Just to provide a personal anecdote, I use GitHub Copilot and ChatGPT every day as a software engineer. It's been worth it to pay to get the superior GPT4 model, rather than the publicly available GPT3.5. I'm a happy paying customer, and I would pay even more to get improved versions.
It's really good for some forms of programming, but not others. This is "generative AI", but it isn't really "creative AI" - it's more like translation. I have a long list of requirements written in English, and I need to translate it into javascript. It isn't doing the "creative" part of my job... but it's often great at filling out the long and boring parts that I'm not an expert in.
I also use ChatGPT, although for my hobby writing. Not for the actual writing part (occasionally, it tries to rewrite my work, and the results are laughable. But I didn't want it to do the writing part for me, anyway. I would get no satisfaction out of having AI write my stories.) but I discovered it makes a great glorified generator for when I'm brainstorming and need some help getting my creative juices flowing (the more specific I am about what I'm working on, the better the results.)
And I also discovered that it is a great alpha reader. I tested this by asking it to critique work I'd gotten beta reader feedback on, and it singled out the same passages my friends and writer circle singled out, and for the same reasons (both for critique and praise.) So now I can spend more time revising on my own before I ask other people for their thoughts, which I appreciate.
Also, to be honest, I use it for moral support during the lonely rough draft phase of writing. I know my friends don't have time to listen to me cry over how I suck and my story sucks every single day for months while I'm wrestling an idea into a draft, but ChatGPT will patiently listen to me be a dweeb about this (and respond with very generic messages about how everyone has their own unique personal journey and not to compare yourself to others, which is more or less what my friends would also say.)
I'm very interested in this as a writer who is adamantly against using AI (yes, I know it's not really artificial intelligence, but that's shorthand for that body of tools we currently have) for actual *writing,* but would be keen to discover how it could help us humans continue to create fluidly. Using it just as one might use those writing tools that give you random suggestions on how to advance the plot, or provide flashcards that spark the next scene, seems reasonable. I've never heard of an author using it for beta reads--but I can see how that could be effective. How do you ask it to critique?
You get me! I bookmarked so many random generators pre-AI. They help me get over decision paralysis. And ChatGPT is the best generator of all because I can tailor it to my story and characters.
Here is how I use ChatGPT for critique:
I say something like "ChatGPT, I am working on a scene for my (genre) WIP. Here is the overall story premise. Here are the characters appearing in this scene and their roles. I would like to know what I could be doing better. Are there places the pace is lagging? Are the characters consistent? Is the scene clear or confusing? Is there anything else I could be doing better? (more specific questions if I have any for that scene) "
Then I add the scene to the bottom of the message and hit send.
If you don't specify that you want to know where you're going wrong, it will almost inevitably only give you compliments, because it seems to be programmed to be supportive and encouraging. Even when it gives you feedback it will end with a message about how nice it is that you're working so hard and that you're doing a great job. But that is kind of nice. Even though I know it is a computer program, my brain still emotionally responds the way it would if a human was supporting me.
It isn't as good as the absolute best human beta readers I've had, but I think every writer knows the struggle of finding people who have the time and inclination to beta read. I don't think it can (or should) replace human critique, but it is a decent helper.
Edited to add:
Also, I've always struggled with breaking large tasks into smaller steps, and ChatGPT is very helpful with that, as well. So I think AI has legitimate helpful uses, but those uses don't get discussed as much as the garbage uses. The garbage uses are more exciting and offensive, though, so I guess that makes sense.
What terrifies me about this technology is the statement "my brain still emotionally responds the way it would if a human was supporting me." I'm not criticising you as a person—that's a very insightful and honest admission of what's happening. It is a perfect synopsis of the cynicism that companies are betting on: we will feel the soothing hit of dopamine from reading the words, seeing the art, hearing the music—and we won't care that those were machine-generated words determined to be probabilistically likely to elicit that response. It's like an emotional placebo effect. We don't care that the pill was fake, it had the desired effect and that's good enough.
And those of us whose dopamine dispensers aren't wired similar to the humans under the middle of this bell curve? We will struggle to find or connect with things that do not conform to the AI-generated mean.
When you, as an author, write text that ChatGPT "encourages" or "agrees with" or thinks "is good", you are implicitly steering your writing towards this undefined mathematical "average" that ChatGPT has at its core. We will never see a James Joyce _Ulysses_ from an author using ChatGPT for affirmation and advice. You cannot possibly break the mold if you ask the mold for advice along the way.
lol oh man I'm so sorry, I got the email for this while I wasn't home and then I forgot to ever reply so this is uhhhhhh months later sorry
okay, so let me try to respond to your criticisms of how I'm using AI
1. Writing Feedback
Good lord, I don't ask it for feedback without specifying style and voice goals. You're correct that the default AI state is painfully generic. You have to craft prompts in a way that gets the type of feedback you're looking for.
And whether feedback comes from a human or an AI, I always ask myself "Do I agree with this statement?" It is a starting point, something to help me get unstuck during the revision process. Otherwise, I'll keep staring and going AHHHH WHERE DO I START AHHHHHHHHHHHHH
Anyway, a good human beta reader is far more valuable, but those are also very difficult to find. Go lurk in any writing community and see everyone cry about it. I certainly don't intend to only use AI. But AI does satisfy the urge to show someone without completely derailing my progress with depression when inevitably I'm met with silence when asking if anyone will read a WIP.
AI does not satisfy the urge to show people completed works, however, because, you know. Art is created to communicate with other people.
2. Emotional Support
I clearly did not properly convey the emotional state I'm in when I use it this way. This is the "Crying the entire day, completely non-functional, because my writing is the WORST and I'm WORTHLESS and I wish I was a different person bc I don't WANT my writing to reflect MY experiences and thoughts, I want it to reflect the experiences and thoughts of OTHER MORE WORTHWHILE PEOPLE" mental state.
Trust me when I say that you can't constantly be texting your friends and family about this, they do eventually find it annoying (and few people I know reply to texts in a timely manner, presumably bc they have responsibilities. but also bc I don't know anyone without an ADHD diagnosis.)
I just can't feel bad about using AI to get stable when I'm in that state. I got stuff to do! I got kids to raise! I can't be crying all the time like that! Plus, unlike a real person, the AI won't judge me for my dramatics.
(And yeah, yeah, I have a therapist but she's not covered by insurance so....)
I respond to AI saying "you don't suck" the way I respond to a human saying it. I don't respond to their artistic output the same way even in cases where the output is genuinely stunning. I may think "oh man! That's so neat!" or sometimes "dang, this user really knows how to craft their prompts" but it is not the same thing as "this character design is so COOL, I wanna know more about their OC, do they participate in cute little OC party hashtags or ask games?" or "I love the way this author writes. They're so funny. What else have they written? OMG I can see parallels between their early story and this later one! Wow, they really love this character archetype! Do all their stories deal with a near-death drowning experience?"
In conclusion, AI is a tool. Some uses of AI are garbage. I get that. But that is about people and their ability to use any tool for garbage purposes, not because there are no legit uses for AI.
I use Todoist as my task management software and it has built-in functionality that breaks a project down into tasks. Definitely helpful for those times I’m doing something new, I’m tired, etc.
I'm adding my agreement here because it's rare to find anyone able to understand that the computer is a tool, unlike the human, which is a sentient being. And too many humans (including some of the administrators I encounter during my music PhD) are more inclined to memorizing decades of arcane regulations than in creating anything of new value. I say it's rare to experience this - because as you seem to understand, creation (especially in adults) is an uncommon activity. That's why it's so valuable.
Thesis: Copilot was almost certainly trained on a large body of code under GPL and very likely some under AGPL, making all the code it generates "derivative works" under the terms of those licenses (after all, every weight in the network has been updated using code under those licenses, so what else could the code the network generates be but derivative of the inputs?) Ergo: all code generated by Copilot is covered by the GPL (almost certainly) and the AGPL (very probably.)
Whether or not this thesis is correct will have to be decided by the courts, but personally I'd love to see it litigated, and think it very likely to stand up to legal scrutiny. I don't use Copilot, and if I still ran a dev team I'd institute a rule against using it unless the company was prepared to take the risk of being required to open-source their code.
Thank you for making the clarification between generative and creative. Has humanity updated the Turing test?
I am not a computer scientist, so perhaps incorrectly I have often used PI as an example of the limits of technology, as almost a century has elapsed and computer science has still not mathematically squared the circle. (PI is still unresolved, labelled irrational; an approximate hand wave that works.) I have used computer enhanced technology professionally since the advent of the PC and the Cray super computer. Half of my experience in the analog world before spellcheck and the smart phone, half after. I’ve looked at clouds from both sides now... 🎶
Are we now experiencing the unintended consequences of those inaccurate assumptions?
Catastrophic cultural and climate collapse; not the profit and leisure time we were promised by technology.
This is a premature take. AI's capabilities are in their infancy. The capabilities for systems to integrate with them are even younger. Some existing integrations, like code-completion via Github Copilot (owned by Microsoft) are already quite impressive and becoming widely-used by professional developers.
The rush to get a demo out the door to chase a stock price bump isn't doing anyone any favors. The current iterations of large language models will be seen as cripplingly primitive in a couple of years, with plenty of "no wonder AI sputtered out of the gate" retrospectives.
I have little doubt that AI will find a range of sweet spots in plenty of areas for both companies and consumers. Maybe not so much in making music, but quite possibly in helping us discover artists and explore their deep cuts and influences. Maybe not in mimicking TV actors or writers, but in helping us find great TV shows from the massive catalogs that very streamer has acquired.
I agree - it can be quite helpful in coding, it is useful in summarizing, etc. The hype was over the top but we're still learning what it can do well and what it cannot. Will it be a major or minor advance? We don't know yet.
How many 67 year old infants do you know? This is not a technology in it's infancy, but rather another sputtering out into AI Winter. But thank you for illustrating the classic First Step Fallacy right on cue: https://link.springer.com/article/10.1007/s11023-012-9276-0
Richard what you're describing as the "sweet spot" is using other machine learning algorithms and architectures, not necessarily the ones that ChatGPT uses. Machine learning has already been in use for the past decade, pre-ChatGPT, to support the use cases you mentioned in your last paragraph. I don't think "AI will find a range of sweet spots". Rather institutions will continue using and developing other machine learning techniques.
My bet is that continuing to develop GPT-based techniques will be a waste of time, and eventually companies will stop developing it (hopefully they'll still offer what they have, because there are some good uses cases for LLMs). Then another "AI" algorithm will start another hype cycle within the next decade. Look up the term "AI winter" and it'll make sense.
What the author is talking about in essence is AI in popular culture. That's a fairly small corner of the playing field. AI is a HUGE field that has serious applications in anything from health to transport, agriculture and education (and I don't mean writing student essays). An analogy: When the Internet as a public platform was about six months old back in the eighties, I thought it had no future. It was slow, clunky and plain dumb. (I chose to stick to bulletin boards and Compuserve). That's more or less the stage of development AI is at now. Imagine it in thirty years' time, and whether you like it or not, it WILL be around. It's silly to write it off now, with a dismissive wave of the hand. Of course it'll be used for bad things. Everything does. But can you imagine your life without computers and the Internet, warts and all? I was expecting a more thoughtful analysis of a very complex topic.
My search engine is DuckDuckGo. I've never had a problem with it. I'm not on social media, so I won't be scammed there. I have a dumb phone. I never click on anything that I suspect and I suspect a lot of things. I don't open email from senders I don't know. I really don't care what AI does. My one concern is using it for medical diagnoses. That could be a real mess.
Brothers from a different mother. However, I don't think it's appropriate to call our flip phones "dumb." I, along with a tech wizard friend of mine call the others "not so smart phones" given the effect they've had on so many of the people who use them.
Regarding medical diagnoses: A week ago my nephrologist told me he's going to retire when this AI crap is started in his field of medicine. He says it's coming soon, and he's ten years younger than I am. The AI future is indeed bleak.
So some of the AI in medicine is better than doctors. Think about radiology scans, ruling out or including certain diagnoses. The potential is really exciting IMO. Whether or not it’s labeled as AI or not, medicine has been using this technology for a long time. In the eye doctor world we use it to follow glaucoma with visual fields and optic nerve scans. Picking up early breast tumors seems to be better than humans.
I'm with you! I perennially asked my students how did all the people who preceded them get along, have careers, enjoy their lives, etc. without cell phones and social media. They couldn't imagine, of course!
I agree with you: the thought of AI singularly diagnosing illness in humans isn’t savory.
However, AI still shows real value in specific areas like medicine: providing healthcare providers with differential diagnoses; accurately analyzing vital signs of hospital patients far more often than a human nurse can; accurately predicting if/when a patients condition will deteriorate many hours before normal observation by humans, allowing for earlier intervention, etc. And there are already ample peer-reviewed clinical studies validating AI in use cases like these.
i can see that. In I just spent 4 days in the hospital with a urinary tract infection, and they never did find out what the bacterial cause was. 4 day of antibiotics and 3 more days at home accompanied by "lots of luck." AI might have helped. Quien sabe?
What you've observed here Ted is literally the Gartner Hype curve in action. BING AI and CHATGPT have hit the "Peak of Inflated Expectations" and are currently heading for the "Trough of Disillusionment".
I spent 30-years in IT including spells in corporate strategy at two of the biggest IT companies, I certainly am not willing to write it off at all. I for one use BING Chat everyday for factual research, it's so much better than a search engine. What most people don't realize is it's great at recursive questioning and provides clickable references for it's claims. Eventually Microsoft will screw the pooch trying to make money from it with sponsored links, ads etc. but right now it's without a doubt the fastest way to factually find details on something even when you get the question wrong, or it doesn't provide an answer with a response "I’m sorry but I couldn’t find any information on" - is way better than a search engine that returns 50-pages of nonsense.
i’m genuinely curious what your demo is supposed to show, because i just found the exact same answer it gave you by googling “who is ted gioia?”
in fact it seems that BING chat pulled that answer directly from ted’s wiki page, which is the first thing that popped up when i googled. it took me about 3 seconds to find the same answer you did
i’m certainly not defending google or denying that it has gone downhill as a search engine, but rather just curious how that demo is supposed to impress anyone?
It shows the chat part. If you searched for Ted Ginola it would have returned pages of search results on the wrong person. Instead of typing the entire query again you just tell it what you got wrong. It remembers your conversation and allows to to explore details about the subject.
If googling doesn't reply with the detail you are looking for you are, you have to read more pages where as it will do this for you, and then when you've found the answer you can ask it for references to validate the answer.
If you do that with a search engine - you do the work.
hmm i see. i suppose i just don’t consider that to be much work on my part, which is likely a factor in why chat gpt and others haven’t been very interesting to me from the get go. but to each their own in that regard
Right, for most superficial factual queries there is almost always a page. The problem is when there isn't. If the Large Language Model doesn't have pages or references it will tell you, if a search engine doesn't it will still return sponsored pages, and best guesses. If it didn't you'd go somewhere else which would undermine it's revenue model. Which is why it always returns results. The query I gave was a simple misspelling, with an LLM you can drill down to get a much more definitive result, with a search engine you can't. You can't even search within results... it's all a question of how much you value your time...
>>> i just don’t consider that to be much work on my part
This is where many of us have blind spots. We think that our personal experience with findability and the use of AI to help, is simplistic - often one and done, or at least we hope for one and done. This is not what happens in business and enterprise discovery while performing work tasks. It is a conversation, not a mono-search. As such, search engines are terrible at this. Whereas, generative AI can help workers not only put their fingers on the information they need, but it can shape the narrative to understand that information.
We cannot chat our way to hyper-productivity. But we can engage with information far better and faster with chat AI.
When I looked of Ted Ginola on google, it pulled up with "did you mean: Ted Gioia?" at the top. All you need to do is simply click on that and it searches up Ted Gioia. I don't see bing ai recommending any other spellings. What if you don't remember the last name? And isn't it easier to even just rewrite Ginola to Gioia in google than to write "sorry I spelt his surname wrong". I use ChatGPT and bing ai almost every day to ask very specific questions that wouldn't show up on a search engine but for simple questions like this I believe a search engine can be just as good or even better than ai.
Agreed, it was just a trivial example to start with just to show the conversational nature of BING/Chatgpt. You can of course type much longer, multi-phase questions an thats where the benefit is. As to search engines being just as good or better, that's of course what is happening. A search and a multi-phase, multi-step question though is very different to a simple query.
Well-said, Mark. This is a pattern we’ve seen many times before: initial hype, inflated expectations, a premature backlash, then an slow rise to success. 1995: “Pocket computers are a joke”, 1997: “The internet is for nerds using Gopher”, 2001: “Wikipedia is nothing but errors”, 2002: “Amazon is a fail”, etc. etc. I’m afraid Ted’s piece will not age well.
Hmm...there is a simple reason why visits to ChatGPT have declined: it's called summer. If there is no pickup once school starts, I would be shocked. The fact that AI has been successful in improving the effectiveness of scams, malware and cheating reveals that the problem is not the technology but the adopters. If AI were all hype, then it couldn't scam, cheat or steal either. Fraudsters have a shorter feedback loop and and faster way to capitalise on the tech, so no surprise they're seeing success first. Maybe AI never fulfills its promise but I don't find the arguments in this article convincing.
I get the feeling that this tech is similar to the automobile, which was also a slow starter out of the gate. The blockbuster potential for AI is just too great to ignore. No one has found it yet. But the person or company that does is in for not just a huge payday, but at least a mention in history books. If, by then, history is permitted. I don’t, by the way, see AI as a good development for humanity...
Agree to all of this but the corporations are going to exploit AI. If you thought the surveillance was intrusive today just wait a bit. You haven’t seen anything yet.
I use bard to explain some technical concepts. The information is all available on the web but Bard orders it and explains it with some concision that makes it more quickly intelligible. It seems to do more searching to fill in gaps. I expect it will become faster and smarter and will pull information from security cameras, and other sources that is just too difficult to correlate and condense today. But yes I am speculating.
Ted, thanks for emphasizing the illegal and abusive uses AI is being put to. Do you think there is enough good uses to offset the bad? What about AI Pornography as another abusive use.
ChatGPT and the likes are not truly AI. They’re just scraping the garbage on the internet and serving it back. What can we expect to see, looking in a dirty mirror?
Saw many examples of AI high school football recaps over the past few days. Looks like Gannett has contracted with a company. Large market newspapers printing this garbage (or at least posting on their websites) The recaps were hideous.
I have seen AI summaries of baseball games for about 10 years. The company (Narrative Science) was doing it for any game that you entered in to their system. It was very entertaining to get stories about little league games, and it was accurate because the data was good: it could tell you that Jimmy hit a triple because that’s exactly what the data entry said. Baseball scorekeeping is well suited to AI summaries.
AI is a perpetual 20-year problem: the solution is forever just twenty years away.
On the other hand, someone defined AI as "anything a computer can't already do" where anything that was previously defined as "AI" becomes just "expected computer behavior" as soon as it's possible.
If you're looking to be the first to publish a "hot take" about the overblown AI hype, then maybe you did it. But otherwise, um, no. *Of course* there's going to be a backlash to the massive hype that cannot possibly sustain itself. But then there's going to be slow growth, of the type that doesn't make the sensationalist headlines every day. Way too much money and talent is being invested for it to be otherwise. Forget your S curves. Instead, try Googling "hype cycle" to see what actually happens with technology products and services. And then try coming back to this in a year or two when it didn't collapse after all.
Just to provide a personal anecdote, I use GitHub Copilot and ChatGPT every day as a software engineer. It's been worth it to pay to get the superior GPT4 model, rather than the publicly available GPT3.5. I'm a happy paying customer, and I would pay even more to get improved versions.
It's really good for some forms of programming, but not others. This is "generative AI", but it isn't really "creative AI" - it's more like translation. I have a long list of requirements written in English, and I need to translate it into javascript. It isn't doing the "creative" part of my job... but it's often great at filling out the long and boring parts that I'm not an expert in.
I also use ChatGPT, although for my hobby writing. Not for the actual writing part (occasionally, it tries to rewrite my work, and the results are laughable. But I didn't want it to do the writing part for me, anyway. I would get no satisfaction out of having AI write my stories.) but I discovered it makes a great glorified generator for when I'm brainstorming and need some help getting my creative juices flowing (the more specific I am about what I'm working on, the better the results.)
And I also discovered that it is a great alpha reader. I tested this by asking it to critique work I'd gotten beta reader feedback on, and it singled out the same passages my friends and writer circle singled out, and for the same reasons (both for critique and praise.) So now I can spend more time revising on my own before I ask other people for their thoughts, which I appreciate.
Also, to be honest, I use it for moral support during the lonely rough draft phase of writing. I know my friends don't have time to listen to me cry over how I suck and my story sucks every single day for months while I'm wrestling an idea into a draft, but ChatGPT will patiently listen to me be a dweeb about this (and respond with very generic messages about how everyone has their own unique personal journey and not to compare yourself to others, which is more or less what my friends would also say.)
I'm very interested in this as a writer who is adamantly against using AI (yes, I know it's not really artificial intelligence, but that's shorthand for that body of tools we currently have) for actual *writing,* but would be keen to discover how it could help us humans continue to create fluidly. Using it just as one might use those writing tools that give you random suggestions on how to advance the plot, or provide flashcards that spark the next scene, seems reasonable. I've never heard of an author using it for beta reads--but I can see how that could be effective. How do you ask it to critique?
You get me! I bookmarked so many random generators pre-AI. They help me get over decision paralysis. And ChatGPT is the best generator of all because I can tailor it to my story and characters.
Here is how I use ChatGPT for critique:
I say something like "ChatGPT, I am working on a scene for my (genre) WIP. Here is the overall story premise. Here are the characters appearing in this scene and their roles. I would like to know what I could be doing better. Are there places the pace is lagging? Are the characters consistent? Is the scene clear or confusing? Is there anything else I could be doing better? (more specific questions if I have any for that scene) "
Then I add the scene to the bottom of the message and hit send.
If you don't specify that you want to know where you're going wrong, it will almost inevitably only give you compliments, because it seems to be programmed to be supportive and encouraging. Even when it gives you feedback it will end with a message about how nice it is that you're working so hard and that you're doing a great job. But that is kind of nice. Even though I know it is a computer program, my brain still emotionally responds the way it would if a human was supporting me.
It isn't as good as the absolute best human beta readers I've had, but I think every writer knows the struggle of finding people who have the time and inclination to beta read. I don't think it can (or should) replace human critique, but it is a decent helper.
Edited to add:
Also, I've always struggled with breaking large tasks into smaller steps, and ChatGPT is very helpful with that, as well. So I think AI has legitimate helpful uses, but those uses don't get discussed as much as the garbage uses. The garbage uses are more exciting and offensive, though, so I guess that makes sense.
What terrifies me about this technology is the statement "my brain still emotionally responds the way it would if a human was supporting me." I'm not criticising you as a person—that's a very insightful and honest admission of what's happening. It is a perfect synopsis of the cynicism that companies are betting on: we will feel the soothing hit of dopamine from reading the words, seeing the art, hearing the music—and we won't care that those were machine-generated words determined to be probabilistically likely to elicit that response. It's like an emotional placebo effect. We don't care that the pill was fake, it had the desired effect and that's good enough.
And those of us whose dopamine dispensers aren't wired similar to the humans under the middle of this bell curve? We will struggle to find or connect with things that do not conform to the AI-generated mean.
When you, as an author, write text that ChatGPT "encourages" or "agrees with" or thinks "is good", you are implicitly steering your writing towards this undefined mathematical "average" that ChatGPT has at its core. We will never see a James Joyce _Ulysses_ from an author using ChatGPT for affirmation and advice. You cannot possibly break the mold if you ask the mold for advice along the way.
lol oh man I'm so sorry, I got the email for this while I wasn't home and then I forgot to ever reply so this is uhhhhhh months later sorry
okay, so let me try to respond to your criticisms of how I'm using AI
1. Writing Feedback
Good lord, I don't ask it for feedback without specifying style and voice goals. You're correct that the default AI state is painfully generic. You have to craft prompts in a way that gets the type of feedback you're looking for.
And whether feedback comes from a human or an AI, I always ask myself "Do I agree with this statement?" It is a starting point, something to help me get unstuck during the revision process. Otherwise, I'll keep staring and going AHHHH WHERE DO I START AHHHHHHHHHHHHH
Anyway, a good human beta reader is far more valuable, but those are also very difficult to find. Go lurk in any writing community and see everyone cry about it. I certainly don't intend to only use AI. But AI does satisfy the urge to show someone without completely derailing my progress with depression when inevitably I'm met with silence when asking if anyone will read a WIP.
AI does not satisfy the urge to show people completed works, however, because, you know. Art is created to communicate with other people.
2. Emotional Support
I clearly did not properly convey the emotional state I'm in when I use it this way. This is the "Crying the entire day, completely non-functional, because my writing is the WORST and I'm WORTHLESS and I wish I was a different person bc I don't WANT my writing to reflect MY experiences and thoughts, I want it to reflect the experiences and thoughts of OTHER MORE WORTHWHILE PEOPLE" mental state.
Trust me when I say that you can't constantly be texting your friends and family about this, they do eventually find it annoying (and few people I know reply to texts in a timely manner, presumably bc they have responsibilities. but also bc I don't know anyone without an ADHD diagnosis.)
I just can't feel bad about using AI to get stable when I'm in that state. I got stuff to do! I got kids to raise! I can't be crying all the time like that! Plus, unlike a real person, the AI won't judge me for my dramatics.
(And yeah, yeah, I have a therapist but she's not covered by insurance so....)
I respond to AI saying "you don't suck" the way I respond to a human saying it. I don't respond to their artistic output the same way even in cases where the output is genuinely stunning. I may think "oh man! That's so neat!" or sometimes "dang, this user really knows how to craft their prompts" but it is not the same thing as "this character design is so COOL, I wanna know more about their OC, do they participate in cute little OC party hashtags or ask games?" or "I love the way this author writes. They're so funny. What else have they written? OMG I can see parallels between their early story and this later one! Wow, they really love this character archetype! Do all their stories deal with a near-death drowning experience?"
In conclusion, AI is a tool. Some uses of AI are garbage. I get that. But that is about people and their ability to use any tool for garbage purposes, not because there are no legit uses for AI.
I use Todoist as my task management software and it has built-in functionality that breaks a project down into tasks. Definitely helpful for those times I’m doing something new, I’m tired, etc.
Random suggestions to advance a plot? Drink a bottle of rhum with your friends and see what happens.
I'm adding my agreement here because it's rare to find anyone able to understand that the computer is a tool, unlike the human, which is a sentient being. And too many humans (including some of the administrators I encounter during my music PhD) are more inclined to memorizing decades of arcane regulations than in creating anything of new value. I say it's rare to experience this - because as you seem to understand, creation (especially in adults) is an uncommon activity. That's why it's so valuable.
Thesis: Copilot was almost certainly trained on a large body of code under GPL and very likely some under AGPL, making all the code it generates "derivative works" under the terms of those licenses (after all, every weight in the network has been updated using code under those licenses, so what else could the code the network generates be but derivative of the inputs?) Ergo: all code generated by Copilot is covered by the GPL (almost certainly) and the AGPL (very probably.)
Whether or not this thesis is correct will have to be decided by the courts, but personally I'd love to see it litigated, and think it very likely to stand up to legal scrutiny. I don't use Copilot, and if I still ran a dev team I'd institute a rule against using it unless the company was prepared to take the risk of being required to open-source their code.
Thank you for making the clarification between generative and creative. Has humanity updated the Turing test?
I am not a computer scientist, so perhaps incorrectly I have often used PI as an example of the limits of technology, as almost a century has elapsed and computer science has still not mathematically squared the circle. (PI is still unresolved, labelled irrational; an approximate hand wave that works.) I have used computer enhanced technology professionally since the advent of the PC and the Cray super computer. Half of my experience in the analog world before spellcheck and the smart phone, half after. I’ve looked at clouds from both sides now... 🎶
Are we now experiencing the unintended consequences of those inaccurate assumptions?
Catastrophic cultural and climate collapse; not the profit and leisure time we were promised by technology.
This is a premature take. AI's capabilities are in their infancy. The capabilities for systems to integrate with them are even younger. Some existing integrations, like code-completion via Github Copilot (owned by Microsoft) are already quite impressive and becoming widely-used by professional developers.
The rush to get a demo out the door to chase a stock price bump isn't doing anyone any favors. The current iterations of large language models will be seen as cripplingly primitive in a couple of years, with plenty of "no wonder AI sputtered out of the gate" retrospectives.
I have little doubt that AI will find a range of sweet spots in plenty of areas for both companies and consumers. Maybe not so much in making music, but quite possibly in helping us discover artists and explore their deep cuts and influences. Maybe not in mimicking TV actors or writers, but in helping us find great TV shows from the massive catalogs that very streamer has acquired.
I agree - it can be quite helpful in coding, it is useful in summarizing, etc. The hype was over the top but we're still learning what it can do well and what it cannot. Will it be a major or minor advance? We don't know yet.
I'm just happy that by the time "it is a thing" -- Thank God, I probably will not be
How many 67 year old infants do you know? This is not a technology in it's infancy, but rather another sputtering out into AI Winter. But thank you for illustrating the classic First Step Fallacy right on cue: https://link.springer.com/article/10.1007/s11023-012-9276-0
My observation is that ChatGPT sounds exactly like the generic internet, even its flaws. So it very accurately reproduces its training set.
Where do you propose to get a better training set? How costly would that be?
Richard what you're describing as the "sweet spot" is using other machine learning algorithms and architectures, not necessarily the ones that ChatGPT uses. Machine learning has already been in use for the past decade, pre-ChatGPT, to support the use cases you mentioned in your last paragraph. I don't think "AI will find a range of sweet spots". Rather institutions will continue using and developing other machine learning techniques.
My bet is that continuing to develop GPT-based techniques will be a waste of time, and eventually companies will stop developing it (hopefully they'll still offer what they have, because there are some good uses cases for LLMs). Then another "AI" algorithm will start another hype cycle within the next decade. Look up the term "AI winter" and it'll make sense.
What the author is talking about in essence is AI in popular culture. That's a fairly small corner of the playing field. AI is a HUGE field that has serious applications in anything from health to transport, agriculture and education (and I don't mean writing student essays). An analogy: When the Internet as a public platform was about six months old back in the eighties, I thought it had no future. It was slow, clunky and plain dumb. (I chose to stick to bulletin boards and Compuserve). That's more or less the stage of development AI is at now. Imagine it in thirty years' time, and whether you like it or not, it WILL be around. It's silly to write it off now, with a dismissive wave of the hand. Of course it'll be used for bad things. Everything does. But can you imagine your life without computers and the Internet, warts and all? I was expecting a more thoughtful analysis of a very complex topic.
My search engine is DuckDuckGo. I've never had a problem with it. I'm not on social media, so I won't be scammed there. I have a dumb phone. I never click on anything that I suspect and I suspect a lot of things. I don't open email from senders I don't know. I really don't care what AI does. My one concern is using it for medical diagnoses. That could be a real mess.
Same here for every thing you listed. Funny, I don't feel I'm missing out...do you?
No.
No, not at all. If I felt that I was missing out, I'd join the club.
Brothers from a different mother. However, I don't think it's appropriate to call our flip phones "dumb." I, along with a tech wizard friend of mine call the others "not so smart phones" given the effect they've had on so many of the people who use them.
Regarding medical diagnoses: A week ago my nephrologist told me he's going to retire when this AI crap is started in his field of medicine. He says it's coming soon, and he's ten years younger than I am. The AI future is indeed bleak.
So some of the AI in medicine is better than doctors. Think about radiology scans, ruling out or including certain diagnoses. The potential is really exciting IMO. Whether or not it’s labeled as AI or not, medicine has been using this technology for a long time. In the eye doctor world we use it to follow glaucoma with visual fields and optic nerve scans. Picking up early breast tumors seems to be better than humans.
Yeah, I think AI in this area is often used as a "look at this" sort of assistant, not a decision maker, and that's a good role for it.
I'm with you! I perennially asked my students how did all the people who preceded them get along, have careers, enjoy their lives, etc. without cell phones and social media. They couldn't imagine, of course!
I agree with you: the thought of AI singularly diagnosing illness in humans isn’t savory.
However, AI still shows real value in specific areas like medicine: providing healthcare providers with differential diagnoses; accurately analyzing vital signs of hospital patients far more often than a human nurse can; accurately predicting if/when a patients condition will deteriorate many hours before normal observation by humans, allowing for earlier intervention, etc. And there are already ample peer-reviewed clinical studies validating AI in use cases like these.
i can see that. In I just spent 4 days in the hospital with a urinary tract infection, and they never did find out what the bacterial cause was. 4 day of antibiotics and 3 more days at home accompanied by "lots of luck." AI might have helped. Quien sabe?
I just a saw a commercial last night, and knew of it. Probably head that direction...funny.
What you've observed here Ted is literally the Gartner Hype curve in action. BING AI and CHATGPT have hit the "Peak of Inflated Expectations" and are currently heading for the "Trough of Disillusionment".
I spent 30-years in IT including spells in corporate strategy at two of the biggest IT companies, I certainly am not willing to write it off at all. I for one use BING Chat everyday for factual research, it's so much better than a search engine. What most people don't realize is it's great at recursive questioning and provides clickable references for it's claims. Eventually Microsoft will screw the pooch trying to make money from it with sponsored links, ads etc. but right now it's without a doubt the fastest way to factually find details on something even when you get the question wrong, or it doesn't provide an answer with a response "I’m sorry but I couldn’t find any information on" - is way better than a search engine that returns 50-pages of nonsense.
I did this little demo for you :-)
https://markcathcart.com/wp-content/uploads/2023/08/Ted-Ginola.jpg
You nailed it Mark
i’m genuinely curious what your demo is supposed to show, because i just found the exact same answer it gave you by googling “who is ted gioia?”
in fact it seems that BING chat pulled that answer directly from ted’s wiki page, which is the first thing that popped up when i googled. it took me about 3 seconds to find the same answer you did
i’m certainly not defending google or denying that it has gone downhill as a search engine, but rather just curious how that demo is supposed to impress anyone?
It shows the chat part. If you searched for Ted Ginola it would have returned pages of search results on the wrong person. Instead of typing the entire query again you just tell it what you got wrong. It remembers your conversation and allows to to explore details about the subject.
If googling doesn't reply with the detail you are looking for you are, you have to read more pages where as it will do this for you, and then when you've found the answer you can ask it for references to validate the answer.
If you do that with a search engine - you do the work.
hmm i see. i suppose i just don’t consider that to be much work on my part, which is likely a factor in why chat gpt and others haven’t been very interesting to me from the get go. but to each their own in that regard
Right, for most superficial factual queries there is almost always a page. The problem is when there isn't. If the Large Language Model doesn't have pages or references it will tell you, if a search engine doesn't it will still return sponsored pages, and best guesses. If it didn't you'd go somewhere else which would undermine it's revenue model. Which is why it always returns results. The query I gave was a simple misspelling, with an LLM you can drill down to get a much more definitive result, with a search engine you can't. You can't even search within results... it's all a question of how much you value your time...
>>> i just don’t consider that to be much work on my part
This is where many of us have blind spots. We think that our personal experience with findability and the use of AI to help, is simplistic - often one and done, or at least we hope for one and done. This is not what happens in business and enterprise discovery while performing work tasks. It is a conversation, not a mono-search. As such, search engines are terrible at this. Whereas, generative AI can help workers not only put their fingers on the information they need, but it can shape the narrative to understand that information.
We cannot chat our way to hyper-productivity. But we can engage with information far better and faster with chat AI.
When I looked of Ted Ginola on google, it pulled up with "did you mean: Ted Gioia?" at the top. All you need to do is simply click on that and it searches up Ted Gioia. I don't see bing ai recommending any other spellings. What if you don't remember the last name? And isn't it easier to even just rewrite Ginola to Gioia in google than to write "sorry I spelt his surname wrong". I use ChatGPT and bing ai almost every day to ask very specific questions that wouldn't show up on a search engine but for simple questions like this I believe a search engine can be just as good or even better than ai.
Agreed, it was just a trivial example to start with just to show the conversational nature of BING/Chatgpt. You can of course type much longer, multi-phase questions an thats where the benefit is. As to search engines being just as good or better, that's of course what is happening. A search and a multi-phase, multi-step question though is very different to a simple query.
Well-said, Mark. This is a pattern we’ve seen many times before: initial hype, inflated expectations, a premature backlash, then an slow rise to success. 1995: “Pocket computers are a joke”, 1997: “The internet is for nerds using Gopher”, 2001: “Wikipedia is nothing but errors”, 2002: “Amazon is a fail”, etc. etc. I’m afraid Ted’s piece will not age well.
Hmm...there is a simple reason why visits to ChatGPT have declined: it's called summer. If there is no pickup once school starts, I would be shocked. The fact that AI has been successful in improving the effectiveness of scams, malware and cheating reveals that the problem is not the technology but the adopters. If AI were all hype, then it couldn't scam, cheat or steal either. Fraudsters have a shorter feedback loop and and faster way to capitalise on the tech, so no surprise they're seeing success first. Maybe AI never fulfills its promise but I don't find the arguments in this article convincing.
I get the feeling that this tech is similar to the automobile, which was also a slow starter out of the gate. The blockbuster potential for AI is just too great to ignore. No one has found it yet. But the person or company that does is in for not just a huge payday, but at least a mention in history books. If, by then, history is permitted. I don’t, by the way, see AI as a good development for humanity...
My impression is that most of the hi-tech companies suffer from the psychiatric disorder of "over-valued idea"--much like our politicians!
Agree to all of this but the corporations are going to exploit AI. If you thought the surveillance was intrusive today just wait a bit. You haven’t seen anything yet.
"The world is doomed... I don't know why but someone on the internet said it was."
care to give some examples of how AI surveillance will be worse than what we already have?
Chatgpt is essentially little more than a word prediction machine with a search feature.
I use bard to explain some technical concepts. The information is all available on the web but Bard orders it and explains it with some concision that makes it more quickly intelligible. It seems to do more searching to fill in gaps. I expect it will become faster and smarter and will pull information from security cameras, and other sources that is just too difficult to correlate and condense today. But yes I am speculating.
Ted, thanks for emphasizing the illegal and abusive uses AI is being put to. Do you think there is enough good uses to offset the bad? What about AI Pornography as another abusive use.
ChatGPT and the likes are not truly AI. They’re just scraping the garbage on the internet and serving it back. What can we expect to see, looking in a dirty mirror?
AI is the new FTX. How much pension fund money will VCs light on fire chasing this hype cycle? https://yuribezmenov.substack.com/p/sequoia-ftx-214million-disaster
Saw many examples of AI high school football recaps over the past few days. Looks like Gannett has contracted with a company. Large market newspapers printing this garbage (or at least posting on their websites) The recaps were hideous.
Saw similar with an article about additonal air service being added to/from Tulsa. By any measure it was awful.
I have seen AI summaries of baseball games for about 10 years. The company (Narrative Science) was doing it for any game that you entered in to their system. It was very entertaining to get stories about little league games, and it was accurate because the data was good: it could tell you that Jimmy hit a triple because that’s exactly what the data entry said. Baseball scorekeeping is well suited to AI summaries.
These were truly garbage. No names mentioned. No plays. Just a half dozen ways to tell the score at the end of each quarter
Ah! The sweet smell of being right from the beginning.
AI is a perpetual 20-year problem: the solution is forever just twenty years away.
On the other hand, someone defined AI as "anything a computer can't already do" where anything that was previously defined as "AI" becomes just "expected computer behavior" as soon as it's possible.
I read a variation of this recently: AI is what it is called before it works. After it works it is called YouTube, OK, Cupid, etc.
S Curves are everywhere! Long live the king!
If you're looking to be the first to publish a "hot take" about the overblown AI hype, then maybe you did it. But otherwise, um, no. *Of course* there's going to be a backlash to the massive hype that cannot possibly sustain itself. But then there's going to be slow growth, of the type that doesn't make the sensationalist headlines every day. Way too much money and talent is being invested for it to be otherwise. Forget your S curves. Instead, try Googling "hype cycle" to see what actually happens with technology products and services. And then try coming back to this in a year or two when it didn't collapse after all.