I was thinking about this while biking a local mountain and listening to Fahrenheit 451. Everything text based is going to change. Did I really read that incredibly long article I submitted to...
I was thinking about this while biking a local mountain and listening to Fahrenheit 451.
Everything text based is going to change.
Did I really read that incredibly long article I submitted to Tildes, or did I just like the Kagi Summary?
Was that book review written by a person who actually read the book, or an AI bot?
Did my boss send me that reply, or was it autogenerated by AI?
Is that inflammatory social media post genuine, or AI generated nonsense? Am I debating a human or a bot intended to cause political division?
Clearly AI is ready to disrupt a number of industries. Email responses. Customer Service. Content Generation. Coding. The real disruption will be how this makes our lives better or worse.
Worse. It will make it worse. I used to be a techno optimist, of the breed of 80s and 90s nerds that thought information wanted to be free, that the internet would democratize learning and...
Worse. It will make it worse. I used to be a techno optimist, of the breed of 80s and 90s nerds that thought information wanted to be free, that the internet would democratize learning and dismantle the old hegemonies, that embracing the open-source movement and internalizing the rugged rebellious hacker spirit would free everyone and usher in a better world where people were rational, educated, and reasonable.
That hasn't happened at all. The technology that created the internet, built largely by people from Caltech and Stanford and MIT with that same attitude was co-opted by corporations and people who used it to get insanely rich and tighten control over society instead.
I lack any sort of optimism for AI either. Its already controlled by large corporations. Open-source implementations of large language models aren't even close, and the computing power they require to train will always mean that they lag behind closed source corporate backed implementations. The potential for cutting labor costs using large language models is absolutely huge, and our economic system isn't set up to distribute wealth created by efficiency gains by its very design. Every increase in efficiency will continue to just make a few people richer and everyone else poorer.
I fear that things are going to get a lot, lot worse.
I already feel like we live in a pretty bad cyberpunk setting; the kind of places I read about as a kid and thought "wow that's truly awful, luckily no one would allow corporations to become that powerful", but we literally live there now. I can't help but think it's going to get even more unbelievably oppressive.
Just gonna throw in that the publicly available and permissively licensed stable diffusion text-to-image model is comparable to DALL-E 2, made by the multi-billion dollar company OpenAI.
Just gonna throw in that the publicly available and permissively licensed stable diffusion text-to-image model is comparable to DALL-E 2, made by the multi-billion dollar company OpenAI.
I'm not sure if I've maybe fallen all the way through that thinking and out the other side, because I very much know what you mean and I've had some pretty similar thoughts on tech in general, but...
I'm not sure if I've maybe fallen all the way through that thinking and out the other side, because I very much know what you mean and I've had some pretty similar thoughts on tech in general, but I've got a sense of something you could very justifiably call either optimism or fatalism that says there's no going back now, so the only hope is to keep pushing forward until we as a society figure things out.
I think the really transformative (pun intended) part of GPT and it's kin is that they offer a more naturalistic way of learning. For a "traditional" search engine, it is keyword based. The more...
I think the really transformative (pun intended) part of GPT and it's kin is that they offer a more naturalistic way of learning.
For a "traditional" search engine, it is keyword based. The more keywords you know about a topic, the faster and (hopefully) more accurate your results. When you're at least moderately fluent in a topic, this is fine. However starting out with no keywords, you can try asking google about things but it's going to be fairly unhelpful unless your exact question is frequently asked or you stumble onto a relevant forum.
Being able to ask natural language questions, and get a response that sounds similar to if you had asked a professor/mentor/peer who knows more than you about topic is incredibly useful. Being able to ask followup questions that take your previous question into context is incredible. Google is mostly (setting aside some very basic customization based on your search history) idotempitc ideotempatic idiotempic idempotent 1, it does not take your previous search into account when you search for something else.
Google and search engines transformed how everyone at all levels of society learned, studied, retained knowledge, and transferred knowledge. ChatGPT and it's derivatives seems like the next major iteration in human's ability to access knowledge. I can't begin to speculate how it will transform society.
[1] I couldn't remember this word, so (obviously) I asked ChatGPT for it. It took 2 tries
The difference between Tom Scott and me: I figured out that labels issue six years ago, and I've been tagging my shit properly since. In fact, I had even built up the habit to NEVER use the...
The difference between Tom Scott and me: I figured out that labels issue six years ago, and I've been tagging my shit properly since. In fact, I had even built up the habit to NEVER use the archive button, and always re-tag; and it got to that point where I didn't know why I was doing it like that anymore...
until this video reminded me why. Thank, Tom.
You’re still calling ChatGPT a tech demo?
Buddy, it’s not a demo, it’s a warning of what’s to come.
It wasn't my barber who unpromptedly told me about "using ChatGPT to write an email". It was an architect I work with. The guy's not a techie, but he was telling me he used it for emails in mid december. And two weeks ago he was showing me various AI-powered apps he's been using.
I feel a lot of what Tom's going through. I wonder though, if I were his age (he is only eight years older than me), would I also feel that existential dread at being "behind the future"? Because that is the one thing I am not connecting with at all in this video. I'm excited. AGI might actually happen in my working lifetime. Hey, maybe space travel's next.
I’m even closer to the same age, and I’m excited too. I’m not yet convinced that AGI is coming, but even if we level off short of that the changes this technology brings are going to be massive. I...
I’m even closer to the same age, and I’m excited too. I’m not yet convinced that AGI is coming, but even if we level off short of that the changes this technology brings are going to be massive. I literally had a conversation about AI, sigmoid curves, and the similarities to the early internet with a friend who was staying with me over new year, and while we both acknowledged that any major change brings upheaval, the main takeaway was that we can’t wait to see what happens.
On a very, very tangential aside, Tom Scott specifically seems to capture my thinking on most things closely enough that it’s actually kind of jarring when a video like this diverges from how I feel about the topic! I’m not sure if there’s a word for it - not quite parasocial, it’s not that I’m under the impression that I know the guy personally and understand how he thinks or anything - more a kind of subversion of expectations after learning to assume from most videos that his take will mirror my own.
[Edit] Come to think of it, I wonder to what extent apprehension is proportional to satisfaction with the status quo here. I’m already fortunate in a lot of ways, for sure, but part of my excitement is absolutely the thought of how I might ride this wave to change my life for the better. If I were already a successful public figure spending my days travelling the world to document interesting places and people, maybe I’d be less enthusiastic about a possible shake up?
Yeah. I have already used this technology professionally and plan to quit my job this year to work with my best friend on a new AI startup. So my enthusiasm is filled with hopes and dreams of...
Yeah. I have already used this technology professionally and plan to quit my job this year to work with my best friend on a new AI startup. So my enthusiasm is filled with hopes and dreams of early retirement.
I'm not very interested in things like chatGPT or Dall-E. They don't do a great deal that's actually new, they just do old stuff in a slightly novel way. A human can do the things they do (and do...
I'm not very interested in things like chatGPT or Dall-E. They don't do a great deal that's actually new, they just do old stuff in a slightly novel way. A human can do the things they do (and do it better). Sure, there's a great technical achievement in creating such things and they're certainly a step along a very specific road which likely ends with a load of low-level word/pixel pushing jobs disappearing as a result but still, not very interesting to me.
ML/AI in other applications where they're doing things humans can't do has much more potential. Autodesk's generative design stuff is fascinating and that's pretty old tech already. AI logistics systems are already online and making massive differences to how things work, saving time, money and resources in a way humans couldn't do. Medical applications, engineering, design, planning, all sorts - anything where there is complexity and patterns to be spotted and exploited and optimised. Of course a lot of that stuff is already happening, because there is a lot of money to be made with even a few percentage points of improvements.
I think true AGI is as far away now as the James Webb Space Telescope was from that day in 1766 when Newton presented his telescope to the Royal Society. Probably further. Even with Carmack on the case. There's a big question of how we'd even know if we had made an AGI, because we barely know what a "GI" looks or works like, despite arguably being one ourselves.
Tangent to your tangent, I used to hang out on b3ta at the same time as Tom and do vaguely know him (well, vaguely did, 20+ years ago) so it's definitely weird when that thing happens to me.
I think even on the text/image generation side, the impact of being four orders of magnitude faster and cheaper than a human for mostly good enough quality is likely to cause some pretty...
I think even on the text/image generation side, the impact of being four orders of magnitude faster and cheaper than a human for mostly good enough quality is likely to cause some pretty interesting ripples in the way we do things - either way though you're absolutely right, there are whole horizons of really cool uses that seem a bit more esoteric to most people and don't generate anything like the same publicity compared to their impact, and it's going to be fascinating to see the changes they bring.
Tangent to your tangent, I used to hang out on b3ta at the same time as Tom and do vaguely know him (well, vaguely did, 20+ years ago) so it's definitely weird when that thing happens to me.
Now that's a name I've not heard in a long time! You're making me wonder how many overlaps there are in the Venn diagram of "elder millennial British computer nerd" culture now... I'm not surprised that link adds an extra layer to the strange for you!
I expect I am a decade younger than you, and I feel that same sense of dread. He articulated it very nicely. Space travel would be rather cool, though.
if I were his age (he is only eight years older than me), would I also feel that existential dread at being "behind the future"? Because that is the one thing I am not connecting with at all in this video. I'm excited. AGI might actually happen in my working lifetime. Hey, maybe space travel's next.
I expect I am a decade younger than you, and I feel that same sense of dread. He articulated it very nicely. Space travel would be rather cool, though.
When you grow up at the cutting edge of technology, you wouldn't be able but to invest a part of your identity into it. At the time, there's no worry the thing might crumble. To not be able to be...
would I also feel that existential dread at being "behind the future"?
When you grow up at the cutting edge of technology, you wouldn't be able but to invest a part of your identity into it. At the time, there's no worry the thing might crumble.
To not be able to be at the cutting edge anymore because you haven't been for a while and a thing made it very clear for you... That's the "middle-age crisis" territory.
(Though, for what it's worth, Tom's handling it very well. Being one of the most popular pop-sci channels on YouTube must be feeling good. His delivery hits the nail, too. Not bad for not being at the vanguard, huh?)
I connect with the existential dread a bit. Tom is a few years younger than me, but I think because I work in higher ed and interact with students regularly I've already seen some signs of how I...
I connect with the existential dread a bit. Tom is a few years younger than me, but I think because I work in higher ed and interact with students regularly I've already seen some signs of how I could easily be left behind as technology and society continue to change rapidly. So I understand his dread, but I am also kind of excited by how AI/ML can potentially improve everyone's lives.
True AGI though, terrifies me. Computers move at amazing speeds, that's kind of why we built them. Setting aside the classic movie tropes about AI turning on it's creators, there are plenty of other ways an AGI could destroy mankind as a side effect. The thing that terrifies me is the speed with which it could happen. By it's nature, an AI could go from helpful to malignant in the literal blink of an eye, and we might not even notice until hours or days have gone by and the entire US power grid (or whatever) is offline.
In the Internet age almost all software depends heavily on the internet to operate. It's considered basic functionality to toss bits around the world. In the Transformer age I think it will be the...
In the Internet age almost all software depends heavily on the internet to operate. It's considered basic functionality to toss bits around the world. In the Transformer age I think it will be the same. You can use them as part of your UI and even as part of your app's logic. I've prototyped a system for converting natural language into procedures that use your app's internal API. Think about a world where everything has at least a basic if-this-then-that built in.
"Tildes, if I get a DM from Deimos respond with a message that I'm out of town"
Think about how much feature work gets taken care of with that one piece of functionality. Open AI's Codex model is really good at things of that complexity. There are already apps showcasing natural language interfaces to small bits of code, like a natural language website navigator (I forget the name of the company but it's pretty impressive). And that might just be the beginning.
I don't think you need machine learning to implement auto-reply, but it might help you find the setting to do it. There are often search boxes for settings. Also, I will often use a Google search...
I don't think you need machine learning to implement auto-reply, but it might help you find the setting to do it. There are often search boxes for settings. Also, I will often use a Google search to find out how to set something.
Siri and Google Assistant were supposed to solve this. Perhaps they will get better?
That's not about using machine learning to implement auto-reply, it's about using machine learning to implement much of the runtime code for your site. That's auto-reply and also a million other...
That's not about using machine learning to implement auto-reply, it's about using machine learning to implement much of the runtime code for your site. That's auto-reply and also a million other one-off single user features.
It's not just about auto reply. It's about huge amounts of automation and functionality. Right now it's possible to add features that do just about anything the user could want, but at some point,...
It's not just about auto reply. It's about huge amounts of automation and functionality. Right now it's possible to add features that do just about anything the user could want, but at some point, they become so undiscoverable that they may as well not exist.
Large language models will make it so the average person is able to access this huge set of functionality easily. I imagine the LLM won't be the last step, we will probably use it to generate some kind of automation script which will be shown to the user to confirm.
I'm imagining letting users use an LLM to write code (even if at several removes) that runs on my systems and now I have got that feeling of dread Tom was talking about. But for very different...
I'm imagining letting users use an LLM to write code (even if at several removes) that runs on my systems and now I have got that feeling of dread Tom was talking about. But for very different reasons.
I was thinking about this while biking a local mountain and listening to Fahrenheit 451.
Everything text based is going to change.
Did I really read that incredibly long article I submitted to Tildes, or did I just like the Kagi Summary?
Was that book review written by a person who actually read the book, or an AI bot?
Did my boss send me that reply, or was it autogenerated by AI?
Is that inflammatory social media post genuine, or AI generated nonsense? Am I debating a human or a bot intended to cause political division?
Clearly AI is ready to disrupt a number of industries. Email responses. Customer Service. Content Generation. Coding. The real disruption will be how this makes our lives better or worse.
Worse. It will make it worse. I used to be a techno optimist, of the breed of 80s and 90s nerds that thought information wanted to be free, that the internet would democratize learning and dismantle the old hegemonies, that embracing the open-source movement and internalizing the rugged rebellious hacker spirit would free everyone and usher in a better world where people were rational, educated, and reasonable.
That hasn't happened at all. The technology that created the internet, built largely by people from Caltech and Stanford and MIT with that same attitude was co-opted by corporations and people who used it to get insanely rich and tighten control over society instead.
I lack any sort of optimism for AI either. Its already controlled by large corporations. Open-source implementations of large language models aren't even close, and the computing power they require to train will always mean that they lag behind closed source corporate backed implementations. The potential for cutting labor costs using large language models is absolutely huge, and our economic system isn't set up to distribute wealth created by efficiency gains by its very design. Every increase in efficiency will continue to just make a few people richer and everyone else poorer.
I fear that things are going to get a lot, lot worse.
I already feel like we live in a pretty bad cyberpunk setting; the kind of places I read about as a kid and thought "wow that's truly awful, luckily no one would allow corporations to become that powerful", but we literally live there now. I can't help but think it's going to get even more unbelievably oppressive.
Just gonna throw in that the publicly available and permissively licensed stable diffusion text-to-image model is comparable to DALL-E 2, made by the multi-billion dollar company OpenAI.
I'm not sure if I've maybe fallen all the way through that thinking and out the other side, because I very much know what you mean and I've had some pretty similar thoughts on tech in general, but I've got a sense of something you could very justifiably call either optimism or fatalism that says there's no going back now, so the only hope is to keep pushing forward until we as a society figure things out.
I think the really transformative (pun intended) part of GPT and it's kin is that they offer a more naturalistic way of learning.
For a "traditional" search engine, it is keyword based. The more keywords you know about a topic, the faster and (hopefully) more accurate your results. When you're at least moderately fluent in a topic, this is fine. However starting out with no keywords, you can try asking google about things but it's going to be fairly unhelpful unless your exact question is frequently asked or you stumble onto a relevant forum.
Being able to ask natural language questions, and get a response that sounds similar to if you had asked a professor/mentor/peer who knows more than you about topic is incredibly useful. Being able to ask followup questions that take your previous question into context is incredible. Google is mostly (setting aside some very basic customization based on your search history)
idotempitc ideotempatic idiotempicidempotent 1, it does not take your previous search into account when you search for something else.Google and search engines transformed how everyone at all levels of society learned, studied, retained knowledge, and transferred knowledge. ChatGPT and it's derivatives seems like the next major iteration in human's ability to access knowledge. I can't begin to speculate how it will transform society.
[1] I couldn't remember this word, so (obviously) I asked ChatGPT for it. It took 2 tries
The difference between Tom Scott and me: I figured out that labels issue six years ago, and I've been tagging my shit properly since. In fact, I had even built up the habit to NEVER use the archive button, and always re-tag; and it got to that point where I didn't know why I was doing it like that anymore...
until this video reminded me why. Thank, Tom.
"A warning of what's to come" -- I knew I had used that exact phrasing before. On the HN thread about John Carmack leaving Meta to work on AGI full-time.
It wasn't my barber who unpromptedly told me about "using ChatGPT to write an email". It was an architect I work with. The guy's not a techie, but he was telling me he used it for emails in mid december. And two weeks ago he was showing me various AI-powered apps he's been using.
I feel a lot of what Tom's going through. I wonder though, if I were his age (he is only eight years older than me), would I also feel that existential dread at being "behind the future"? Because that is the one thing I am not connecting with at all in this video. I'm excited. AGI might actually happen in my working lifetime. Hey, maybe space travel's next.
I’m even closer to the same age, and I’m excited too. I’m not yet convinced that AGI is coming, but even if we level off short of that the changes this technology brings are going to be massive. I literally had a conversation about AI, sigmoid curves, and the similarities to the early internet with a friend who was staying with me over new year, and while we both acknowledged that any major change brings upheaval, the main takeaway was that we can’t wait to see what happens.
On a very, very tangential aside, Tom Scott specifically seems to capture my thinking on most things closely enough that it’s actually kind of jarring when a video like this diverges from how I feel about the topic! I’m not sure if there’s a word for it - not quite parasocial, it’s not that I’m under the impression that I know the guy personally and understand how he thinks or anything - more a kind of subversion of expectations after learning to assume from most videos that his take will mirror my own.
[Edit] Come to think of it, I wonder to what extent apprehension is proportional to satisfaction with the status quo here. I’m already fortunate in a lot of ways, for sure, but part of my excitement is absolutely the thought of how I might ride this wave to change my life for the better. If I were already a successful public figure spending my days travelling the world to document interesting places and people, maybe I’d be less enthusiastic about a possible shake up?
Yeah. I have already used this technology professionally and plan to quit my job this year to work with my best friend on a new AI startup. So my enthusiasm is filled with hopes and dreams of early retirement.
I'm not very interested in things like chatGPT or Dall-E. They don't do a great deal that's actually new, they just do old stuff in a slightly novel way. A human can do the things they do (and do it better). Sure, there's a great technical achievement in creating such things and they're certainly a step along a very specific road which likely ends with a load of low-level word/pixel pushing jobs disappearing as a result but still, not very interesting to me.
ML/AI in other applications where they're doing things humans can't do has much more potential. Autodesk's generative design stuff is fascinating and that's pretty old tech already. AI logistics systems are already online and making massive differences to how things work, saving time, money and resources in a way humans couldn't do. Medical applications, engineering, design, planning, all sorts - anything where there is complexity and patterns to be spotted and exploited and optimised. Of course a lot of that stuff is already happening, because there is a lot of money to be made with even a few percentage points of improvements.
I think true AGI is as far away now as the James Webb Space Telescope was from that day in 1766 when Newton presented his telescope to the Royal Society. Probably further. Even with Carmack on the case. There's a big question of how we'd even know if we had made an AGI, because we barely know what a "GI" looks or works like, despite arguably being one ourselves.
Tangent to your tangent, I used to hang out on b3ta at the same time as Tom and do vaguely know him (well, vaguely did, 20+ years ago) so it's definitely weird when that thing happens to me.
I think even on the text/image generation side, the impact of being four orders of magnitude faster and cheaper than a human for mostly good enough quality is likely to cause some pretty interesting ripples in the way we do things - either way though you're absolutely right, there are whole horizons of really cool uses that seem a bit more esoteric to most people and don't generate anything like the same publicity compared to their impact, and it's going to be fascinating to see the changes they bring.
Now that's a name I've not heard in a long time! You're making me wonder how many overlaps there are in the Venn diagram of "elder millennial British computer nerd" culture now... I'm not surprised that link adds an extra layer to the strange for you!
I expect I am a decade younger than you, and I feel that same sense of dread. He articulated it very nicely. Space travel would be rather cool, though.
Apologies, but:
Username checks out. :D
When you grow up at the cutting edge of technology, you wouldn't be able but to invest a part of your identity into it. At the time, there's no worry the thing might crumble.
To not be able to be at the cutting edge anymore because you haven't been for a while and a thing made it very clear for you... That's the "middle-age crisis" territory.
(Though, for what it's worth, Tom's handling it very well. Being one of the most popular pop-sci channels on YouTube must be feeling good. His delivery hits the nail, too. Not bad for not being at the vanguard, huh?)
I connect with the existential dread a bit. Tom is a few years younger than me, but I think because I work in higher ed and interact with students regularly I've already seen some signs of how I could easily be left behind as technology and society continue to change rapidly. So I understand his dread, but I am also kind of excited by how AI/ML can potentially improve everyone's lives.
True AGI though, terrifies me. Computers move at amazing speeds, that's kind of why we built them. Setting aside the classic movie tropes about AI turning on it's creators, there are plenty of other ways an AGI could destroy mankind as a side effect. The thing that terrifies me is the speed with which it could happen. By it's nature, an AI could go from helpful to malignant in the literal blink of an eye, and we might not even notice until hours or days have gone by and the entire US power grid (or whatever) is offline.
In the Internet age almost all software depends heavily on the internet to operate. It's considered basic functionality to toss bits around the world. In the Transformer age I think it will be the same. You can use them as part of your UI and even as part of your app's logic. I've prototyped a system for converting natural language into procedures that use your app's internal API. Think about a world where everything has at least a basic if-this-then-that built in.
"Tildes, if I get a DM from Deimos respond with a message that I'm out of town"
Think about how much feature work gets taken care of with that one piece of functionality. Open AI's Codex model is really good at things of that complexity. There are already apps showcasing natural language interfaces to small bits of code, like a natural language website navigator (I forget the name of the company but it's pretty impressive). And that might just be the beginning.
I don't think you need machine learning to implement auto-reply, but it might help you find the setting to do it. There are often search boxes for settings. Also, I will often use a Google search to find out how to set something.
Siri and Google Assistant were supposed to solve this. Perhaps they will get better?
That's not about using machine learning to implement auto-reply, it's about using machine learning to implement much of the runtime code for your site. That's auto-reply and also a million other one-off single user features.
It's not just about auto reply. It's about huge amounts of automation and functionality. Right now it's possible to add features that do just about anything the user could want, but at some point, they become so undiscoverable that they may as well not exist.
Large language models will make it so the average person is able to access this huge set of functionality easily. I imagine the LLM won't be the last step, we will probably use it to generate some kind of automation script which will be shown to the user to confirm.
I'm imagining showing the user a block diagram / workflow like with the iPhone's shortcuts system.
I'm imagining letting users use an LLM to write code (even if at several removes) that runs on my systems and now I have got that feeling of dread Tom was talking about. But for very different reasons.