Most of the prompt answers are smart sounding bullshit. Maybe that's why the headlines never stop - the people who like to make smart sounding bullshit are the ones who love ChatGPT.
I understand it's 'just' a language model, but it doesn't matter - that's not how it's perceived, and it is actually rather impressive anyway.
What you're seeing is just a manifestation of a rather major (sociological) event, and while I understand that the amount of hype is a bit over the top, to a large extent it makes sense.
If somebody thinks it's just a bullshit generator and 100 million people using it after 2 months are wrong, the problem is with the person who didn't put in the effort to learn to use it effectively.
Alternative explanation: Most people are happy to generate bullshit, and love that they now have a way to do it with zero effort. Bullshit copy, bullshit art, bullshit code — the sky's the limit!
Attention and profiling is money and power.
AI will steal attention by spamming identities, voting for/against products and opinions, discriminating in ways hard to reveal, at scale never seen before. It will read all your HN comments and clip a tag onto your ear. AI war will not be a T1000 walking on skulls, it will happen in our feeds, and platforms/governments will have no idea who’s artificial or russian this time.
Honestly it’s pretty concerning when I think about what can be done with the internet and societies by bad actors automating and weaponizing existing chatgpt and pic/video tech alone.
edit: I get very strong flashbacks to when wikipedia was new and people had to learn the hardway it wasn't always correct/up-to-date/etc.
For me, this is what gets exhausting. It is an impressive language model, and it is going to change a lot of things, but it isn't the Apocalypse and it isn't the Messiah. On HN I expect a certain level of technical understanding that is weirdly absent on this topic.
An example: every time the question of accuracy comes up people seriously suggest that you just need to prompt it for its sources. Or we just need to improve the training data. There's no recognition that there are tasks that a language model is just fundamentally unsuitable for.
I'm genuinely impressed about ChatGPT, and have been thinking about many times in the past when having such a tool at hand would have been massively helpful. Natural Language Processing is a damn hard problem, and ChatGPT seems to be a huge advancement in that regard.
But I actually laugh at all the people that think that this will replace humans in any meaningful capacity. If your job is only giving known answers to known problems, then you have something to fear. Otherwise this will only be a powerful productivity tool.
A Language Model will replace software developers much like Excel replaced accountants.
But Planet Money also found that there were 600,000 more jobs for regular accountants. After all, crunching numbers had become cheaper, more versatile, and more powerful, so demand went up.
The point is not really whether 600,000 is more than 400,000: sometimes automation creates jobs and sometimes it destroys them."
Just throwing numbers on a spreadsheet didn't do the trick.
Sounds to me the existing jobs are more skilled than the former ‘accounting clerks’ which just sounds like data entry people.
Back in the day my mom used to do the books for the grocery store she worked for and all she really did was tally up all the data from the multiple cash registers and send it off to the corporate accountants. Not a whole lot of skill needed aside from attention to detail to ensure the totals were correct.
To me, the most impressive part of ChatGPT was not that it could give mostly correct answers to known problems. In a sense, internet search could do it already (just in a much more cumbersome way), with similar degrees of correctness.
The most impressive part for me was actually how seamlessly it parses and produces fluent natural language. Text generated by it reads like something a human would type.
So far I didn't try to fool it by purposefully asking something ambiguous (something that is a characteristic of natural languages), or ask about something that has an ambiguous answer to see how it handles it, but so far I'm impressed.
But I never considered that people may restrict the research of AI to language models due to the rampant success of this avenue of research. I hope this is not the outcome, but I wouldn't be surprised (i.e. the success of ChatGPT works as a blackhole for investment in the area, with everyone racing to cash in on it).
I’m really not sure where you get that from. I think the best we can say is that they are currently unsuitable. Yes, today the applications are limited, but you can’t blame people for projecting this incredible progress into the future a bit and seeing glimpses of how much this could change work and productivity. Insider reports are surfacing that both Google and Microsoft will be adding LLM outputs to their search engine result pages. Seems like a big deal to me considering transformers are only about 4 years old.
Of course they are—they'd be incredibly stupid not to ride the hype wave while it lasts. A business choosing to take advantage of current trends is not evidence that those trends are founded in anything substantial.
This is not possible. Of course, someone providing GPT as a "service" could provide bogus "sources" for particular output and people might take them at face value.
It is a wakeup call though. Considering how quickly this was developed, the likelihood of either of these manifesting within our lifetime became much more plausible.
This hyperbole is exactly why people like OP are getting sick of the headlines. It may be a major event sure, but only hindsight will be able to say that for certain. It may be another novel gimmick though, and this pronunciation will seem ridiculous in a year or two if that's the case. It's just too early to tell.
We asked some questions about her field, and even asked it to draw a cell through SVG. My mon confirmed that all if its output was correct.
Mom was completely amazed at what was happening. She couldn't believe it was answering like that, and so fast. For people 60+ ChatGPT is StarTrek science fiction.
These are the guys that had to write dead tree letters to ask for a paper in the other side of the world. And shuffle through stacks of physical bibliographic cards. Also they spent hours looking through tomes of Encyclopedia Britannica in search of info.
Us younger folks have just been slowly boiled like the frog with all this new technology. But it is indeed amazing.
So, every time I see it mentioned I feel depressed and sick on the inside.
I will say, however, that so far it has actually enhanced my productivity in my current projects, and that’s fine. I just don’t think that its impact will remain so limited for long.
Of all the tech that’s been invented, this is the one I fear the most in terms of its negative impact on jobs. Hope I’m wrong!
If anything, I would say regular office workers are the most threatened, particularly if the jobs revolves around digesting and passing information.
As a programmer, I gotta say if you're not at all concerned, then either you haven't been paying attention or you're in denial. Sooner or later it's coming.
The newer research papers coming out are where I would focus if I'm really worried. ChatGPT is really the industrial grade implementation of ideas that aren't exactly new. And the idea itself (LLM) does not contain ability to generate novel logic or solve unlearned problems. In fact, I would go further and say that it does not see your prompt input as a logical problem, but rather a collection of words, and its output involves a transformation of words with its extensively training set. Which would contain logic that was in the training material but itself did not add anything to it, with no guarantee of correctness, only training weights.
There are no guarantee that it didn't also train on the question part of stackoverflow... and you only get answers as good as its training material.
Where I work we attempt to automate away repeatable problems with deterministic frameworks, which are imo far better. I would agree that ChatGPT will probably improve my writing if I were to use it on email and documents, but only in language, not in ideas.
Not with language models. A language model can parse natural language, and with enough training data, give out what it thinks the answer is based on the data it was trained with. It is not General AI.
It cannot reason a solution for a problem that had an unknown answer. It won't be able to reflect logically on a context to foresee problems within this context. It cannot have a meaningful conversation. It won't be able to understand that one of the things it "knows" was incomplete, untrue, or just plain wrong, and fix itself.
It's a powerful tool, a game-changing tool. Perhaps as game-changing as the advent of computers, internet, or wireless communication. But it still won't replace humans.
General AI for now is science fiction. Perhaps this is unfortunate. I wouldn't mind an AI that can replace humans, even if I too am made obsolete with it.
Maybe I’m optimistic, but I feel like we need AGI to reach the next level of development as a civilization. If software engineering jobs are the price to pay so be it. World hunger, medical science, energy, space travel, if we can get all of these to take a ride on something resembling Moore’s Law we are in for a one hell of a fantastic future in our lifetimes.
I also think it will solve things such as energy and space travel. World hunger, medical science (among other human problems) will become meaningless.
"Rejoice glory is ours / Our young men have not died in vain / Their graves need no flowers / The tapes have recorded their names"
There only needs to be one intelligent species in any ecosystem. If AI becomes that species, humans will be relegated to the role of horses - only good for menial physical labor. That is, of course, only until the AI invents cars.
When my job is at risk it will be because we have AGI, not a better language model, and at that point everyone is at risk. Worrying about job security in the face of AGI is like worrying about what you'll do after your city gets nuked: it's unlikely to happen and there's nothing you can do if it does.
The programming job market going away will be the last thing on my mind at that time, I think it would be one of the biggest shifts in human history.
OK, ignore everyone who doesn't understand the technology. Of those of who do, I'm utterly amazed how pessimistic many are that this "isn't capable" of leading to AGI. Probably not Transformers specially, but LLMs show that intelligence is remarkably easy. You don't even need to put anything in the neural architecture designed to perform reasoning tasks, but they can be learnt regardless, because Transformers are flexible enough to learn to emulate computation (Turing machines) with bounded space and time, going beyond the famous result that 2-layer MLPs are universal function approximators.
That sounds interesting. Can you provide a reference to this research?
A more interesting example of transformers learning a process may be [1].
There's a large literature on applying language models to reasoning tasks, but not many on what's actually going on inside them. But see for example [2]. Also https://transformer-circuits.pub/ has a body of work on it, but still at a very early stage (see in particular "In-context Learning and Induction Heads").
[1] Extraction of organic chemistry grammar from unsupervised learning of chemical reactions https://www.science.org/doi/10.1126/sciadv.abe4166
[2] Analyzing the Structure of Attention in a Transformer Language Model https://arxiv.org/abs/1906.04284
LLMs show that language is remarkably easy. Ever since GPT-3 was released, I've been convinced that language comprehension isn't nearly as big a component of general intelligence as people are making it out to be. This makes some intuitive sense: I recall a writer for a tabloid expressing that they simply turn off their brain and start spinning up paragraphs.
But so far, I haven't seen any of these models perform logical reasoning, beyond basic memorization and reasoning by analogy. They can tell you all day what their "reasoning process" is, but the actual content of any step is simply something that looks like it would fit in that step. Where do you derive this confidence that advanced logical reasoning is a natural capability of transformer models? (Being capable of emulating finite Turing machines is hardly impressive: any sufficiently large finite circuit can do that.)
"X is the key to intelligence"
computers do X
"Well actually, X isn't that hard..."
rinse and repeat 100x
At some point you have to stop and reflect on whether your concept of intelligence is faulty. All the milestones that came and went (arithmetic, simulations, chess, image recognition, language, etc) are all facets of intelligence. It's not that we're discovering intelligence isn't this or that computational feat, but that intelligence is just made up of many computational feats. Eventually we will have them all covered, much sooner than the naysayers think.
Why should I have to care about those weird milestones that some other randos came up with once upon a time? I've never espoused any of those myself, so how is this supposed to prove anything about my thought process?
> It's not that we're discovering intelligence isn't this or that computational feat, but that intelligence is just made up of many computational feats. Eventually we will have them all covered, much sooner than the naysayers think.
Well, it certainly appears to me like there's a big qualitative difference between the capabilities you mentioned (arithmetic and simulations are just applications of predefined algorithms; chess, image recognition, and language are memorization, association, and analogy on a massive scale) and the kind of ad-hoc multi-step logical reasoning that I'd expect from any AGI. You can argue that the difference is purely illusory, but I'll have a very hard time believing that until I see it with my own eyes.
Because its the same thought process that animated theorists of the past. Unless you have some novel argument to demonstrate why language isn't a feature of intelligence despite wide acceptance pre-LLMs, the claim can be dismissed as an instance of this pernicious pattern. Just because computers can do it and it isn't incomprehensibly complex, doesn't mean it's not a feature of intelligence.
>Well, it certainly appears to me like there's a big qualitative difference between the capabilities you mentioned... and the kind of ad-hoc multi-step logical reasoning that I'd expect from any AGI.
I don't know what "qualitative" means here, but I agree there is a difference in kind of computation. But I expect multistep reasoning to just be variations of the kinds of computations we already know how to do. Multistep reasoning is a kind of search problem over semantic space. LLM's handle mapping the semantic space, and our knowledge from solving games can inform a kind of heuristic search. Multistep reasoning will fall to a meta-computational search through semantic space. ChatGPT can already do passable multistep reasoning when guided by the user. An architecture with a meta-computational control mechanism can learn to do this through self-supervision. The current limitations of LLMs are not due to fundamental limits of Transformers, but rather are architectural, as in the kinds of information flow paths that are allowed. In fact, I will be so bold as to say that such a meta-computational architecture will be conscious.
> Where do you derive this confidence that advanced logical reasoning is a natural capability of transformer models?
("Advanced logical reasoning" is asking a lot, more than I wanted to claim.) I was going off papers like [2] which showed very high accuracy for multi-hop reasoning by fine tuning RoBERTa-large on a synthetic dataset, including for more hops than seen in training (although experiments "suggests that our results are not specific to RoBERTa or transformers, although transformers learn the tasks more easily"). While [3] found "that current transformers, given sufficient training data, are surprisingly robust at solving the resulting NLSat problems of substantially increased difficulty" but "transformer models’ limited scale-invariance suggests they are far from learning robust deductive reasoning algorithms". I think that low scalability is to be expected, transformers don't have a working memory on which they can iterate learnt algorithmic steps, only a fixed number of steps can be learnt (as I was saying).
Unfortunately, looking for other papers, I found [4] which pours a lot of cold water on [2], saying "a deeper analysis reveals that they appear to overfit to superficial patterns in the data rather than acquiring the logical principles governing the reasoning in these fragments". I suppose you were more correct. I still think there's more than just memorisation happening here, and it isn't necessarily dissimilar to intuitive (rapid) 'reasoning' in humans, but as with everything in LLMs, everything is muddied because capability seems to be a continuum.
[1] Hofstadter, 2001, Analogy as the core of cognition, http://worrydream.com/refs/Hofstadter%20-%20Analogy%20as%20t...
[2] AI2, 2020, RuleTaker: Transformers as Soft Reasoners over Language, https://allenai.org/data/ruletaker
[3] Richardson &al. 2021, Pushing the Limits of Rule Reasoning in Transformers through Natural Language Satisfiability https://arxiv.org/abs/2112.09054
[4] Schlegel &al. 2022, Can Transformers Reason in Fragments of Natural Language? https://arxiv.org/abs/2211.05417
I just wanted to chime in and remind about the other part of my argument: my job is not threatened until we have AGI, and AGI would be so earth-shattering to the entire premise of our economy that there's literally no point in worrying about it as an individual. We can and should talk about society-level changes like UBI, but having individual anxiety about your own personal job is a strange response to the end of the entire global economic system.
Until then, language models just regurgitate sentences. Anything that relies on the sentences being logically coherent cannot rely on these models. This includes engineering, finance, and journalism (ideally, although I am aware of CNET's experiment).
After then, we salaried humans might be in a pickle.
Edit: sentence.
If any given music synthesizer company could produce it's own DAW (digital audio workstation) by employing a single programmer or a contractor, the demand for programmers could increase.
Just speculating, it's hard to know.
until that happens what's the point of worrying? the year of the linux desktop joke got retired anyway so let's use self driving cars and "ai programmers" :)
A big part of the failings of cryptocurrency and self-driving cars was that predictions of mass disruption were contingent upon the tech improving linearly from the current state. That didn't happen.
The AI wave is going to be bigger than the internet in terms of societal impact.
We're watching humanity's swan song in real time.
Swan song that started as the singers started regarding IQs of 0 as acceptable actors.
and in general, actions are based - prescriptively - on concretions of layers of developed intelligence (wisdom), intelligence which is constantly applied and exercised (just with different occasional effort, for resource management). If somebody went sparingly, their own practice and our damage - if they are active.
So, again: a "sub-bohemian" idea that "anything goes" is the swan song. Wait till you actually needed intelligent action, to prevent critical losses.
--
I will rephrase coincisely: you are supposed to exercise intelligence at any action. If you do not do it, your vice.
No offense, but if writing the actual code is the part slowing you down there's something very wrong.
> The hardest parts of software are nailing down requirements, scheduling work, and fitting square pegs into round holes for legacy systems at a conceptual level.
Then why do we hire PMs and have engineering ICs? The only reason any of this is hard is because 1) there are a lot of stakeholders and moving pieces and 2) humans are subject to context switching productivity losses. Machines will absolutely be able to inject themselves into these business processes and chip away at our inefficiencies.
I bet you you're wrong. I quit my $400k+/yr TC job to focus on AI because I believe in this so strongly.
Let's check back in five years.
I'm hiring, btw.
From what I’ve seen so far the ‘impactful’ models take tons of cash to initially train, tons of cash to continue to train and tons of cash to provide “free” access to the public to get the hypetrain chugging along.
Not criticizing but this is what I think people should be worried about, the “future” locked behind a handful of super rich corporate firewalls. Not even mentioning they have real-time data feeds on virtually everything they can pump through their AI for whatever purposes they can dream up. And I’m not even paranoid…
Actually, this I agree with. Chat bots in their current form would help bridge technical knowledge gaps between stakeholders.
Whether it would significantly lean up the less technical staff seems unlikely. Getting rid of engineering is even less likely.
> there are a lot of stakeholders and moving pieces
Yes. There are so many extending all the way out to internet discussions like this one and beyond to consumers, investors, etc... I think all this discussion is necessary to actually making anything of value. Will consulting chat bots instead of hacker news be the future, or is that where we admit there's something pathological going on?
> humans are subject to context switching productivity losses
I don't think letting people figure out what they want is a "productivity loss".
This (in context) is not the «AI wave». Artificial Intelligence is that which automates solution finding normally tasked to a professional. It has to be qualitatively adequate and competitive (which you could take as the real meaning of the "Turing test" - otherwise, "being scammed by con artists" was already a known factor and perspective at the time of the formulation). If any output fitted the definition, a random number generator (or a constant output) would fit as AI.
Go back to Popper (this name just for example): there has to be a metric for "success".
What should happen is teachers and students are required to up their skills to the point that student are required to produce things better than AI. But the likely result will be lots of futile surveillance.
We’re certainly behind where we expected to be, but the future is indeed here!
https://www.theverge.com/2023/1/29/23576422/san-francisco-cr...
> Months later, a Cruise AV “ran over a fire hose that was in use at an active fire scene,” and another Cruise vehicle almost did the same at an active firefighting scene earlier this month. Firefighters say they could only stop the vehicle from running over the hose after “they shattered a front window” of the car.
I don't know about per-mile, but that's worse than a human for sure.
Among them: I don't think I've ever demonstrated more creativity than chatGPT, only equalled it; and while it sure does make mistakes, when I look back at my old code (or even blog posts) I realise I sure did a lot of that too. I'm pretty surprised it's even as capable as it is, given my understanding of how it works and what it's "goals" are (the task "predicting tokens" doesn't seem like it should be able to do this much).
My fear is that the reference class for this is Go, where someone writing an AI for the game thought it was a decade away from beating humans less than two year before AlphaGo beat Lee Sedol: https://www.wired.com/2014/05/the-world-of-computer-go/
It doesn't matter if the thing a submarine does is "swimming", after all.
We are supposed to want the submarine to swim well.
In context: Dijkstra noted that IT discipline had strong regional tints, enabled by some "«malleability»" (over which the particularizing forces are applied). This "malleability" follows very fuzzy definitions of purpose, which he exemplified through a von Neumann that resembled medieval scholastic philosophers and an Alan Turing that proposed (to the judgement of Dijkstra) "irrelevant" perspectives in the same direction, such as "whether submarines swim" (for "whether computers think").
Now: the context instead of this branch of discussion is about those "signs" noted by OP, which I note are overwhelmed by evidence that those signs are doubtful. The point is not in the vague metaphors that Dijkstra found confusing, but in the opposite flat matter that "whether it swims or it "submarines" [as a verb], it has to do it properly". Which is not in Dijkstra, because he was speaking of something else.
If chatgpt can do things you would have given an junior or assistant and it's now faster to just do it yourself it will have an impact and it might raise the bar.
I would neither under or overestimate what it can replace in a few years.
And surprisingly a lot of time and energy is used to teach people old things just because they also need to learn.
It just might be e easier to teach it once to chatgpt or whatever is next, than people.
And at least in my team it definitely feels like I teach every new hire/intern/young person the same old things.
I could also just starting to teach an ai those things.
And this multiplies. If everyone of us is only teaching the same ai everything once, than this is much more effective.
I don't think it has enough context to be a real coder yet, but it has some ability.
Having the token limit unlocked in the future might change a lot.
Most of us haven't.
We are reusing combining etc.
All self driving companies have this bias. If it works in US and Europe, it works everywhere. It’s like the same old saying of “it works on my computer. Looks good. Let’s put it on production”. We all know how the story ends ;)
If the problem of figuring out where to drive isn't even solved yet, how can you claim that "self driving cars work right now". They don't, not for any useful definition of the term.
Unless they send a human out to every building and mark out which door is the correct one the algorithm is just guessing. I use a “professional” GPS for my job and I don’t trust it at all to get me to the correct entrance, I have to study satellite images and type in the coordinates manually and even then it’ll decide to reroute to the wrong place on occasion because it doesn’t know there’s a gate in the way of its optimal path.
Bit of a hassle, really. If you ever see a big truck stuck on some random road with nowhere to turn around it’s probably the GPS’s fault. One of the first things new drivers need to learn is the GPS will actively try to kill you and can’t be trusted.
Self-driving cars only work on some roads specifically adapted for self-driving cars, and even then they require a team of specialists constantly monitoring them (so it's not very self-driving, the driver just moved to another place).
If they require a specially designed roads, they don't replace cars, they replace trams. And trams are much more efficient, so they don't really replace even them.
You can't put a self-driving car on a random road and expect it to work.
I don’t think that is actually a thing. The places self driving cars are being used right now do not have roads that were especially adapted to support self driving cars.
The problems have been well documented, and the part where model makes up info and presents it as fact is a huge problem. Whether it’s solvable is up for debate until it is actually solved.
What this article has shown is that the automation part can work well in somewhat limited circumstances, but in the end it is a form of automation and perhaps a novel user interface options. It is not creativity, it is not something genuinely new. Automation is powerful and it enables productivity goals, which is usually profitable.
It will make 10% 10x more valuable and 90% obsolete.
There will be attempts at improving reliability but at some point it becomes too expensive and ineffective to try to improve it. People that know how to use it will really thrive.
At some point it will be self improving. Either through humans telling it to quit spewing bullshit or by learning fact-checking is a good system of improvement.
Because, by the time AI is good enough to be replacing developer jobs (5-10 years), you will be an experienced engineer who would be negotiating requirements, inputting the prompts, reviewing the output, working on the architecture and process framework within which the AI operates, etc.
There has never been technology that blew up like "chatGPT" before that made a real impact on society. That's why my take is its all a scam. Youtube "chatGPT" you will see how everyone and their dog are making $50,000 a month passive income using GPT. Of course they will tell you all their secrets... as long as you watch commercials attached to their videos.
Chatgpt-backed tools will become the norm in the same way that every knowledge worker uses Google on a daily basis today. Note that I did not say that people will use it to replace search - search is but a small bit of how it can be used.
On one hand those are huge advancements but at the same time just another kind of software that doesn't work properly when I need it to.
Let's not forget that this is iteration 1 of the technology. It is only going to get better and better. The fact that it can already do all these independent tasks without having seen additional data for the task is a complete game changer.
> The fact that it can already do all these independent tasks without having seen additional data for the task is a complete game changer
There are some impressive examples, yes, but ironically they aren't among the ones getting hyped. Right now AI is only usable for one kind of application: where a few incorrect results don't matter at all. That means for me projects like Copilot are out - if I want buggy code I can just write it myself, and at least I understand my own code. ChatGPT quickly falls apart when you actually start testing it for something productive. I don't doubt others have some limited uses for it.
GPT-3 has been available for more than a year and there has been no large scale impact at all.
The few AI generated articles are easy to spot and full of mistakes.
ChatGPT is great at one thing: rambling.
Msft literally just rolled out Teams premium. That's not nothing IMO. We're seeing it start to get integrated in actual products.
Microsoft is jumping on the hype train to make a quick buck out of a cheap API, why shouldn't they? That doesn't mean it actually brings any value.
It's exactly the same as every video game company trying to shoe-horn NFTs into their games.
As a side note, it doesn't look like Microsoft is using GPT for its Teams premium, it's mostly about voice recognition and things like that than content generation...
The output of this software would have been considered absolutely incredible to most of the Computer Science field 10 years ago, and would have been considered sorcery 20 years ago.
Yes, if you're one of the 0.003% of the world who are AI researchers, ChatGPT is probably somewhere on the spectrum between boring and kinda-impressive. To the rest of us, it's astounding.
"Naming a framework can be a crucial aspect of its success. It's important to choose a name that is memorable, catchy, and represents the core values and goals of the framework.
Here are a few suggestions:
Prodigy: A framework that aims to make web development simple and efficient.
Velocity: A fast-performing framework that emphasizes speed and agility.
Eon: A modern and innovative framework that incorporates the latest web development trends and technologies.
Apex: A framework that emphasizes scalability and reliability, helping developers build robust and scalable applications.
Fusion: A framework that seamlessly integrates different technologies, tools, and libraries to create a unified and seamless development experience."
I'm looking forward to hiring a prodigy.js expert with 10 years of prodigy.js experience.
Tried it, it did okay on some high-level stuff but couldn't get to working code: https://imgur.com/a/HModV0f
Do you think it's fair to hold it to the task of instantly creating a new JavaScript framework from scratch and implementing it? This is a task that multiple experts planning and collaborating in different areas of software engineering could perform together with extreme difficulty over a long period.
On the other hand, I was surprised that it came up with working, "correct" code for a difficult task:
Pastebin for the code:
I checked it and it is correct for n up to hundreds of thousands. (I'm not sure if that is really how matrix powers are supposed to work, but overall, its function returns correct exact solutions for the problem and worked without any modification.)
I find both conversations I just linked pretty impressive in their own ways.
Sort of in line with the usual switchbait behavior certain companies are well known for, except at a grander scale: first compel people to share information openly and publicly, then use it against them.
We wouldn’t give Google a free pass if it stopped sending us traffic and started only showing aggregated information on its pages. Heck, there was an outcry when they merely started showing questions before search results, even though they actually credited the website. Somehow we are fine when OpenAI, the for-profit company with antithetical name, and Microsoft, its largest partner with which billions it plays, do the same.
On the off chance there is at some point some case law that builds around this - who cares. IP laws are garbage and problematic and in 99.9% of cases only serve to enrich millionaires and billionaires.
You may allow this by distributing it under a permissive license, or the platform you use to publish it may have terms that give it certain rights to your work, but by default you own the work and can fully control how it is allowed to be used.
Which principle OpenAI is basically throwing out of the window.
People who do the actual work do so because they take pride in it. The work is their child, and they feel valued because people know and appreciate them for doing it. If you believe that most people who do the actual original work do so without any pride or concern for recognition or control over their work, then you are living in a different world than I.
1. I'm just naturally averse to any sort of hype. Maybe something I should talk to a therapist about because it's only rational to some degree. I've been building software for decades, and almost every hype that came up meant more boring/annoying work for me - usually around resolving misconceptions non-technical people get from the news. Rewarded only by the grim satisfaction that people finally get it a few years delayed ("blockchain technology" is my favourite example here).
2. I can only see the bad LLMs can bring: A further decline of quality in a ton of areas, having to deal with worse writing and user interfaces in my private life, and worse code and tools in my professional life. It's just largely dystopian in my mind, I don't see any benefits. Code writing speed is hardly a bottleneck for the kind of work I do, I actually find it to be a small, particularly satisfying part of it. Is it more fun for a violinist to not touch the instrument and instead just tell a machine what to play? Certainly not for me. Generating elaborate wiki pages and SEO spam? Hell no, it's bad enough without LLM assistance.
The strategy I developed over the years is to learn enough about the hyped thing to feel like I don't miss out on information I need to have in case it comes up, and then to largely ignore it. Things are rarely eaten as hot as they are cooked. Sometimes, I start to get into the hyped thing a little later on the hype cycle, having waited for the early adopters to figure out what it's actually valuable for. At that point I usually am not appalled by it anymore. There's nothing wrong with _not_ being an early adopter of something. And it's also not as if you can do anything to stop it, whatever is gonna happen, happens. All you can control is how you react to that, in what you do, as well as emotionally.
I heard this trend is everywhere around at multinational corps in our country.
The CTO is just a guy keeping the balance sheet in order and nodding at what Sales/CFO/CRO says.
Not only devs, but also designers, testers and data analysts.
Most people are very intimidated by a blank page or empty code editor. Most people need a helping hand and someone to help them navigate a problem. Most people have a lot of difficulty producing individual work on their own.
Think of this like flipping the other end of a ping pong table up so you can bounce the ball back. Sure there’s no real person there, your shots are just being reflected back, but without something to bounce against, there’s no way you can play solo. Maybe that’s not you, but it is a lot of people.
These recent AI developments are the discoveries of the decade. Let folks have their hype.
The past six months have had me excited about tech for the first time decades. I'm just happy to be here, experiencing it.
Dropbox/google drive/etc?
Git maybe, although obviously not for all office jobs.
GPS navigation?
Keurigs and instant pots?
I strayed a bit from “office jobs” but yes I have lived through a number of things that save 30 minutes a day for a lot of people.
I get how it could replace certain jobs almost entirely, if those jobs involve churning out a lot of bullshit text. For anything that actually matters, checking/fixing ChatGPT output is going to take roughly as long as doing the research/writing oneself to begin with.
Poster's name of 'sanitycheck' happily checks.
It could help write emails faster, especially ones that require a lot of care. Or summarize meeting notes.
I think it’s actually not a bad thing if there are ‘too many’ posts around a particular piece of tech. If something sticks for a while I usually take it as signal to investigate. Other things have a brief moment of hype but don’t actually stick around too long if you watch the ebb and flow.
Instead of getting upset, consider spending less time on here. :)
I prefer/recommend not to use the "hide" button too much in the front page, but in some case you can use it for a week if the ChatGPT posts get too annoying.
But on the other hand, no. This is actually a significant technological development that has real implications, positive and negative, for our entire society. It's not some BS nonsense like blockchain. This is almost as big as the rise of smartphones. We need to be talking about it.
But I'm worried when it gets to the point where a future ChatGPT is no longer that unreliable, and then millions of people given that hammer will feel justified enough to keep ranting about it and injecting it into every random forum thread and comment section forever just because it's possible for them to.
Yes, I'm tired of it, because it's not an endgame tool but many treat it like it is. And I have a feeling I'm going to get even more tired of it the more it improves - that is, the more people it's able to convince it's no longer bullshit.
But here's the thing, that little anecdote perfectly exemplifies the excitement and novelty surrounding language models like ChatGPT. Sure, some of the answers generated might seem like "smart sounding bullshit," but that's part of the charm! We're still discovering the capabilities of these models, and it's exhilarating to imagine what else they might be able to do in the future.
So let's not get tired of the ChatGPT headlines. Instead, let's continue to engage in discussions and explore the potential of AI technology. The future is full of opportunities, and language models like ChatGPT are going to play a major role in shaping that future. Who knows, maybe one day they'll even be able to write their own hilarious chicken crossing the road jokes!
I would be impressed if the LLM was a fresh never seen blob. These models have been in development and have been massaged by humans for years. It wouldn't surprise me if one of the developers added it to the system as a test case or an easter egg.
You have to keep in mind that these apps have zero intelligence. They are not smart. This is the hype cycle of these products. The AI companies need money, customers and developers so they are trying to get everyone excited about the product.
OpenAI's CEO is talking about how General AI will change everything but I'm sure he knows that LLMs will not get us there. On a scale of 1 to 10 we are at 1, if that. They are impressive but they have no cognition.
That's genuine utility value. The only utility I've ever had using crypto is buying mushrooms over the internet. Now with them decriminalized in my state and easy 1-2-3 spore kits available, I don't even need it for that any more.
There is nothing intelligent about ChatGPT, it's just a statistical trick, no understanding, no unique ideas. It's this years's self driving car.
Speaking of which, I'm been waiting for my self driving car since 1982. At this rate better make it a hearse.
(kind of a plug, I haven't updated the Repo in a while and just run it locally).
Plus, on a different note, I fear this mega hype will eventually reach those that decide in a company, and most of them in my experience do not possess proper technical knowledge, which will trigger them to buy shitty AI-generated/based products because of their obsession to save costs, scale a company and replace people.
The way I see, medium/big companies do already or will personalize this AI for their own domains and package it to the end users which will have to interact not anymore with a human but with a machine that can't yet reason very well.
How often do we complain that "algorithms" deleted an account and all that? I imagine chatGPT will make all this and much worse available on a much larger scale.
This is also why I find these articles pretty annoying.
As an AI language model, I don't have feelings or personal opinions. My purpose is to assist and provide information. If you're feeling overwhelmed by ChatGPT headlines, perhaps take a break and come back later.
Like CNet generating their articles with ChatGPT... from the few times I've read CNet lately, I bet that didn't change their article quality one bit. They were blogspam, they stayed blogspam.
At first, they were all in the format of "I asked chatGPT to write a blog post, here it is" and "I used ChatGPT to talk to my mom so I don't have to"
Then it quickly moved onto people pushing out prototypes "Use ChatGPT to generate pick-up lines" and "ChatGPT as a therapist"
And now, more recently, it's all about taking down ChatGPT. ChatGPT is flawed, ChatGPT can be detected, ChatGPT is a dirty liar, or some other library is better than ChatGPT.
You mean that language models will never get better than ChatGPT? Or that ChatGPT itself won't be improved.
Because I guess the latter could end up being true if openai decides to call whatever user facing tool that GPT-4 powers something other than chatgpt.
I’m sick to death of seeing the concentric circles of HN readers parroting the same critiques of ChatGPT over and over again. “It’s a language model!” they’ll say, and with each iteration there’s a higher chance that the person that writes that has no idea what the implications are. The comments usually end with a hand-wavey explanation as to how ChatGPT having flaws means that it won’t be remotely disruptive. There’s usually some weird elitist / classist vibe to the comments (effectively: “none of the jobs that actually matter will be impacted”). Then there’s some other hand-wavey blah-blah said about creativity. As if the typical HN user, myself included, isn’t a rank and file software developer / manager at Tech Company 1527.
The funniest thing about it is that these comments are all written presumably at least in part because the writer thinks that they’ll provide value. You wouldn’t even need GPT-3 to construct one of these comments. Previous-generation models could do it.
I work in the education sector. I’ve no doubt that ChatGPT et al are going to have a long-term impact on the product that I work on. There are always kids that want to find ways to cheat. Before I get lectured by someone with absolutely no relevant experience: ChatGPT can write wholly believable essays in various genres.
We’ve all been worn down by AI hype for years and years now. We’ve especially been worn down by all of the Elon-bait around FSD. To let that evolve into “be unproductively critical of every AI advancement” is…not original thought.
I’m sick of the recycled BS ChatGPT think-pieces yep. But they’re no different to the comments that put them in the front page, including my comment. They’re also no different to a bunch of BS thinkpiece articles on HN.
It's certainly entertaining, albeit Midjourney is more exciting imo, but there are far more challenging problems in the field of Computer Science and sciences in general than NLP.
“ChatGPT but for whatever” will probably be the most effective way to separate venture capitalists from their money in the near future, much like “crypto for whatever” and “Uber but for whatever” and “an app but for whatever” in the past.
Unrelated, the irony of this post made me chuckle. By posting this, you're adding to the ChatGPT "headlines".
People see it as the second coming of Christ, as the Google killer, as a disruptor of educational systems, etc. when all it is is something that's good about making bullshit look legit.
AI has always followed a cycle of hype and then total lack of interest when people realize it's empty promises. ChatGPT isn't changing that.
personally, I like it and see it as a big win for reality where tech at least from my perspective of an average joe seemed to stagnate for years..
Yes, trends are usually not very "deep" or intellectually enriching.
Thanks to Firefox + uBlock Origin, _the name of the CEO of Twitter and SpaceX_ name has completely vanished from my everyday browsing.
I can do the same for ChatGPT. Or any other thing or topic that swtiches from initial hotness to constant noise.
edit: it is so effective that if even affects my own commenting here. I had to replace its actual name by "he name of the CEO of Twitter and SpaceX" to be able to post my reply.
Just waiting on that bubble to burst.
Some people may be annoyed by it I suppose, same as people get annoyed when they see new JS frameworks/libs.
If you don't notice how much money-momentum is behind it, you're burying your head in the sand. I think it's an interesting time because we are seeing how the market squeezes value out of emerging technology.
It's an amazing tool nonetheless.
Anyone remember that voice conversation app everyone was pushing and I can't even remember its name?
edit: Clubhouse
Yes, ChatGPT is largely overrated. Yet it is still a modern marvel. It WILL change the world.
But most of the stories about it are still garbage and I'm very tired of endless breathless and misleading headlines.
Anyway, Chinese balloons are infinitely more important.
What problem domain do you use it in that you find underwhelming?
The answers were "...it is not scientifically proven" BS.
Something I already proved with my own life, GPT is not willing to answer and risk its good, politically correct name for.
Google is here for stay until people are ready to give up spoon fed BS
Though I'm not sure it's quite as bad as the countless Wordle headlines from a couple years ago.
I have genuinely found it incredible in certain situations. I don't feel like I am getting as much value from it as others might be so for the moment I like the discussion
that's most of the ML/AI community
There are some tropes that need to go away, though. Most of them some form of lame middlebrow dismissal.
- "It doesn't actually understand": nobody outside of a small niche cares how the sausage is made, you are not making a substantive observation that hasn't already been made by 100,000 other people.
- "Most of the answers are bullshit": factually incorrect and the growing sidebar of my chat history with it can prove as much.
- "Okay if not bullshit then so riddled with errors as to not save any time or be useful" see above.
I would suggest that people who have such a hard time with content they are not personally interested in to make use of the hide button rather than spewing hate all over every related thread.
Paul Graham (PG) walks onto the stage, the audience applauds.
PG: Thank you, thank you. Good afternoon everyone, it's great to be here today to talk about AI. But today, I want to focus on a specific type of AI that has taken the world by storm - ChatGPT.
Cut to:
INT. NEWSROOM - DAY
A JOURNALIST is typing away on his computer. He looks up, frustrated.
JOURNALIST: (to himself) Another day, another set of ChatGPT headlines. I mean, what's not to love?
Cut to:
INT. CONFERENCE HALL - DAY
PG: Now, I know what you're all thinking. "Oh no, not another ChatGPT talk." But I want to give you a different perspective.
Cut to:
INT. DINNER PARTY - NIGHT
PG, surrounded by friends and colleagues, is holding court.
PG: You know what, let me tell you a story. I was at a dinner party the other night, and someone brought up ChatGPT.
FRIEND: (excitedly) Oh, I love ChatGPT!
PG: (smiling) Well, I decided to have a little fun and ask the model some absurd questions. And you know what it responded with?
FRIEND: What?
PG: (grinning) "Why did the tomato turn red?" "Because it saw the salad dressing!"
Everyone at the table erupts in laughter.
Cut to:
INT. CALL CENTER - DAY
PG is observing a customer service representative at work.
PG: (to himself) And then there was the time I watched a machine powered by ChatGPT resolve an issue faster and more efficiently than any human representative.
CUSTOMER SERVICE REPRESENTATIVE: (smiling) Well, it looks like ChatGPT has done it again.
PG: (nodding) Indeed.
Cut to:
INT. CONFERENCE HALL - DAY
PG: So, instead of complaining about the constant stream of ChatGPT headlines, why not embrace the excitement and keep pushing the boundaries of what's possible with AI technology? The future is a blank canvas, and it's up to us to paint it with the limitless potential of AI. And who knows, maybe one day, these models will be able to write the next great American novel, or maybe even perform open-heart surgery. The possibilities are truly endless, even if it means cracking a joke or two along the way.
The audience erupts in applause, and PG takes a bow. The scene ends.
So no, more crazy MVPs and open source libraries please
every moderately large news organization has been using A/B testing for over a decade to choose how to present the headline, with sometimes tens or hundred of possibilities being quickly swapped out/tested as people click, training "models" for presenting the most catchy headlines
now we just have a 'confidence percentage' before actually publishing
It literally does not "know" or "understand" what is true or false. It is just capable of imitating true-sounding things (which often happen to be true). That's a massive difference, and it's dangerous that people take what it says as true.
"I asked ChatGPT this question and now its my medium column"
etc etc
It just seems at the moment that most of the content is like that.
We have created intelligence. We aren't alone in the universe anymore. At this stage, it's still an "incredibly well-read moron", but it's still intelligence.
especially because it's based on google tech
Gmail/gsuite android maps YouTube
Thanks to Firefox + uBlock Origin, "Elon Musk" has vanished from my everyday browsing.
I can do the same for ChatGPT. Or any other thing or topic that swtiches from initial hotness to constant noise.
Just vote and move on, hide the submissions if you have to.
I lived through the rise of the web, mobile internet, and the smartphone so I have my doubts about ChatGPT being the biggest technological breakthrough of my life.
Not to mention all of the things outside the realm of computer science competing for that title. Fusion power, solar power, alternatives to antibiotics, cures for diseases that kill millions, etc.
LLMs have a long, long way to go to get anywhere near that. All the predictions about replacing jobs are making dubious assumptions about radical improvements over the current state of the art. Maybe that will happen, maybe it won't.
The question in my mind is: is the curve just linear or is it getting closer to exponential at this point?
https://en.wikipedia.org/wiki/Interpolation
Here is ChatGPT's response fwiw.
Prompt: is the curve of improvement in LLMs linear or more exponential?
ChatGPT: The improvement in language models has generally followed an exponential trend. With the advent of deep learning and the increasing availability of large amounts of data, the performance of language models has improved dramatically in recent years, leading to breakthroughs in various NLP tasks. However, it is important to note that the improvement curve of language models is influenced by multiple factors, including the amount of training data, the size and complexity of the model, and the computational resources available. The improvement curve of language models may also become more linear as the state of the art approaches certain limits.
> The improvement curve of language models may also become more linear as the state of the art approaches certain limits.
This part of ChatGPT's response is the crux of my original question.
It might be notable that I was able to focus and possibly learn from the response to my question via interaction with an LLM, and not via interaction with a human.