I am puzzled by all the hot takes stating that ChatGPT (and other similar tools) outputs garbage and is being "discredited" when I have personally found it to be an incredibly useful tool that I have been using daily pretty much since it was first released. The only thing being discredited, in my mind, are the people who rush to be contrarian about something they don't actually seem to understand or have taken any time to use.
There seem to be two kinds of people: ones who see the mistakes ChatGPT makes and the ones who don’t.
If I was a scam artist today I would be very happy to find people of the second kind to victimize.
Secondly I’d be very concerned that the second kind of person is working at the hospital dispensing drugs or on the beat as a cop or doing any kind of job where mistakes could have consequences. They're not just going to be making mistakes and letting other people's mistakes go by, they are going to be denying it and making excuses about it.
If ChatGPT has a core competence it is that people give it the deference that they give to high status people, who are used to just saying whatever comes into their head and having people just nod. I think of the story of “the Emperor’s new clothes” where the Emperor pulls off something that not everybody could do.
These concerns converge in the 419 scams which deliberately filter for the second kind of person with deliberately poor grammar and spelling. Amazingly there are people who have $10 million dollars to lose who fall for this! I wonder if there is somebody evil, bright, hard-working and ambitious who is HFRL training a model right now to do romance scams.
I'm acutely aware of ChatGPT's mistakes, have spent lots of time investigating them, and still find immense value in the tool.
While it's true that many less technical people will end up treating it as an oracle, let's not conclude that it's impossible to understand what it is and use it productively despite its flaws.
I’ve frequently been the ‘final assembly programmer’ for projects that were started by some fresher that got to what looked like 90% done by management but it was unsound underneath and that last 10% took 50% or more of the work.
There is nothing I find more fatiguing than pushing a bubble around underneath a rug in a system that eventually has to be carefully taken and inspected bit by bot and a huge amount of work put into figuring out what exactly is wrong with it and fixing it.
I’m afraid the ChatGPT enthusiast is going to walk away from a looming disaster thinking they and ChaptGPT is so brilliant, probably never realize the harm they did, and if they do they’ll be contemptuous of the low-status ‘grinds’ who take so long to make things right.
> There is nothing I find more fatiguing than pushing a bubble around underneath a rug in a system that eventually has to be carefully taken and inspected bit by bot
Couldn't agree more
> I’m afraid the ChatGPT enthusiast is going to walk away from a looming disaster thinking they and ChaptGPT is so brilliant
Also agree.
But none of this detracts from ChatGPT's value or incredible power.
It just means that it will also cause many problems. Maybe once you net everything out the balance sheet is negative. I think you could make that argument for social media, for example. But the point is moot, because it's here, and its influence is only beginning.
> There seem to be two kinds of people: ones who see the mistakes ChatGPT makes and the ones who don’t
Agreed; it quickly became apparent once I found myself entering the 'seeing its rough edges' camp.
Once you've _tried_ breaking it and can see the limits of an LLM you'll find yourself wishing for an AI assistant, but knowing what is and isn't possible makes managing one's expectations with new technologies easier.
I'm still generally an optimist with these new techs because they're still very cool and potentially useful interfaces to existing technologies, but I agree with you in that it's too easy to get caught up deferring to it like it's some techie oracle (and that the tendency for people to want to do so is concerning).
I used to think people wouldn't "simply accept" the types of systems in _Minority Report_ or _Psycho-pass_, to the extent that I found it to be immersion-breaking. But, concerningly, it seems folks are happy and willing to give up that deference (in what's probably a well-studied sociological observation I'm simply unaware of the term for). Scary stuff.
I'm one of the ones who sees ChatGPT's mistakes. It makes lots of them. Some are very dumb.
But the idea that it's a fraud or a scam or whatever label the naysayers choose to put on it is just wrong. It makes mistakes, but it also provides a lot of value when used correctly.
Consider the many use cases where you don't need to rely on its "knowledge" at all - just this morning, I used it to write some marketing emails. I've got a paragraph about the company that I give it, plus some specifics about the particular email, and it knocks out something that's entirely useable in seconds.
There are plenty of tasks like this, in which it's creating something based on your input, in which it is immediately and clearly useful. Just because there are use cases where it fails doesn't make it a scam.
I would wager that the recipients of marketing emails do not consider them valuable. The prospect that in the future, these will not even have human creativity flowing into them just adds insult to injury.
I've got a few thousand people who have signed up for them and haven't yet unsubscribed. Whenever I send out a marketing email to that list, a number of people immediately make purchases.
Why would that be the case if people didn't consider them valuable?
When my "genius" boss first heard about ChatGPT etc., he was convinced that this would end up replacing all the developers in our org. I told him that the more likely COA was that it would replace all the middle managers who just collate TPS reports to push to the VPs.
I think of the story from Lem’s Cyberiad about the 8 story high ‘thinking machine’ that insisted 2+2=5, got angry like the Bing chatbot when it was told it was wrong, then broke loose from its foundations and chased Trurl and Klapucius into the mountains.
People will learn the hard way there is no market for machines that get the wrong answers. There are plenty of places where people will accept one kind of imperfection or another and that’s fine, but when it comes to an accounting bot that screws up your taxes it is not fine.
(Funny I have seen a few chatbots that claim they are busy as soon as you tell them they are wrong about something and I wonder if that is because they’ve been trained on many examples where things really went south when somebody called out somebody’s mistake.)
>People will learn the hard way there is no market for machines that get the wrong answers.
Eh, in machines there are markets for machines that get it occasionally wrong as long as the defect rate is lower and or cheaper than the human equivalent defect rate cost. So I'd say that's a really bad take on the last few hundred years of industrialization.
In accounting there are really two different sets of things going on at once. There are 'the numbers' of which we'd run in a calculator, but then there is interpretation of the written rules in relation to the numbers you input. If your business is in any way complex, take your numbers to 3 different firms and see if even two come back with anything close to the same number. Hell, in the same damned firm we commonly see that two auditors will come up with different numbers and a supervisor had to look at the rule in question and make a judgement call on what they think the IRS agent would accept.
Now, don't confuse that with me thinking that using ChatGPT is a good idea to do the above. We are not there yet.
I share your reflection on the current state and foreseeable future effectiveness of AI, but
> People will learn the hard way there is no market for machines that get the wrong answers.
I fear this is wildly untrue.
To take your example: The wealthy won't use the accounting bot. Everyone else won't have the time/energy/means to recoup whatever it cost them.
In software development one only needs look around to see a world of "markets" - that is, profitable opportunities - for wrong answers (bad designs, useless products, orders of magnitude of inefficiency).
On the one hand, GPT and AI is immediately usable and arguably quite cool to the layman. It's not a perfect system, but it can write Bash scripts and handle your resignation email. Anyone who does copywriting or programming for a living will probably find a couple uses for it. It's fairly cheap, and it's definitely "worth it" more than analogs like Cryptocurrency.
...on the other hand, though, AI is kinda a bubble. It's our nature to put our hopes and dreams (and money) into the most-promising fields of computing; that's fine. But AI truly has limited application - anything that requires accountability, determinism or trust is a bad fit for AI. When you really whittle down the places you can apply AI, it looks a bit like those 4D theaters from the early 2000s - a cool concept for some stuff, but experience-ruining for others. The eager over-application of AI will be the bubble, not the technology itself.
The bing stuff can get very wild. But I think the reason they aren't making it widely available yet (and won't for a while) is that it's just a test for "ChatGPT that can search the web" the same way that ChatGPT was (is?) a public test.
Sure, it's an unending stream of things, but here are some recent examples from my history:
-Had it extract some gnarly string manipulation logic in a MySQL where clause into a function. It wrote clean, well commented code, and even explained what a regex did in a comment.
-Had it explain to me how macOS plist files worked, how to view them, how to edit them
-Asked it what some lines in an ssh config file did and it clearly explained it to me
-Dropped in some XML from a spring applicationcontext file that was doing something I didn't quite understand and it explained it to me. Asked how I would change something about it and it told me how, explained how it worked, and provided example xml
-Gave it a SQL Server CREATE TABLE statement and asked it to create the equivalent CREATE TABLE statement for MySQL 8, which it successfully did
My only complaint has been how sometimes the system has been unavailable. I now pay for the $20/month subscription and that has been resolved.
Thank you for those examples. Converting sql manipulation into a function is an excellent example. You can do a lot in a few case statements especially with nesting but translating that into something else is valuable.
I asked it to write some queries in relational algebra and it was close but it was invalid in ways that a non-database person wouldn't understand. Kind of looked right.
There exist problems where finding the solution is hard but verifying the solution is easy. example in the general case: prime-factoring numbers.
Are you claiming that for all possible problems posed, verifying the correctness of the output of an LLM is equal or higher difficulty than solving the problem itself? If so, that seems like a claim more arising from emotion than reason.
In the general case, you can have it generate code in a proof based language that "proves" the code is correct against formal specification. Unless you consider math itself to be "bullshit" too.
The people who are skeptical that it can be easier to verify correctness than to produce a correct work might as well try to prove P=NP (which is the theoretical and formal statement of such).
They also probably never asked people to work on a task either (hint: people can get things wildly wrong, even generally competent ones)
If it is a well known problem, I might as well find the solution on stack overflow, where actual the actual humans that trained this system discuss it.
As for proof based languages, can it actually do that? Have you tried?
Within the constraints of technical systems, it's pretty obvious. It's not a topic that you can really bullshit - things work or they don't. It makes mistakes sometimes and what it says won't be consistent with other things it's saying or that you know to be true. If you ask it if it's sure, it often will correct itself. Worst case, you verify with other sources and/or try it out.
It's not really any worse than asking someone knowledgable about a subject face-to-face. They know a lot, but they can be mistaken about things. At some point you will tease out the misunderstanding.
The best way to understand the value of the tool is to try it for yourself.
My partner is looking for a new job, and ChatGPT has been an amazing help in writing the cover letters that recruiters can't stop asking for. Previously, she would spend hours molding her previous cover letter to the job description. Now she can just give the job description and a bullet point list of her relevant experiences, and spend 15 minutes on the ChatGPT output to polish it a little bit and correct any potential misstatements.
As they should be! I do not know why they are not already, but as I said recruiters can't seem to stop asking for them. But be assured that I won't mourn them once they are gone, and if ChatGPT gets us there faster then I am grateful for it.
Edit: on second thought, I believe "Editing ChatGPT output" might be a valuable job skill in the future, and maybe a good cover letter would give recruiters some signals in that way.
I tried using chatGPT to write me a cover letter and targeted resume given my full resume and the job description. It gave me the cover letter and the resume - but it made stuff up to match the job description. I have a masters degree but it gave me a PhD, for example (I'm not sure why it did this as the job requirements were Masters of PhD, I guess it figured a PhD was better). I'm pretty sure I would have easily gotten an interview for the position based on that cover letter + resume as it was essentially a perfect fit to the job description, but the interview would've gone badly.
That's why the 15 minute review process afterwards, at least for a cover letter the factual errors are glaring enough that they can be caught easily, and simple enough that they can be edited out in that time. It's not at the stage where I can write my partner a job application bot that scrapes job boards and fires off 50,000 applications per second, but the productivity gain is obvious.
This to me is what I see want to see as the fallacy in the anti-ChatGPT argument, but at the same time have to acknowledge as the real danger. I think AI paired with a human, as you are describing, can be more effective and efficient than either on their own. However, I've never seen it stay that way in commercial applications. So I'm afraid that the smallest drive towards cost-cutting will lead to removal of the human in the loop or at least tighten throughout expectations so much that the human intervention is in practice more often missing than not.
Not sure what to do about this. I want ChatGPT as a tool available to me, but I don't want any service provider I interact with to use it because I know they'll fuck it up eventually and I'll get worse service.
> I want ChatGPT as a tool available to me, but I don't want any service provider I interact with to use it because I know they'll fuck it up eventually and I'll get worse service.
This is just straightforward selfishness and narcissism.
No software developer should be so naive to assume they'll always do the right thing with a powerful but incomprehensibly dangerous tool.
You know how many minutes I "spend molding [my] previous cover letter to the job description"? 15 minutes. If you spend more than that, you are clearly doing it wrong.
Not them, but a short list based on my chat history:
- Explained why a python variable in a closure was saying it was not assigned, after several minutes of googling/SO failed me (needed the nonlocal keyword, different from js)
- Explained the difference in several slightly-different C++ lists, which I couldn't google because the differences were mostly special chars
- Wrote working powershell code to output a list that has every string in $list1 that is not in $list2, on the first try, after I struggled for over an hour to do the same (and google/SO failed me).
- Explained the difference between two batfiles for the Intel Fortran compiler after google failed me
- Generated some really cool D&D statblocks
-Unbeknownst to us players, generated a really creative heist dungeon that our DM ran us through (he thanked ChatGPT as his "co-DM" at the end). This one is particularly notable because the man works at a hiring agency, he's not exactly a techie.
> after several minutes of googling/SO failed me
**
> and google/SO failed me
This is what I've been noticing as well. There have been many times when I spend 5-10 minutes trying to figure something out with Stack Overflow and Google, not getting anywhere, and then ChatGPT immediately gives me the answer. I'm not sure how anyone who considers Stack Overflow useful would think that ChatGPT is useless.
Do I need to verify the answers? Sure. But I have to do the same thing with answers on Stack Overflow. Even if I'm using official documentation, I need to verify that what I think it's saying is what it's actually saying (particularly when it's as unclear as it often is).
What's even more interesting is that I keep running into people who are finding completely different uses for it.
The critique here is like a lot of other critiques I've seen. They seem to come from people who haven't spent much if any time working with it, but have read some articles about how it's not infallible and decided that meant it was junk (though this piece goes the extra mile when it hints that big GPU is behind it all). What's striking to me is the lack of intellectual curiosity on display.
+1 for the DMing. I mostly use it to generate data for my dev/test tables, but the two times I used it for DMing, I gained hours. Also live when the player go somewhere really unexpected (it wasn't my campaign).
Tangent:
Google is less and less useful (as is SO for 'old' languages). I look into a mix of reddit, github issues, github code and SO depending on the issue. Hopefully this advice helps you if Chatgpt/copilot can't.
If you treat it as a surpercharged search engine on a compressed snapshot of the internet in 2021, then its quite useful. If theres ever a function I forgot, but I know how to explain what it does in natural language, chatGPT most of the time can find me what I'm looking for. On some more obscure bugs or if I'm sorting through a new codebase, chatGPT can help me out from time to time to understand new topics.
Of course, we shouldn't rely on chatGPT. It has give me wrong and insecure code before. However, its a nice tool to have around
>Of course, we shouldn't rely on chatGPT. It has give me wrong and insecure code before. However, its a nice tool to have around
You 100% should verify any code generated by ChatGPT - but this goes for any code found off the internet. I have come across bad Stack Overflow code committed to codebases before.
I think what's making these hyperinformed hallucinators useful for coding right now is that fact checking is fundamentally a big part of our work. There's a bare minimum that must be done to even think anything has been accomplished at all, that it runs and does what we expected on the most basic input, but then we are also used to writing tests and reading code critically.
I wonder what a learning model with the ability to test feedback from executing its code in an interpreter would look like? I know there are different groups testing things like integrating the ability for LLMs to use tools, wondering how this will pan out in the end.
If you think cryptocurrencies are in any way comparable to the recent progress in artificial intelligence, I’m sorry but you’re a Luddite.
If you’d have presented anybody reading this with ChatGPT two years ago, they simply would not have believed it was possible.
This kind of step change in AI ability is interesting enough in itself without having some kind of insidious and completely unsubstantiated conspiracy theory behind it.
> This is not a coincidence.
I despise this kind of writing — writing with 100% confidence and then providing zero evidence. It reminds me very much of my reading of writing from “the election was rigged!!!” conspiracy morons. 10 minutes of my life I’ll never get back.
> If you’d have presented anybody reading this with ChatGPT two years ago, they simply would not have believed it was possible.
2 years ago, Google Lambda was available internally for dogfooding. The hype was high for a couple months, and then it dropped off after the entertainment and novelty value wanned. It even led to a series of articles about a person claiming it to be sentient.
So yes, 2 years ago, plenty of people would have believed it was possible.
How i wish such PR firms would truly exist. I fully get the point of the article and still disagree. You know what could be also true? That some people don't get the appeal of popularity which is fine but unfortunately the hard currency in product success
> If you’d have presented anybody reading this with ChatGPT two years ago, they simply would not have believed it was possible.
I'm having a hard time believing it now, it's just an amazing amount of progress in a short amount of time.
> I despise this kind of writing
Me too. It's suggesting that there's some kind of cabal that somehow decided it was time to crash crypto and get everyone to go after AI. Are there bandwagons? Definitely and they eventually go too far as with crypto - and quite possibly AI will experience something similar if it doesn't deliver on it's promises. But it's not like there's a bunch of guys in a smoke filled room somewhere who are driving this. Crypto was largely driven by cheap money (0% interest rates for several years) and artificially low electricity prices. Investment in AI now with much higher interest rates means that some people think they're going to get a better return on their money. It seems less bubbly in this kind of monetary environment.
That actually wasn't what he said. He said the hype around cryptocurrencies was comparable to the hype around AI. That's a very different statement than you're putting into his mouth. If you've ever read any of his books (they tend to be very good), you know that he's not a Luddite and has a lot to say about future technologies including AI.
Cryprocurrencies are juxtaposed in the article to ChatGPT-like AI in two ways.
* 3 years ago, BTS / ETH / a whole slew of other coins were all the rage, the centerpiece of popular speculation. Now AI takes their place, even though without the get-rich-quick overtones.
* Many cryprocurrencies and AI both massively benefit from the use of GPUs, so once the crypro bubble burst and a lot of GPU compute power was freed, it can now conveniently be used for AI.
LLM based chatbots are already more useful than all of crypto combined. Whereas crypto wastes energy as "proof of work" AI models try to be as energy-efficient as possible. The energy needed to train these LLMs is also 10,000 times less, so let's not act like they're in the same universe.
Moreover, training the base layer of a LLM has to be done only once (for each generation) and subsequent training for domain specific purposes can happen on top of the base layer. This means these AI startups are not going to require an unlimited number of GPUs.
Another big difference is that fundamental AI insights are what drives these AI innovations. Throwing more training data at a neural network is not what makes ChatGPT different from older generations of chatbots (although ChatGPT certainly does benefit from its huge training data set). Transformers for example are a new development, first introduced in a 2017 paper "Attention is all you need".
It's unclear if LLMs will turn out to be a big deal. Maybe this will turn out just to be another AI summer. Too early to tell. But cynicism is unwarranted here.
> Another big difference is that fundamental AI insights are what drives these AI innovations.
Bigger models, more compute, more data, better compute have driven a lot of the performance of ML post-2013.
For ChatGPT specifically, the secret sauce is the addition of human-vetted conversations [1].
In general fundamental AI insights come when
- (1) somebody does something that works incredibly well and others try to explain it. Transformers are arguably an example of this although there’s a ton of attention literature before that.
- (2) a ton of people make incremental steps towards a goal and at some point it becomes actually useful. ChatGPT is an example of this one.
The point I’m trying to make is fundamental AI is boring, slow, rarely drives AI innovations and that’s not a bad thing. What drives AI innovations is really good UI and good data both of which are hard to nail down.
I think this sort of skepticism is healthy — I think explainability and overfitting doesn't get enough attention for example — but AI strikes me as a bit different from crypto in that it's creeping into our everyday lives. Instructors are wrestling with assignment submissions, you wrestle with the uncanny valley in customer support lines, and people are posting about their troubled relationships with Bing in various places.
If crypto had been showing up everywhere to the same extent, I think these discussions would be a bit different. That's not to say there isn't rampant hype in AI but I do think there's a continuum from blatant fraud to truly life-changing tech with plenty of grey, and AI is closer to the genuine end of the spectrum.
My guess is AI isn't monolithic either, just as I don't think crypto was. Some areas of AI will probably end up having been overhyped more than others.
I agree with the sentiment of the post that sometimes I think semi-fraudulent to outright fraudulent overhype is a defining feature of our time. I wish the institutions responsible for public discourse would approach things accordingly.
> but AI strikes me as a bit different from crypto in that it's creeping into our everyday lives.
Taking this further: Crypto was creeping into our lives, but only in the FOMO sense; People thought they needed to invest in crypto because other people were going to invest in crypto and they didn’t want to miss out. The actual use cases of crypto aren’t attractive to most people, aside from ideological or law-evading circumstances.
AI is different because people are rushing to actually use it for their own benefit. Investors are rushing to place their bets, but we already see big companies deploying AI in the real world in ways that are being used by the general public.
I really have little respect for any opinions that shame the use of electricity for things like training AI models. Being mad at it for crypto I can at least understand. But this should be at about the millionth place in the list of things to complain about electricity waste. And if that's how you prioritize your worries, I'm not sure how to trust any other opinions.
It makes me wonder, what is the energy consumption of a what it takes to nurture a human to be an expert in something?
Discussions of the intelligence in the LLMs as an emergent property of differentiating the model on troves of quality data, raise the question of how a human gains its intelligence. Starting from scratch, a child and a LLM, how can you compare their intelligence when provided with the same training data? What happens? What is the comparable intelligence of the two?
We sit here analyzing the energy cost of training a LLM, developing artificial intelligence. We have to compare that to the energy costs of human intelligence. One may argue we are here because of the previous energy use of civilizations prior.
An LLM doesn't start from scratch, it's ran on computer hardware...
The particular problem with talking about humanity is our biological functions are deeply intertwined with our intelligence functions.
Some things are programmed in, like cellular behavior, organ building, what to do with these systems after you eat food. The two bottom parts of the brain are rather well developed at birth that control these functions.
The cerebrum is rather undeveloped in comparison, and that's where the "thinking" happens, at least in terms of what we call intelligence and consciousness. While it comes preprogrammed with any number of subroutines, the amount of what we would consider data is minimal, and in our young years a huge amount of energy is spent exercising the brain so it works properly in the reality it exists. In children that is deprived from stimulation and real world experience (think extreme abuse), the brain underdevelops and it can become impossible for them to learn some things or think in particular ways.
Do not underestimate the amount of work that must be performed for an infant to become a thinking machine.
Can one argue that the random initialization of a model is akin to this starting point? A human brain has reached a morphogenetic starting point thanks to much evolution, a process full of randomness.
Before ChatGPT I spent around 200 usd per month to practice German with German native speakers for 2 months.
Then I use ChatGPT to practice German and I am very happy about the experience. AI can't replace real human yet. But I'm already willing to practice with ChatGPT given my current use case and my budget. It's not 10x better, but 20x cheaper. From my experience, AI is not hype.
Imagine integrating a more fine tuned AI and text-to-speech feature to make a better AI teacher. Such an app already exists, I believe. I'm waiting for a German and Japanese one.
This is a great example of a concrete use case. It is well known that tutors are far more effective than generalized learning, and everybody’s anecdotal experience is that immersion is the best way to learn a language (suggesting that the interactive effect of tutors is at least as strong in this domain)
A quick Duck Duck Go search suggests that capturing just a fraction of this market is worth >>$1B. To me that feels categorically different than crypto, even if a few grifters are also adding froth to the space.
Where did you find these German native speakers to practice with? And you said "such an app already exists", but not for German and Japanese. Do you know the name? I have a lot of experience building language learning services so am very curious to hear what you feel is currently missing in your current tools to learn German/Japanese.
What I don't like about a real tutor is because it's expensive and I have to book in advance.
ChaptGPT is free and I can practice with it whenever I want. Besides, ChaptGPT is not a real human so I don't feel embaraced when I made stupid mistake.
How much does it currently cost you (per session) to book someone on iTalki?
And which would you prefer: a cheaper, on-demand version of iTalki (you get matched with someone already online versus booking in advance), or an AI-based app/service that you can chat back-and-forth with whenever you want?
I pay $15 to $20 per session (around 45 ~ 60 mins), depending on teachers.
The problem with matching with someone already online is that it might not be the right person. There are tutors that are very good, helpful and inspiring. I don't think I want to talk to random person online. A long term relationship with several tutors are helpful because we know each other better and I'm more comfortable speaking in front of them.
Just because a bunch of crypto grifters are finding a new field to dip into doesn't mean that there hasn't been a lot of genuine progress in AI in the past 6 months.
Aside from dumping a very large amount of compute and human-supervised learning into existing technology, what major progress has been made in the past six months?
I mean, Stable Diffusion came out mid August (almost exactly 6 months ago) and the past six months have seen tons of refinement by hobbyists, so we went from having what amounted to a tech demo with practically closed access to a really functional diffusion model that you can run on your average gaming machine. I'd say that's kind of major.
Read it again, but this time turn off your preconception. I asked a question I wanted to hear a legitimate answer for. I am very interested to hear where this is headed (to the best of our ability to predict). The results we can get with sufficient computing power is quite interesting, but is it reaching the point of rapidly diminishing returns? What's the next step in this evolution, beyond making the training faster or adding more parameters to the model?
I think that's an interesting question that you won't be able to get an answer for, because most new things are still not accessible to the average HN reader. I mean, you can look up ML research papers to see what's new, but none of that is going to seem anywhere near as impactful as ChatGPT because it's at the level of research.
It's not like when the first theoretical models that make up the basis for GPT came out, people were instantly sure where we'd end up in 5 years.
I'm sure people within OpenAI or other AI companies would be able to tell you of very interesting advances, but they won't do so on HN until the things are public.
You mean things like GPT showing theory of mind at the level of a 9 year old, something that many people were arguing may not ever happen and/or if it did we were still a long way away from it just a few years ago?
I suspect that there are plenty of Crypto enthusiasts who would say the say the same thing about crypto - the tragedy is that if the grifters are successful it's going to poison the AI market for another 5-10 years.
I think there is a fundamental difference when you have an actual usecase.
You can do useful things with AI, the market can reward people for being useful, and push out the stupid things.
You can't really do useful things with crypto. The ecosystem was a gradient between straight up fraud, to sketchy financial engineering projects that well technically not a scam are right on the border of being one. There is no room to reward actual progress so obviously the money is going to get dumped into the sketchy shit.
It's mind blowing that we have something blossoming out of nowhere that has real-world use-cases and is genuinely improving people's lives.
I have tools to write the perfect e-mail, remove the background from photos, write copy for my clients' websites, and tools to help me code faster. It's all very brilliant stuff.
> Much of what passes for "journalism" these days is just stenographers feverishly copying the press-releases they're spoon-fed. Real journalism is a niche sector, and unless you subscribe to the exhorbitantly priced newsletters of the high-end analysts who are paid to work full-time studying the sector, what you're seeing on the news websites and in the newspapers is the product of PR firms paid to push AI. And you really need to ask who is paying them.
I think people who seem overly eager to call chatGPT and stable diffusion and similar models "bullshit generators", and dismiss them because of that is missing something large.
In Machine Learning itself, we consider two classes of models/algorithms, discriminative and generative. Up until six months ago, the best work we had in ML was all discriminative, and humans were irreplaceable when it came to content generation. All of a sudden we have good generative models and a way to cheaply generate vast amount of content and ideas, some of which are good and some of which are terrible. I predict this shift will also shift the associated humans' roles from "generator" to "discriminator", who will be cherry-picking and editing AI generated articles or images (much faster) rather than creating them from scratch (much slower). Yes, it will take a slighlty different set of skills in the human worker, but if you can't see how that productivity boost can help humanity, I ont know what to say.
Disapppointing summary from someone whose imagination should allow them to see more.
Generative text models might spit out "garbage", but this "garbage" has been more interesting than the garbage we have seen before in how it breezes through the Turing test.
It's telling that not a word was said about image generation, which is benefitting from the same GPU usage, has not stopped progressing, and produces stuff that creates economic value. Like in this example: https://arstechnica.com/information-technology/2023/02/us-co...
The author seems to identify with VCs too much, if all the AI hype means is more grift. Sure, that might be a factor, but the general condemnation of the sector is too much.
I really hope one of the AI startups (stealth or already public) gets some more traction, especially among smaller scale hardware. Nvidia is an AI tyrant.
But I guess we are probably stuck with Intel/AMD/Apple... because who else is going to make affordable PCIe desktop cards, much less squeeze into laptops?
Within the next five years or so there will be a completely new hardware paradigm based on something like "crossbar arrays" or memristors that is orders of magnitude more efficient for ML tasks. This provides an opportunity for new startups. They will eventually probably be acquired by Nvidia, Intel or AMD though.
This is my take on it too. Any technology that significantly reduces the cost of running AI is of immediate value to any company performing a lot of AI simulations currently.
I nearly pulled my hair out trying to get a Stable Diffusion UI running on Graphcore's free POD4 instances, and... gave up after I couldnt even download their SDK to update it. Even though its theoretically extremely fast.
Exactly. That's the eigenrhythm of present day capitalism. Throw money to likely candidates, see what sticks. Speculators always capture headlines. Crytpo is chugging quietly along.
While the author's prediction may or may not come true, this text has a weird conspiracy vibe in my opinion. Especially parts trying to imply that by some mysterious way NN are trained on GPUs and manufacturers are pushing AI to sell more cards, not unlike the rush to get GPUs for crypto. It's almost as if GPUs have been specifically built to handle massive parallelism with specialized models offering machine learning optimizations! However, the cards used for these vocations are different: while you used to be able to mine cryptocurrencies with consumer GPUs, training AI models is done on professional hardware with more emphasis on VRAM. Training something serious like ChatGPT is far from easy (if at all possible) for a startup. And AI projects have arguably become more useful to the everyday consumer than crypto currency ever was. Either way, AI is certainly a huge buzzword, but for now, it's living up to the hype. Who knows what'll happen three years into the future.
I disagree with Charlie Stross's comparison of the hype surrounding AI applications to the cryptocurrency bubble. He suggests that much of the AI hype is driven by the same grifters and hucksters that were previously involved in the cryptocurrency market. While both industries have seen a surge in investment, they are fundamentally different. Cryptocurrencies were based on speculative investments with little intrinsic value, whereas AI has the potential to revolutionize many industries and improve human lives. Charlie's skepticism and pessimism towards AI startups seems misplaced, as there are already many successful AI applications in use today. While caution is always advised when investing, writing off the entire AI industry based on comparisons to cryptocurrency seems shortsighted.
> If you're thinking about investing in AI startups now? My advice is to avoid them like the plague [...] it's too late, and you'll wake up one morning only to discover you've had your pockets picked.
> I give it about 1-3 years until the crash. (Although I tend towards optimism: the cryptocurrency bubble took a bit over a decade to implode, but back in 2011 I prediced its demise within, yeah, 1-3 years.)
What I'm reading is, I should invest in AI startups now, and cash out 90% of my investment in 3 years time while things are still going well. Maybe cash out 30% of whatever remains each year after that, until the crash.
While some of the recent hype around AI may be due to VCs and the industry looking for the next big thing, it’s a bit absurd to think that AI orgs like OpenAI and Stability were holding back research and then only released it once the crypto bubble popped.
This article also misunderstands the differences between the various types of GPUs and seems to think they are fungible.
Large models like GPT-3 and Stable Diffusion are trained using cloud compute, most likely clusters of Nvidia A100s. No one was ever using those for mining because it wouldn’t have been profitable. Similarly, no large models are being trained on second hand GPUs used for mining.
Machine learning has a decade-long track record of adding value to consumers and companies in various sectors. Crypto does not. There's probably too much hype, but it's not completely unfounded.
A huge difference between ChatGPT and crypto, though is that the latter literally offers to print money.
The former, what, lets you make low-quality internet content a little more easily? It's harder to share the long-term 'get-rich-quick' benefits (though, of course many are trying right now). There's an initial bump right now as various content mediums try to cope with ChatGPT content, but once that stabilizes, I don't see the long-con grift that the author is mentioning being possible in the same way.
The hype around ChatGPT is massive! I haven't seen this much buzz since the iPhone launch. Back in 2008, I remember having debates with my friends about the future of Nokia just a few days after it launched. Since then, I've witnessed the rise and fall of several tech waves, including mobile apps, cloud, SaaS, and crypto. It's been exciting to see so many people ride these waves and create wealth in the process. Will I be one of the lucky ones this time around is the only question.
There are a lot of "bullshit jobs". Not needing people to do those jobs anymore will be a benefit to anyone paying for them, less so for those previously getting paid.
I agree with many of the top-level comments about the relative utility of AI applications and blockchain applications as they exist today. But I'll add another point that no one else seems to have made yet.
The reason for the widespread grift in the crypto space is not just that blockchain applications are a relatively new and overhyped technology. Another huge factor is that many blockchain applications, by their very nature, are designed to circumvent securities regulation in order to raise money from unsophisticated retail speculators.
I'm sure some shitty AI startups will raise capital from retail speculators on crowdfunding platforms and the like. But I expect it to be a lot less widespread compared to the crypto space, and it will happen within the confines of (mostly) US securities regulation, which will mean much better disclosure and much less outright fraud and misappropriation. The main "victims" this time around will be, like, the LPs of second-rate VC funds, and in many cases those fund managers are well aware of all of this but are incentivized to follow the trend anyway.
I know this forum is full of businessy types but it's still amazing to me to hear so many people saying crypto has no use cases and AI is already proving useful. I have my own healthy skepticisms about both, and see some really interesting use cases in both. Obviously there are and will continue to be donothing hucksters with magic potions, like in any industry, especially new and misunderstood ones. They will do their ruinings with no concern for actual progress.
However can we take stock of a few things here?
Crypto:
- opt-in, democratizing
- privacy preserving, data protecting
- open, transparent, auditable, provable
- peer to peer
- extremely cheap to participate
AI:
- trained and used without consent
- massive data ingesting and processing
- closed, blackbox, censored, inscrutable
- centrally, corporately controlled
- prohibitively expensive to participate
So when I hear people saying that crypto is dead and has no use cases, and AI is useful already and is solving real world problems, I think I know what kinds of problems they are trying to solve.
The examples I'm seeing so far of people getting "real world" uses out of AI are like downright grotesque to me. Writing cover letters? Making "content" for marketing websites? Writing emails?
Who wants this? Only the makers of these things want them. Not the receivers. As if the corporate internet couldn't get any more boring and derivative, let's put it all in a big pile and then churn out infinite replicas of it in a vast grey fog.
Upvoted because someone is publicly defending cryptocurrency on HN. I appreciate your bravery. You are right on those points.
The AI take has some valid criticisms but overall is the wrong assessment. People can use it to make spam, sure, but they can also use it for a thousand other things. To generate pretty much any type of text content. The only way that is not useful is if you just don't place value on natural language at all. Its a pretty important part of our world. But beyond that it can be used to create a natural language interface to many types of systems that would otherwise need to be programmed by a person.
Crypto has almost no use cases for the average person over the system the average person is already using. Crypto has a huge amount of risk over the average system the average person is already using. This dramatically reduces the perceived value of Crypto. Again, Crypto has uses, but not at the value put on individual tokens when compared to cash.
When looking at your AI description, if used in the same light, Linux should have won the world and Windows and all other proprietary software shouldn't exist. And yet here we are with billions/trillions in market value around proprietary software.
Some variation or piece of GNU/Linux is running on the vast majority of computing devices on the planet. One could argue it and its core philosophies have enabled most of the tech industry in part by being free. If that's not winning, I don't know what is. No one has starry eyes about it because people aren't making out like bandits being gate keepers to the tech. It's free, it's open. It's so free and ubiquitous that it's not even considered. It's just a bedrock tool. Like the hammer, or a fire. Has fire "lost" because it has a low market value?
Regarding crypto use cases, while they're obviously still being explored, the core features remain:
These are things you get when you use an open platform like, say, Ethereum, or whatever next iteration.
"The average person," whatever that means on a planet of 8 billion, may not need these things among stable, prosperous governments, with low corruption, upstanding law enforcement, and equal protection under the law, but anyone who lives outside those golden scenarios, either today or in the future, might desire such features. They might even help them survive.
There are plenty of LLMs that are completely open source that you can use (like BLOOM for example). There's also a massive original use case of machine translation for LLMs which is a real world problem. AI image generation seems WAY more useful thing to me than say NFTs.
It's kind of a non-sequitur. That's like if I said that Linux is opt-in and then you said "I don't think that individuals have a choice in the matter when they get kidnapped by a drug cartel and forced to run their IT infrastructure". Yeah, nothing is opt-in if you're being forced by a criminal to use it. What?
Now what you've just done is a complete fanciful strawman.
> It's kind of a non-sequitur.
Nope, the difference with my argument is that this is actually happening in the real world. Ransomware gangs trying to destroy companies data and they can only use crypto to save their documents, which is one of the reasons why the crypto is very appealing for criminals.
Nothing entirely wrong with that, except that proves my point. There are no legitimate use cases of crypto, and usually the current financial system does better if not more than what crypto tries to make even worse.
A lot of white collar professionals that I personally know, who didn't write a single line of code in their entire life's, use ChatGPT daily. Sure, when you just ask it questions, it hallucinates bullshit a lot. But when you ask it to process your input, for example, to write a professional sounding document based on your hectic bullet points, it's stellar. And a lot of white collar work is exactly this.
GPUs aren't need for blockchain. Bitcoin is stuck in the past and is a shame it's what everyone looks at, along with the misnomer "cryptocurrency". Ethereum is the leader in the space and the only one innovation and uses PoS which can run a validator on very little hardware, such as a raspberry pi.
Crypto is still way bigger. I don't think the comparison holds. ChatGPT is mostly jsut a tool, whereas crypto is hyped as somehow replacing the current financial system or even 'world order'. There is considerable hype, but not on the level seen with crypto.
Both right and wrong. There will be bubble, grifters and there will be a lot of pets.ai , but I think that unlike crypto the technology is sound, impressive and promising.
The advice to be cautious with investing is a sound one.
The thing I find most suspicious/fishy/smelly about the current article is that Charley is a cracking good writer who has just found that a machine might be able to take his job.
Now, of course, AI is all a giant scam. It's been steadily progressing in a way that connected people could see since GPT-2[0]. But only now, with the clouds on the horizon is it clearly and obviously some conspiracy.
"It is difficult to get a man to understand something, when his salary depends on his not understanding it."
It is funny how NFTs and crypto fell off the radar so quickly. It happened so quickly that prices didn’t collapse because hardly anybody trades anymore.
I think people like Eliezer Yudkowsky are worked up about ChatGPT because it bullshits so much better than them.
which can do 13 different things and text generation is only one of them. Most of them involve setting up well defined problems (e.g. being smart) and building training sets (e.g. working hard) and aren’t the shortcut that people think ChatGPT is.
As I see it things that I was struggling with just a few years ago are downright easy, there is no doubt “the new AI” will find useful uses but somehow ChatGPT strikes at a vulnerable place in many people’s psychology.
If you have a real-world text-related task that is somewhat complex then you can use a custom prompt with OpenAI's API to achieve it within a few days. I can imagine someone hiring you for something like that and you charging them $20,000 for two weeks of completely wasted time due to whatever cognitive or philosophical hang up you have with using the more powerful pre-trained model.
