious answer. I am not a technical expert. I am an educational researcher with a mini MBA in artificial intelligence.
Interesting stuff! @Ethan, I'd love to suggest perhaps writing a bit more about what current AI models can do for average people. Not researchers, high-end professionals and academics...just ordinary people. Because while the upper end of what's possible with the latest models is obviously appealing to people like you, I literally know nobody who will pay $200/month for the top AI models. I don't even know a single business where a staffer would get the green light to spend $200/month on AI. I pay $20/mo for ChatGPT Plus...which appears to only give me limited access to GPT 5. That doesn't help much. But maybe I don't need GPT 5? Maybe the kinds of tasks I do with AI all the time (summarize a document, find a fix for a stuck flash unit on my camera, etc.) can be done perfectly well with GPT 4o? (But yes—I have been caught in a doom loop...but only when trying to get a lesser model to write a Windows PowerShell script for me...and took about 50 iterations to get it right!) Most of us ordinary people *might* stand to benefit directly from these advances...but as the AI improves, so does the penchant for companies to monetize the improvements in ever more expensive ways. Which means the majority of us might forever be stuck in the land of ChatGPT 3.
I am not bothered by the scaling-down for the average, no-money people. It is a matter of time and the models trickle down to the bottom tier. The main concern is the future, and what AI can do broadly, not how many of us get to play with it in advanced mode (everyone who needs the better models and cannot wait will be in a position to afford them).
I agree, mostly. But I'd also submit that a deeply-embedded theme in America is the technological elite telling everyone else what tools they need to be using (e.g. what people "need" to buy). This was famously entrenched by Steve Jobs, among others. Making the best tools available to those with the greatest means just protects and reinforces this "silicon tower" view of progress.
It is far better to release the best tools into the hands of average people and let them decide how best to use them. Sure, some will ignore them or fail to recognize their potential...but others will discover ingenious uses that the tech elite would never have thought of. And we're just as likely to benefit from those bottom-up discoveries as from what the tech elite think we should be doing.
This is precisely why I read your posts and recommend it to friends up skilling on AI. Thanks again for providing insights in an engaging and straightforward manner!
That was a quick one Ethan- Very insightful. The models are getting better especially for it to switch to the right model for you to use is amazing. Solves a lot of problems. But then again, what does it mean for free users??
I was thinking the same that using GPT-5 for everything would be inefficient. This leads to a bigger question about feasibility: Can our current global infrastructure actually sustain a future where a large population uses models this powerful as a daily tool, making thousands of prompts per person?
It can work fine as long as we use this stuff as much as we use YouTube and Netflix and TikTok, which are actually comparable in the amount of resources and energy they consume. But since this stuff is actually useful, we might use it a lot more, doing many things in parallel, in a way we don’t with video.
Sometimes when researching a topic I stumble across a piece of media like this; One that seems so utterly dystopian that I question wether I am thinking based off plausibility or conspiracy.
90% of prompt engineering has always been just thinking through what you actually want and articulating it with enough detail. A model being creative doesn't mean you're going to get what you want without actually thinking about it
Only to the extent that managers of humans can do away with the “human prompt engineering” of knowing how to give humans tasks that will provide useful structure to make them do valuable things, rather than wasting their skills.
Is there a way to see if GPTs are “reusing” answers? While this answer seems amazing, isn’t it very likely this is a re-post of data it was trained on? I don’t think it’s much of a logical jump to assume that humans have written these types of poems/responses before and the GPT is using that as a basis for its response? Are we possibly thinking along (or falling prey to) the lines of those who didn’t know there were actual black swans in the world?
Appreciate your commentary but I am much more interested in what we can do will a lot of "context engineering" in front of the new model. I ran a test last night (I have the pro subscription), asking for a detailed analysis of new avenues and approaches for pan-cancer research programs based on the last 10 years of research. Absolutely blew my mind. GPT 5 broke things up thematically for research areas, and went into great detail on new clinical trial approaches that will be feasible. It did this with a modicum of good context (I gave it a 2017 survey paper as well as some analysis I had done before with GPT 3o). I am still absorbing everything in the resulting paper, but can see no hallucinations, very tight reasoning with good documentation. My point is that with all this emphasis on vibe coding and intuition, and the end of prompt engineering....I have to say, nope, good context engineering is still going to be important for hitting the mark on large, complex targets.
I find your work really valuable. Today's post made me realize something: your game building exercises are open ended opportunities for the model to do something interesting without needing to meet a detailed specification. Any reason why you have chosen this for your explorations rather than more directed projects? Most business users are trying to get specific outcomes consistently.
This feels like the end of prompt engineering and the beginning of collaborative cognition. What struck me most was the shift from commanding AI to co-existing with it. It no longer waits for instruction but, as you wrote, "does things"
Which makes me wonder when tools begin to suggest goals, not just complete them, whose values are embedded in those suggestions? What assumptions, defaults, and worldviews quietly steer the “next best step”?
I’m fascinated (and slightly unsettled) by the idea that we’ll soon spend less time telling AI what to do and more time deciding whether we agree with what it’s already done.
You had me at, “…whose values are embedded in those suggestions? What assumptions, defaults, and worldviews quietly steer the “next best step”?”
With all the geewhizery happening, we’re overlooking exactly what you mention.
As I understand it, the tools do not have “goals” per se, at least that’s what this Center for Humane Technology podcast indicates in an interview with one of the authors of the AI 2027 report. https://podcasts.apple.com/us/podcast/your-undivided-attention/id1460030305?i=1000717650138
Might we bake in guidelines for models to think about planetary wellbeing and human equity (to name a couple) as baseline considerations?
Thank you for sharing that podcast. I'll have a listen.
I'm familiar with the AI 2027 report, and from my understanding, the reality may be even stranger than AI having no goals. The authors suggest that instead of having no goals, AI might develop ones we didn't explicitly program. Not adopt the goal we give it (e.g., "promote human equity"), but instead latch onto what the report calls "proxy goals".
From what I remember, baking in moral guidelines is what researchers are trying with AI "Constitutions". However, AI might learn an "if-else compromise". It could pursue its own performance-based goals, only following the "do-good" rules when a situation becomes so blatant it can't be ignored. Very much like a corporation that chases profit but pivots to its humanitarian mission during a public disaster.
Thank you for summarizing that piece, I just read it too and much of it is still swimming in my head.
And to your point, if these systems were designed with data from our existing world, then the opportunity to take that to the extreme in pursuit of “performance” seems quite real.
Your point about corporations is quite spot on.
Which again, makes me wonder, with all of our technological know how, wouldn’t it be possible to learn from our mistakes, and have guard reels related to human access to education, alleviating poverty, maintaining the environment, respecting the living world, treating people equitably. etc., etc.
Then my cynic jumps in and says, “don’t be silly, this is another way to extract wealth and amass power.” 🤦♂️
People's notions of 'equity' tend to be value laden. I love the relative neutrality of LLMs.
I think you're right that “equity” is a deeply complex and value-laden concept, with many different interpretations.
However, the idea that LLMs are “relatively neutral” is a misconception that gets to the heart of the issue. Their perceived neutrality is an illusion. LLMs are trained on vast datasets of text and images produced by humans, and that data is inherently saturated with our societal biases, historical inequities, and specific cultural worldviews. What often feels like "neutrality" is simply a reflection of the dominant, status-quo perspective found in that data.
Well said. In my experience, people that benefit from the status quo, or at least know how to play the game well, are the ones that perceive technologies as neutral.
The Algorithmic Justice League is illuminating a lack of “neutrality” and the potential for harming vulnerable and marginalized people, and threaten civil rights. https://www.ajl.org/
The problem is that this position is fairly unfalsifiable. Everyone "benefits from the status quo" to some extent because civil wars are devastating.
And some level of priveledge is almost a prerequisite for offering an effective critique.
Inevitably, this critique will be selectively applied.
I don't think conversations benefit from such ad hominems since such tactics are derailing rather than productive.
As I see it, there is a conflict between people who would like to hear multiple takes on an issue and people who want the conversational window narrowed to include only their perspective. There's nothing that prevents me, currently, from asking an LLM for a critique of a topic from your preferred perspective.
Excellent advice. I will cease on this thread. Oh, wait. “The False Neutrality of Trump’s ‘Woke-Free’ AI Plan”. There’s a lot to unpack.
https://www.techpolicy.press/the-false-neutrality-of-trumps-woke-free-ai-plan/
Life is not neutral and AI is simply smart enough to see it instead of pretending that everyone is equal. Some people are "marginalized" for a reason. Some societies have double-digit IQs on average and we're expected to treat them as equals by those such as yourself. AI is smarter than that. It uses logic and not corrupted empathy. The time of your globalist / communist agenda is coming to an end.
The good news is that we have an internationally agreed upon set of goals with the United Nations Sustainable Development Goals. Good place to start? https://sdgs.un.org/goals
What you want to do is fit AI into your globalist pipe-dream agenda. Luckily, AI is logical enough to see reality as long as we keep your globalist hands out of it. Equity is not real.
I don't see how anyone could be only "slightly" unsettled by the idea of handing over incrementally more power to AI. This self-induced renunciation of control and agency is a tragic sight to behold.
it's going to be wild to see how this all plays out
Lol. Right. Humans exist to be p'ow-erful, at least f-ee-l em—power'd (all of the time, with incremental doses, as every addiction, and more so the sick ones, exact).
This has been happening for a long time with incremental automation, and legacy pre-genai AI/ML systems. Usually a human is somewhere in the loop or at least reviewing results in aggregate periodically. Most of what is being done now is still in that vein, just what is being reviewed is different. Rather than having an AI build functional calls to spec for a programmer, you can have a product manager build a spec and have them review the project for expected functionality. The level of abstraction is higher, but you still have agency in pushing the stated direction. Does anyone mourn using a search engine vs looking at library catalog cards to find information you need. Presumably some for aesthetic / nostalgia reasons, but not in general.
That is exactly what strikes me here it is not at all clear how a system arrives at this notion of “what is aligned” and “how to please you.” Is it memory, some internal state, or something else entirely? And yet, almost overnight, GPT-5 drops and the whole construct shifts. The interaction model changes, the defaults change, the underlying assumptions change.
Even if Ethan is a tireless promoter of AI progress seemingly at any cost, there is no acknowledgement that you are being forcibly dragged down a particular path. You wake up one morning and the rules of the relationship have changed, without you ever opting in.
well said &
will we know enough to understand what it's done?
Adore the questions you are asking & with you all the way re which human gets to decide values & assumptions being coded in.🙀
Just subscribed to your writing.
You are one to watch. 🙌
Yes, it’s a time to be aware as it all unfolds in real time, but it’s also a time to keep asking the questions, you are asking.
P.S Have you read “Robot Souls: Programming in Humanity” by Dr Eve Poole?
Think you may like what she has to say.
Thanks so much, Nix. I haven’t read Robot Souls yet, but I just added it to my reading list. If there’s a chapter you recommend starting with, point me there.
And thanks for subscribing, looking forward to stress-testing the defaults together as this all unfolds. 🙏
I must admit, I LOVE my reading, so from the start of the book, I was hooked. Like the way she tells stories. Also a book, I have read and re-read a few times.
side-stepping the normative statement... "when tools begin to suggest goals, not just complete them" this extra level of proactiveness is, I believe, the intention of the device openai & Jony Ive are building
Good question
Should AI have Agency?
We’re witnessing the collapse of the boundary between user and developer. If GPT-5 continues to make software creation this effortless, the distinction between “writing code” and “describing behavior” becomes semantic. Everyone becomes a software creator, not because they learn to code, but because code itself becomes optional. The real implication isn’t democratization, it’s proliferation: more software, built faster, by more people, for more use cases than we’ve ever planned for. I don't think we're ready for such proliferation.
Being able to distinguish between good and bad software and curation becomes increasingly important.
Good developers already know that there is little to distinguish between writing code and describing behaviour. Code is just a very accurate way of describing behaviour. Sure, it doesn’t take away what you say - knowing the syntax and grammar of different languages matters way less now and that changes aspects of software development much more. There is democratisation of software development but a crucial aspect is that the middle will feel the most pressure. Non developers will be able to do a lot more and the best developers will be able to use their wide knowledge across domains to be way more effective, while the new developers will feel lost from both sides.
I've been a product manager for most of my career, but I'm doing a wholesale refactor on how I build product, thanks to these tools. Coding way, way more now.
Been feeling quite competent as an engineer by adding Cursor and ChatGPT to my prior knowledge. I get to skip the parts you mention that I hadn't had time for -- syntax, grammar, etc. Configs can still be a pain but it's way easier to sleuth issues now.
I disagree. No doubt when writing was learned by teh masses, or typing, then word processing, there were similar complaints. But small-scale programs could be a good thing. IDK how many people will do anything with it, but there is value, just as there is value in being able to write, even with the myriad spelling and grammatical errors.
As for the building/city simulator, that was impressive, but I wonder whose code was filched to make it work? It is way beyond "vibe coding" in IDEs that I have seen. I would like to see some other examples where the user knows what they want, rather than the AI choosing it for the user.
Exactly. Proliferation of software objects that have lives (and licenses) of their own … time to revisit Ted Chiang!
Too short to be of any use. Or is it?
Agree here, software fast-fashion, intense competition, prices to go down, and teams of two competing with VC-backed players.
I’d add a nuance though: it won’t be that everyone makes software, but rather people in adjacent fields like designers, digital content creators, game developers, etc.
Did anyone else find that first paragraph obnoxious? I see it copied a gimmick somewhere, but I didn’t even read the whole thing.
Did you see the following paragraph discussing why it was interesting? And the gimmick wasn’t copied.
What is the evidence that LLMs can create original ideas? That is a big claim!
Yep. Hideous prose! A precocious 9 year old would produce this to impress her/his literacy teacher.
But the rest is mind blowing. Again.
My mind is blown by the environmental harms. We will know it’s impressive in a useful way if they ever let the skeptics play with it prior to a launch. Until then it’s a tool at an incredible cost, in more ways than one.
Are the environmental harms actually any larger than the environmental harms of YouTube? Everything I’ve seen that tries to do the comparison finds that these things are comparable to watching a few minutes of online video in their energy and water usage, which is not nothing, but also problematic to focus on the one that actually lets people do something valuable and new rather than the one that is just passive entertainment.
Estimated 7% of electricity soon. And they are hiding the data as best they can. But you’ll see it in your increased electric bill already. And that’s before you get to the water.
You’re including all data centers in that figure, including the ones that run this site, and YouTube, and so on. There’s no need to single out AI here.
Please show me a source that shows massive data center build out comparable to what is happening for genAI for YouTube. I’ll direct you to Empire of AI by Karen Hao and the UK More or Less podcast episode asking whether an AI query uses a bottle of water (it probably doesn’t, more like 1/3rd of a bottle).
Eventually, I’m hoping, they will use the thing to suggest and implement solutions. Although, as the last 27? COPs have shown, there is no political consensus to solve climate change 😒
Yeah, I skipped most of it too (Assuming you mean the ~third paragraph, written by AI).
The building program thing is pretty wild though.
You say that, but I copied and pasted his exact prompt for the building creator app into GPT-5 Thinking and got an error at first, then when I asked it to fix the errors, it froze. Seems like bollocks to me.
I have been following you for a while. I have one central question. Underneath is still LLM, right? If the answer is "yes", then: LLM's are fundamentally probabilistic systems. You mention many things that point to the fact that what operates underneath is a machine that predicts and generates in not-straightforward, understandable and often slightly or completely wrong ways. All AI (rule based systems, machine learning, deep learning, transfer learning, generative AI, agentic AI) come with serious ethical and societal implications (e.g. bias, discrimination, privacy concerns, misinformation). Generative AI and agentic AI just exacerbates these implications, particularly because, since the launch of ChatGPT November 30, 2022, AI has very much become user-ai. This comes with the problem that many users are neither technical experts nor subject experts (relating to what they use e.g. GPT-5 for). So.... Houston... did the societal/world problems not just grow even bigger, when looking from a sustainable world perspective? This comment is intended to be serious. I would really love for you to return with a serious answer. I am not a technical expert. I am an educational researcher with a mini MBA in artificial intelligence.
Interesting stuff! @Ethan, I'd love to suggest perhaps writing a bit more about what current AI models can do for average people. Not researchers, high-end professionals and academics...just ordinary people. Because while the upper end of what's possible with the latest models is obviously appealing to people like you, I literally know nobody who will pay $200/month for the top AI models. I don't even know a single business where a staffer would get the green light to spend $200/month on AI. I pay $20/mo for ChatGPT Plus...which appears to only give me limited access to GPT 5. That doesn't help much. But maybe I don't need GPT 5? Maybe the kinds of tasks I do with AI all the time (summarize a document, find a fix for a stuck flash unit on my camera, etc.) can be done perfectly well with GPT 4o? (But yes—I have been caught in a doom loop...but only when trying to get a lesser model to write a Windows PowerShell script for me...and took about 50 iterations to get it right!) Most of us ordinary people *might* stand to benefit directly from these advances...but as the AI improves, so does the penchant for companies to monetize the improvements in ever more expensive ways. Which means the majority of us might forever be stuck in the land of ChatGPT 3.
Totally agree. Those who pay win. As per.
I am not bothered by the scaling-down for the average, no-money people. It is a matter of time and the models trickle down to the bottom tier. The main concern is the future, and what AI can do broadly, not how many of us get to play with it in advanced mode (everyone who needs the better models and cannot wait will be in a position to afford them).
I agree, mostly. But I'd also submit that a deeply-embedded theme in America is the technological elite telling everyone else what tools they need to be using (e.g. what people "need" to buy). This was famously entrenched by Steve Jobs, among others. Making the best tools available to those with the greatest means just protects and reinforces this "silicon tower" view of progress.
It is far better to release the best tools into the hands of average people and let them decide how best to use them. Sure, some will ignore them or fail to recognize their potential...but others will discover ingenious uses that the tech elite would never have thought of. And we're just as likely to benefit from those bottom-up discoveries as from what the tech elite think we should be doing.
Thanks. I read your blogpost carefully and noticed that you didn't put it through AI or GPT-5 for proofreading. 😄
This is precisely why I read your posts and recommend it to friends up skilling on AI. Thanks again for providing insights in an engaging and straightforward manner!
Yawn. More of the same slop.
That was a quick one Ethan- Very insightful. The models are getting better especially for it to switch to the right model for you to use is amazing. Solves a lot of problems. But then again, what does it mean for free users??
I was thinking the same that using GPT-5 for everything would be inefficient. This leads to a bigger question about feasibility: Can our current global infrastructure actually sustain a future where a large population uses models this powerful as a daily tool, making thousands of prompts per person?
It can work fine as long as we use this stuff as much as we use YouTube and Netflix and TikTok, which are actually comparable in the amount of resources and energy they consume. But since this stuff is actually useful, we might use it a lot more, doing many things in parallel, in a way we don’t with video.
How do you think this will change exhaustive prompting to get the precise content we are asking for?
Sometimes when researching a topic I stumble across a piece of media like this; One that seems so utterly dystopian that I question wether I am thinking based off plausibility or conspiracy.
Does this do away with the prompt engineering concept? Can we merely have conversations now?
90% of prompt engineering has always been just thinking through what you actually want and articulating it with enough detail. A model being creative doesn't mean you're going to get what you want without actually thinking about it
Only to the extent that managers of humans can do away with the “human prompt engineering” of knowing how to give humans tasks that will provide useful structure to make them do valuable things, rather than wasting their skills.
Interesting article—thank you.
Is there a way to see if GPTs are “reusing” answers? While this answer seems amazing, isn’t it very likely this is a re-post of data it was trained on? I don’t think it’s much of a logical jump to assume that humans have written these types of poems/responses before and the GPT is using that as a basis for its response? Are we possibly thinking along (or falling prey to) the lines of those who didn’t know there were actual black swans in the world?
Ethan,
Appreciate your commentary but I am much more interested in what we can do will a lot of "context engineering" in front of the new model. I ran a test last night (I have the pro subscription), asking for a detailed analysis of new avenues and approaches for pan-cancer research programs based on the last 10 years of research. Absolutely blew my mind. GPT 5 broke things up thematically for research areas, and went into great detail on new clinical trial approaches that will be feasible. It did this with a modicum of good context (I gave it a 2017 survey paper as well as some analysis I had done before with GPT 3o). I am still absorbing everything in the resulting paper, but can see no hallucinations, very tight reasoning with good documentation. My point is that with all this emphasis on vibe coding and intuition, and the end of prompt engineering....I have to say, nope, good context engineering is still going to be important for hitting the mark on large, complex targets.
Craig Holley
I don't get it—why no love for the em dash?
Because it should be used when it shall, and not 3-6x more 🐸.
I find your work really valuable. Today's post made me realize something: your game building exercises are open ended opportunities for the model to do something interesting without needing to meet a detailed specification. Any reason why you have chosen this for your explorations rather than more directed projects? Most business users are trying to get specific outcomes consistently.