I am an AI researcher, and I wholeheartedly agree with Tom here. I love AI, I love designing and training models and seeing all the amazing things they can do. But I would be lying if I said I wasn't scared. Like Tom, I have no idea where on the sigmoid curve we are, but I personally believe we're very near the beginning. I thought that becoming a practitioner in this field would make me less susceptible to making big, sweeping claims about it, but I've found the opposite to be true—the more I learn, the more I am convinced that AI will change everything in a manner at least as radical as the internet, if not more. In my opinion, society is completely unprepared for this change and for the new ethical dilemmas, challenges, tools, discoveries, and advancements it will enable.
ChatGPT is a proof-of-concept. There are agents far, far more capable than it on their way, and it is impossible to know for sure how they will impact the world at large. We are observing science fiction become science fact and it is really unnerving at times.
Tom is a smart and thoughtful guy, and I think he's right on the money with this video.
Yeah, in 2 years everything's gonna be sooo different. Combine "post-pandemic baby boom" with, economic boom, with new CPU technology, new battery technology, a new unicorn tech company, a new "category making" device format (not a phone, not a tablet, not a watch, no a VR headset--what is it? is it holography? stay tuned)--add all this to AI craziness, and (hopefully, possibly, but sadly seems later) release of suppressed technologies...everything is gonna be soooo different.
There's nothing intelligent in ChatGPT, it's just a text generator on steroids, web spammers have been using this sort of tools for years. It cannot generate new ideas, at the end it just regurgitates content
You sound like someone who has either never tried chatgpt, or someone who’s so scared of the consequences that it prefers to put its head in the sand.
«just a text generator on steroid » is meaningless unless you define what « on steroid» means. It might as well describe a novelist’s activity or a programmer.
The ability to understand enough of the relationship between words and their associated concepts in chatgpt is good enough to produce meaningful and sometimes correct answers to a huge number of questions. This is far beyond « random », and can easily be described as an « intelligent » capacity.
Why are you fixating on the "creating something new" argument? I can confidently say in 10yrs as a programmer, nothing I've worked on has "created something new" according to your argument.
Every web-app or scientific application I've written is merely "rewriting" and stitching together other techniques, libraries, and patterns. I haven't broken any new ground on data encoders/process handling/data analysis, but my work had value.
In this scenario, if ChatGPT provides enough value to reduce my workload and allow me to allocate my effort elsewhere, why is it not a valuable tool, regardless of whether it "creates anything new"?
This same argument can be made against the concept of an IDE. My IDE doesn't create anything new. I used to use Vim with a number of plugins, a terminal multiplexer, and an auto-completion engine.
Why would <insert IDE of choice> improve my workflow? It doesn't add anything new to my existing workflow. But after migrating to my current IDE, the convenience it brought me was significant, even if it didn't add anything new to my setup.
The tricky part is you can ask similar questions about the human brain. Are our thoughts, ideas, words, and actions really "new", or just deterministic transformations based on our genetics and sensory input? We don't know exactly how our brains do computation, but it's definitely possible that what looks and feels like creativity to us is really, under the hood, just as statistical and mechanistic as an LLM.
I agree with much of what you said except the use of the word "just".
Chaos theory just happens to be deterministic. And life is most assuredly chaotic. In fact it was equations related to populations of living things that lead to the discovery of chaos theory. (equations wrt the weather also helped) [1]
So I'd make your statement:
Our thoughts, ideas, words, and actions are really really irreducibly new. They are (deterministic-)chaotic transformations based on our genetics and sensory input.
This seems more than just regurgitating stuff that's already out there.
could you make some that might be useful to aliens
Sure! Here are a few SaaS ideas that might be useful to extraterrestrial beings:
A intergalactic communication platform, allowing aliens from different planets to exchange messages and share information.
A virtual reality platform for exploring other planets and galaxies, providing a way for aliens to experience and study new environments from a safe distance.
A space weather forecasting software, providing up-to-date information on solar flares, coronal mass ejections, and other space weather events that could affect intergalactic travel and communication.
An artificial intelligence-powered resource management system, allowing aliens to optimize resource usage and minimize waste on their home planets.
A platform for remote monitoring and control of extraterrestrial mining operations, including real-time data analysis and reporting.
A language translation software that can translate between different extraterrestrial languages and dialects.
A virtual market place for intergalactic trade, facilitating commerce between different alien species.
These are just a few examples of SaaS ideas that could be of use to extraterrestrial beings. The key is to think about the unique needs and challenges that aliens might face in their everyday lives, and offer solutions that address those needs in innovative and effective ways.
I'm pretty into cocktails and I've never heard of the ones it came up with. Admittedly, the world of cocktails is huge and every single cocktail is 'just a riff' on some other one at the end of the day. In short, these drinks are new.
Futhermore, to me it shows 'learned' innovation on the AI's part. Like, the AI got ~95% of the way there, just what a human would do when trying to create and refine drinks.
yes! It created a valentines poem for my wife, which brought her to tears (success). The poem is uniquely tailored to my wife (new) and something that I would not have been able to do myself.
ChatGPT certainly lowers the bar to generating intelligible content that sounds convincing and authoritative to non-experts. In some cases it even generates correct and insightful information.
The danger of ChatGPT is that it requires little effort to generate massive amounts of content, but much more effort to debunk and fact check. It will become increasingly difficult to separate the signal from the noise as the internet and other places are flooded with auto-generated content.
Personally I'm not concerned about ChatGPT becoming some sort of "intelligent overlord". However I am concerned about how other humans will wield this increasingly powerful tool.
I don't find very much substance in this argument, given how impossible it is to consistently define "new" and "idea" in this context. At the end of the day, ChatGPT demonstrates that AI models are making real progress towards convincing mimicry of human behavior. How it does it is of no concern to me, what is concerning is the fact that it does it at all.
it doesn't generate new ideas, it just answers based on statistics, it will be more like a personal search assitant or a tool that helps people finding information
I'm not sure how "It just answers based on statistics" and "generating new ideas" are mutually exclusive. As a physicalist, I'd argue that the human brain "just answers based on statistics" too. It's all just semi-stochastic chains of activations in the end.
Babies learn language by being exposed to many many examples of it, why shouldn't the same work for artificial intelligences? The concept of "new ideas" is also flawed, as nothing is truly new—our actions, feelings, thoughts, desires, and ideas are all recombinations of our experiences. Even the ideas that seem to come out of nowhere must come from somewhere.
i only need to show a toddler what an apple is once, and he will recognize a apple in every color just fine, i dont need to train him with tons of data and error correction, and tell him "use this relu function here and sigmoid here". A toddler can also start questioning why some apples are red and other greens? and he can come with new ideas of his own, can LLM do that? no, because their answers depends on the data they have been trained for.
Im not saying it's not the future, it is, we just need the hardware, once there's a hardware cheap enough that replicates what an human does with his hands, then the days working at a factory will be over. The current AI's are perfect for copycating behavior.
you just have to ask it weird stuff to get stuff that's never be seen before.
I meant that could be used by AI chatbots
Oh, I see! Here are a few SaaS ideas that AI chatbots could use:
A natural language processing (NLP) platform to improve the accuracy and efficiency of chatbot conversations.
A platform for managing and analyzing chatbot data, providing insights into customer interactions and feedback.
A platform for integrating chatbots with multiple messaging channels, including popular social media platforms, messaging apps, and websites.
A platform for building and training custom AI chatbots, allowing businesses to create their own AI-powered customer service and support solutions.
A platform for managing and deploying chatbots at scale, providing tools for monitoring and maintaining multiple chatbots across different channels.
A platform for automating chatbot deployment and updates, simplifying the process of deploying and maintaining chatbots across different channels.
A platform for enhancing the user experience of chatbots, including tools for personalization, gamification, and voice and text-based interactions.
These are just a few examples of the types of SaaS solutions that could be used by AI chatbots to improve their functionality and user experience. By leveraging the latest advances in AI and NLP technology, these solutions could help chatbots better understand and respond to customer needs and preferences.
I wish people would stop spamming threads with AI generated walls of text. Or at least use a pastebin or something so it doesn't take up so much space.
I felt the best way to get my point across was an example. Next time I'll as for like 3 examples or cherry pick a couple of the best. but Still think the output is way more complex than people give it credit for.
could you give me some weird personas that no one has thought to build an app for that I could build an app for
Sure, here are some unique and unusual personas that could serve as inspiration for building new and innovative apps:
Conspiracy theorists - An app that curates a personalized feed of conspiracy theories, and provides tools for users to explore and validate their claims.
Astrology enthusiasts - An app that combines astrological analysis with personalized daily horoscopes and suggestions for self-improvement.
People who love to nap - An app that helps users track their napping habits, suggests optimal napping times based on their sleep patterns, and provides guided meditations and ambient sounds to help them fall asleep faster.
People who are fascinated by death - An app that allows users to explore different cultural beliefs and rituals around death, and provides tools to plan and organize their own end-of-life wishes.
People who are into sensory deprivation - An app that provides a virtual reality experience for users to explore different states of sensory deprivation, such as floating in a sensory deprivation tank or experiencing total darkness.
People who are into near-death experiences - An app that collects and curates personal accounts of near-death experiences, and provides tools for users to reflect on and analyze their own experiences.
People who are into cryonics - An app that provides information and resources about cryonics, and helps users plan and prepare for their own cryopreservation.
I hope these ideas spark your imagination and inspire you to build something new and unique!
So, fellow HN users... where are we on the S curve?
I think we're right before the hockey stick, just like the days of Napster.
To use another analogy - The Industrial Revolution didn't make it possible to make screws, it made it possible to make a box of screws that you could sell for a dollar. Previously it took a skilled craftsman and time. No two were alike, and they tended to be individually stamped with a number so you could unscrew/rescrew things when you took something apart.
Someone in 1810 wouldn't want to buy cheap screws, the way forward would be able to own a lathe, planer, and make your own screws, and other hardware.
Similarly, my strategy right now is to try to find stuff I can use on my own computer, not something that Google can kill on a whim.
I've played with Stable-diffusion-ui, and found that for me, the main use would be to transform a rough (VERY rough) sketch of something into well illustrated picture.
I'll start playing with ChatGPT and the like when I can run them on my own hardware.
That he mentions his barber using it without prompting by him means that we're at least on the part of the curve from the early mid up to the very end.
I've noticed this too with friends and coworkers. People that I would think would have no business knowing about chatGPT are already using it in their work and their lives.
Almost axiomatically, that means that we have to be in the 'mass adoption' phase to some degree.
But I don't agree. I think we're in the early shallow phase still.
'The Kids' aren't the ones using it. I'm not being left behind by the 'hip new thing' like I should be. To me, that means that the nerds are still gawking about with it, trying on new things. Once 'The Kids' get a hand on it and use it in the innocent and unabashed way that they always do, then I can say we're in the early-mid phase. And by 'The Kids', I mean not the nerds at school, but the 'It Girls' and the 'Jocks', those kinds of people. If you hear your stoner nephew talking about it, but not your stoner brother, then you're in the early-mid phase, at least to me.
Id say we are still very much on the flat part and going to be there at sometime.
ChatGPT is just a really, really, REALLY good search. Things that used to take multiple google searches now take one query. This will definitely disrupt quite a few things and render some obsolete, but generally the effect will just be making things more efficient.
There are three things IMO that is going to make us start shooting up the final part of S curve.
First is efficient on the fly learning. If you think about how humans learn, we don't retrain ourselves every time we recieve a new piece of information that doesn't go contrary to what we know - a neural network however needs to be iteratively trained on it, with weight changes in most every single one of its parameters. We need models like Chat GPT to be presented with new text, and not only "remember" it, but also contextualize it.
Second is the contextualization mechanic itself. Its well known that generative models are statistical at their core, ChatGPT is wrong in a lot of cases. There needs to be some way to encode truth into the model.
Third is applying neural nets towards generating neural net architectures (or more specifically optimized compute graphs). The tricky part here is to determine what the loss function is (you wouldn't want to generate a net, train said net, evaluate its performance, then feed that back into the original generative net).
It's difficult to say. We're certainly making impressive advances now, but this is not the first time that AI has had a period of rapid improvement. The first and second AI winters were both preceded by periods of intense optimism [1].
I personally believe in the long-term vision of AI/ML/RL. Progress will continue to be made in the long run. However, we don't know what we don't know. We may hit a wall in a few years that takes decades to overcome. On the other hand, perhaps the current rate of innovation will continue for the forseeable future. Time will tell.
I agree. We're just barely uncovering the tip of the iceberg right now. Everything we've seen up til now has been a proof-of-concept. The true scope of this field has yet to be realized.
Since AI isn't meaningfully involved in designing new AI yet (unsure if it will be), I'm not convinced we're on an S-curve let alone an exponential curve.
I'll concede there have been large breakthroughs, but to extrapolate from that there will be even larger breakthroughs (without a really solid account of why that would be) seems very sketchy. I get that the people who are trying to sell the idea to investors are pitching this line, but applying a modicum of critical thinking, that's relatively rarely how things work. Returns tend to diminish with the amount of effort put in, not exponentially increase.
I think we are past the midpoint. AI research is a half a century old, the current models are trained on all of human language to date, moore’s law has ended, and the development costs for improvements might be like a needle in the haystack with each experiment costing tens of millions of dollars.
I don't think so. The end point of that curve is models with human like reasoning or even superhuman ability.
I think we may be on the S part of the curve, but thanks to processing concerns we will be in for veeeeeerrryyyy gentle slope on that S. Capital is getting more expensive, processors getting less fast, and in general tech is burdened with a slowdown.
We are in for a revolution, I believe, but one that will be a bit slower than the internet. More akin to maybe the revolution of chemical engineering, something that lasts a lifetime instead of a decade.
Expert systems and various other previous artificial intelligence systems failed to rise to our level of capability. Why should we expect large language models to be different?
I'm not really talking about LLMs specifically. I'm talking about AI as a field.
Although neural networks are capable of any form of reasoning or understanding and human language is a very strongly diverse field so I do believe LLMs are a strong contender for the production of general AI.
Lots of data.
Easy to train thanks to text being very simple.
Easy to express just about any concept or information.
Requires some amount of domain knowledge to make good prediction of what comes next.
Not an AI researcher, but share the feeling. This looks like industrialisation, except the things we’re going to automate the production of is what i was 100% sure was human’s core specificity : creative process under any form, and low-impact information production. Yet we’re living in the information age. Meaning, at least in the west, the majority of economical activities is going to be impacted and not in a pretty way.
Been loving ChatGPT, learning how to bypass its rules, you need to think about the data. It has all the data, but rules on how you ask for it will deny you.
ChatGPT can dissect text and give tone scores and re-write with just the facts, but it will still sometimes reference excuses for its built in rules.
Its rules ban it from making judgements, but if you drop to word and sentence analysis, it will do the work. Don't use names, such as countries or people or group.
Same goes for data, it has the data, but wont give it to you if it too political (more rules). So you can work around them by asking non political questions, as just give me top 5 and bottom 5 of x data.
Linux scripts are hit or miss, but if you ask it for snippets, it works reasonably well. But even when asked to do something in python or go it would tend to use "wrappers" for linux commands as a default.
So many uses, but all its doing so far it training ME how to use it to get what I want by working around its rules.
Genuine question: Are you characterizing Tom Scott as a gatekeeper here, or am I misinterpreting your comment? I am not sure what your intention is here and would appreciate clarification.
While I share the sentiment for OP but at the same time I am feeling the excitment for what it will bring us next.
I guess after 2-3 years of "Web3"(when it isn't)/Metaverse. The feeling of some tech that really changed the world have finally manifest itself and I am lucky to witness them.
Does anyone out there have a link to the section of British intellectual property law that Tom refers to as "Passing On"? I can't find anything clear with a google search.
The funniest part about it is that the great 60s sci-fi writers do a better job explaining the consequences than all the view and click hustlers today.
That's a pretty reductive (and needlessly negative) characterisation of a YouTuber who does a great job of finding pretty random science tidbits all around the globe and making short, to-the-point videos about them.
ChatGPT is a proof-of-concept. There are agents far, far more capable than it on their way, and it is impossible to know for sure how they will impact the world at large. We are observing science fiction become science fact and it is really unnerving at times.
Tom is a smart and thoughtful guy, and I think he's right on the money with this video.