That's essentially where we are going with this.

There is, in fact, a vast amount of talk about that, has been for decades. You might not see a lot of it on HN, but that's mostly because its off-topic.

https://www.nvidia.com/en-us/geforce/news/may-2021-rtx-dlss-...

It's subtle in a game made in 2016 before anyone in the mainstream was talking about this sort of thing, but imagine the possibilities for an indie developer without much funding. They can make their textures at a resolution that's reasonable for their budget, but that would look bad scaled up with traditional methods. They can offer higher resolutions at AAA quality without an AAA performance tuning and texturing budget.

Also I think that once GPT will understand images the designers and front-end devs jobs will start evaporating.

Google -> Stack Overflow

With...

chatGPT --maybe--> Google --> Stack Overflow

I won't make predictions about the future, but it's been pretty transitive for today.

ChatGPT is great for doing some of the heavy lifting. I pasted in a list of raw, unprocessed data and asked it to give me unique values sorted alphabetically. I know I could have done it in code, but just being able to paste it in, give a plain English command and get the same results was fantastic.

Although hmm, on the other hand, it's also arguable that what we may have have coded would be subject to errors too.

A bit ironically, if it is regarding a fact, I perform a search on DDG or google and find an authoritative website or paper to corroborate it. So it still feels like search outside of an LLM is an important tool. If the result is code I can test it myself.

I've had a few times where I didn't like the library that it used to come to the results, I just told it 'use library xyz' and it would rewrite the whole thing using idioms native to that library. It's amazing. It's eliminated like 75% of the drudgery that I've come to hate so much about programming the last years.

I think beginning of a revolution. The ground has shifted under our feet.

It'll be really interesting to see how resistent it is to these problems.

Once upon a time Google Search was a great product.

I believe that gives an answer. But the right answer? Even Wikipedia doesn't guarantee that.

dunno about blockchains, but i'm sure there isn't 100million VR units sold. Plus you don't need to pay upfront to use chatGPT to try it. And everyone who has tried it has been amazed, or at least, impressed at how good the responses are.

Not saying that the language models and AI hype is to be believed, but there's substance there. If you are an investor, you should be wary of missing out. And if you're not an investor, you should be wary of losing your human advantage in the economy.

This is almost the definition of a novelty effect. Consider, for example, 3D.

Stereoscopic 3D has gone through at least 5 waves of people trying it, saying, "Amazing! This will change everything!" But so far every time everything has not been changed. (The 5 waves I count: Brewster stereoscopes in the 1850s, 3D still viewers in the 1940s, 3D movies in the 1950s, VR in the 1990s, and 3D movies/TV starting in 2009. Assuming that we see the current wave of VR/metaverse hype as still open, of course.)

The question for me is what happens once something gets past "amazing". For example, consider Shazam's music recognition technology. It was widely considered amazing at the time. But in very short order it became boring, as it had basically one use case, one that did not matter much to most people.

I rented the Quest for a couple weeks at Christmastime. The first week everybody thought it was amazing, me included. Things tapered off in the second week, with the kids back on their Switches and the Playstation. When I mailed it back nobody even noticed.

I hear a fair number of anecdotes like that. And I find very few people for whom VR games are a daily driver the way, say, mobile games are.

ChatGPT is being used by common people in practical ways, not just going "wow." It's already proven itself.

So: We are already there, all without ChatGPT... ChatGPT (with its cool ways to make it nondeterministic, like: replacing tokens with synonyms and raising the temperature. Or: retraining it...) will make things worse definitely.

Probably the way to go is for {academia,sects,orgs with sense for humanity,...} to build stores of attributable information and then start shunning the branches which produce bullshit (this is what scientific publishing should do already.... it doesn't... even without ChatGPT there's already templated articles strewn around the corpus and noone cares). The main issue there is, that this means that they will also have to stop using a lot of technology (and buy a 32nm foundry...), because many orgs are probably already poisoned.

This was the first result: https://applications.messergroup.com/co2-applications

There were a number of other relevant results. Papers on various applications of CO2, forum posts, etc. Nothing on emissions.

Sure google is a lot worse these days than it used to be, but it still seems plenty usable to me and I have no issues finding what I want with some basic operators.

The problem in my imo is that Google was good enough that basic operators were not required, as such the avg Joe is amazed when finding out that they exist, and Google doesn’t really promote the advanced search page these days.

Sure this talks a lot about emissions, but it also seems to include a detailed breakdown of various industrial uses for CO2, with numbers, projections, statistics and all.

First result for “co2 industrial usage numbers”, without quotes - just like the sibling comment.

You need to specify atleast a few keywords to clarify what you’re looking for - search engine nor llm chatbot nor HN commenters can read your mind…

It's one of the most common patterns I come across. I search word X but find a bunch of word Y stuff that's not what I'm looking for, because word X and Y appear together a lot in Y's context. So I add -Y to the query.

It's become a tool that basically anyone can at least get some information out of. Googling used to be a skill that non-technical people struggled with.

In other words, it's more of a social than a technical issue.

And without commercial buy in, I can’t see why anyone apart from computer geeks and scammers would want to use it. The first question your average users would ask is “why isn’t $SHOP on there”.

Much as I dislike the modern web, it’s what the people have chosen in exchange for free services (like email) and convenience (like online shopping).

And to repeat my earlier point, an EULA isn’t going to keep your vision clean. Malicious actors aren’t going to give a hoot about any EULA. And any business that might take an EULA seriously would simply choose not to use such a minority platform anyway.

Also if this “EULA” is targeting how companies use your platform, then it’s not an EULA. What you’re talking about is a different type of contractual agreement.

I'm not sure more, better rule are going to stop a rule engine from participating. If anything, formalized rules merely makes it easier to train the machine learning system du jour to follow them.

My modest proposal is to go the other direction - fill the entire internet with garbage data so that the generative models trained on it become useless. Clearly there's nothing that could go wrong with this plan.

The masses are still on mainstream platforms, but more and more people are nauseated by the instagrams of this world.

Sooner or later some group will realize it becomes necessary to dissolve the bands which have connected them with another, and you'll have a formal split.

In short, I think people are wired to do better in smaller groups and I welcome this balkanization, albeit with the caveat that the communities can interconnect and grow/shrink/die/emerge organically to suit the needs of the people within them instead of commercial investors.

But on the other hand it's hard to deny that the mainstream internet is getting worse and worse. I grew up with the internet as my nanny, but I'm moving closer and closer to the idea that smaller groups focused on specific interests really do provide better experiences.

Perhaps it was never meant to be, and the sweet spot is just living your life completely offline and using the Internet for Wikipedia and Whatsapp.

[0]: https://www.goodreads.com/quotes/7444411-it-s-like-in-the-gr...

This sounds like a legit use for ChatGPT :-)

Like how facebook or github account depend on your phone number, or free trials depend on your credit card number.

blockchain?

all that pointless CO2 for nothing

ChatGPT is a symptom that perhaps finally raises discussion of the root problem above the noise. Yeah, ironic.

It’s speculated libgen is in there as well.

It was an interesting time, because (from memory) 2000 was still "It's pretty complicated for average person to create and post content on the web" and transitioned to 2010's "There are highly polished, easy to use, free SaaS solutions that a trained dog could use." With mobile versions of same in the next few years. (All dates mass market penetration, not first release)

There was a direct correlation between ease-of-publishing and less intellectual discourse on the web.

"Info decay" was definitely a good way to put it. It wasn't that people were necessarily dumb or angry, but that there was a lot more honestly incorrect information online.

Before, there were crackpots, but it was pretty easy to tell they were crazy.

Having seen both, the banal wrongness is far worse.

I think that's one reason wikipedia really took off ~2005+. People were craving an easier way to get even roughly-right information.

I would argue this is a problem stemming from the way "AI" is presented rather than anything to do with "AI" itself.

That is to say, fucking nobody trusts autocorrect to be correct and "AI" is really just a more complex form of autocorrect.

But people hear "AI" and think AI as in Transformers, Terminator, R2-D2, etc. They're intelligent, right? And that's where the presentation is doing significant harm. There is nothing intelligent, let alone sentient, about "AI".

Many of my coworkers feel the same. And this started way before GPT was conceived.

Besides, your post somehow feels like generic technology-angst to me. It seems like whenever we invent something new, there are those people that predict this is finally our demise...

GGPT is a tool, like every other technology. Use it wisely, and you will likely benefit. Use it without engaging your brain, and you will like see negative effects.

As for ChatGPT, I'm actually using it and it enhances my productivity by a lot. So, I believe this isn't just hype.

While Blockchain only has hype, AI has hype and real applications. While there are enogh bottom-feeder AI bros on Twitter and Discord, AI does have real applications and it solves real problems and makes people real money.

You need to shift your focus from the AGI and LLM hype to real application areas.

I have shipped end-user facing real world AI product myself. It saved each client (non tech companies) 2-3mn USD per year, and made our company money.

I can say with authority that if you think AI and blockchain are the same, you are plain WRONG.

That's why I wrote non-tech companies in parenthesis.

For a tech company, you can sell codegen software where the customers solely determine value.

You can sell another tech company Blockchain Snake Oil and make money off of it.

My work was not such a case. End users, consumers saw real value which made their lives better in a small way, and help save our coustomer, a business- save millions of dollars.

I really care about my work, and rejected many offers where I could have made more money working on snake oil recipe.

> I can say with authority that if you think AI and blockchain are the same, you are plain WRONG.

Is attacking a straw man that you built all by yourself.

But as with both of those and plenty of other novel technology, whether or not I like it doesn't answer most of my questions, which are about what other people will use things for. E.g., I did not like YouTube, but I am just not a video person. So me using YouTube and saying "meh" wasn't informative, because I wasn't the target audience. Similarly, I was wowed by the modern VR experience, but being wowed didn't make me a long-term user, something apparently true for a lot of people.

And that goes double for anything that is supposed to be meaningful human connection. E.g., after my mom died, I got a number of nice notes of condolence from people. I appreciated them. If any of them had felt machine-generated, that person would be dead to me.

So my general feeling about the composition case is that it is going to vastly enable the generation of documents that are, to one degree or another, bullshit. And I don't think that's a net positive for society. It's not like we had a bullshit shortage previously.

And this is a straw man:

> Pulitzer prize email that evokes monumental emotions

There's nothing wrong with a plain and simple email. Indeed, I think that's generally better unless the writer finds it worth the time to write something fancier.

- Customer reviews on sites like Amazon can be gamed with significantly better accuracy, at significantly higher numbers. Presumably there will be a period where this is successful for marketers, and then eventually people will come to wholly distrust customer reviews altogether. This will also likely be the death of the "reddit" trick, where you can just query on reddit for real, live customer experiences with products.

- Significantly increased ads on anything like a social network, where a "user" might be talking about a product.

- Generally cheaper ability to write advertisements in a broad category of platforms.

- Sponsored misinformation (government, corporate, etc.) at scale.

To be clear, I think that ChatGPT is going to be a disaster for the internet.

The recent South Park episode where the boys are using ChatGPT to write the perfect romantic replies to their girlfriends really opened my eyes to how impactful this is going to be, not just business copywriting/education essays/etc but even for social interactions like dating (Tinder is going to get interesting).

It's like an autocomplete for your day-to-day life.

Its going to be integrated in less than 12 months

https://www.reuters.com/technology/chatgpt-sets-record-faste...

First, that just means ChatGPT had 100m people enter their email into the registration page. Second, none of those 100m users had to invest any money, unlike VR or blockchain.

But it's definitely not going to go nearly as far as people are predicting. The more i use it the more it becomes apparent to me that there's no real intelligence behind it, just large-scale pattern recognition that makes it seem real because I'm not intelligent enough to comprehend the patterns.

Does it even need to be to be useful?

A ChatGPT that is legitimately "intelligent" would basically be the El Dorado of tech. There's plenty of value between here and there.

The big issue IMO is people are going to Chatgpt's website explicitly and having certain high expectations. Or they use it via Bing expecting it to be a search engine answer machine. Once such an interface is more directly integrated into our lives in a clean way where people understand what it is and what the limitations are it could easily be an app that basically everyone has installed, but more like a super-useful tool like Google Maps rather than replacing Google. That's still a big business and major development for the industry.

And even if the consumer daily-used mobile app thing or some high-visibility Google integration turns out to be less interesting/useful than predicted, there's still a million niche business and hobby usecases.

> training it off of Reddit posts and then making a surprise pikachu face when it turns out to be an arrogant cunt

The hype is real.

But continuing with your analogy, I'll note that the people who fight financial crimes and fraud have absolutely had to integrate responses to cryptocurrencies. That didn't prove that there was lasting value to the cryptocurrency hype wave.

I will note that you didn't use it for most of the coding. And that looking at your code, you also seem perfectly capable of figuring out the AppleScript if you had wanted to. In an amount of time that would not have wildly increased the total project time.

So overall, this looks very "stone soup" to me. Did it get you over a hump? Surely. Did it actually contribute much in terms of code? Not from what I see in your script.

I do support work myself, and our developers have a function that is broken and shoving too much stuff into the database. This crap is compiled into a DLL so I can't do anything about directly. Went to GPT and said "hey, I have a table named X, when Y updates write a trigger that deletes this"

In a few seconds it dumps out a working trigger. Now, I've never written one of these before, dong SQL stuff isn't really my forte, but with a basic understanding of what I needed done I got working code with an explanation faster than I could parse the how to.

Even if AI doesn't contribute more than that, for lots of use cases it's more than enough.

This sort of reminds me of the 3d-printing hype cycle. People were very excited! We could make any object! We could soon be 3d-printing 3d printers! But in practice, 3d printing is now mostly useful for small, marginally valuable stuff. That's not nothing; I have friends who are excited about their 3d printers. But I also have friends who are excited about woodworking and accordion playing and falconry. But none of these are particularly world-changing.

My interest in it is this: given this (flawed) technology exists, what are the things we can use it for that are genuinely useful and help us build interesting things that we may not have built before.

I've been finding an increasing number of areas where it gives me a 10-30% productivity boost... and a few areas like the above one where that tips me from "not going to do that project" to "doing that project". Those are what's exciting to me, far away from the breathless hype around this stuff.

So in this case the leg-up that GPT gave me was the difference between the project existing and it not existing at all.

It would suggest an incorrect type as an input to a constructor and if you ask it to correct the error, it simply suggests another conceptually similar type to fit in, which also throws an error upon compilation.

A modified chatgpt that was bound under the hood, to a strict set of rules about the structure of c++ and it's relationships, could do much better. The bot gives a strong intuitive sense that we have only solved half or a third of a potential revolution in programming, with more to come.

You should start believing people are using this.

People are comparing LLM hype to crypto and VR hype and that’s really not accurate. Those are both always in a state of “someone is using it now, and a ton of people will use it in the future”. ChatGPT is in a state of “a ton of people are using it now”.

It would remind me of the explosion of social networking, except it doesn’t have network effects! It’s crazy how much it’s getting used by all sorts of people not because they want to interact with their friends (messaging apps) but just because they want to use it.

I was a big AI and even just generally ML doubter until I started using it as part of my workflow and reading about how other people are, too. At that point, you kinda have to update your priors.

To my shock, She started explaining to me how awesome it is and how all of her teacher friends are using it to offload a lot of the boring work like writing comments and stuff for students report cards. They are required to write numerous paragraphs for each student several times.

Something in the past I would take my wife an entire week each time.

These are people that need help setting up Wi-Fi on a laptop. And they are in love with ChatGPT.

She manages 80 students.

Alternate would be school hire a lot more staff.

If I'm understanding right, your wife is coming up with some bullet point style feedback about each of her students as part of a recurring evaluation, then using ChatGPT to generate lengthy prose from those bullet points.

Hailing ChatGPT as a savior here feels something like the tail wagging the dog. Why not just change the nature of the task so that what's required is to provide the quick bullet points, instead of using language model prose generation as a go-between for communication from teacher to student? What value is being added by the middle ChatGPT layer? More importantly, what value was being added by the 36 hours of marginal effort previously spent doing this manually?

That's one of my main hang ups with all the proposed use cases of ChatGPT - it seems we've created arbitrary problems and now have this complicated tool to solve them. Which in turn will enable us to keep making bigger and more arbitrary problems. But the problems don't need to exist in the first place

Reducing internal pushback to customer hostile bureaucratic requirements imposed by management on employees is a harmful effect of ChatGPT, not a benefit.

I agree with what you say. However she is not willing to a fight worldwide bureaucracy.

Employee pushback isn’t just individual.

What were they doing before ChatGPT?

Teachers tend to work very long hours because of having to grade work, write these kinds of reports, etc, all in their "non-teaching" time.

My entire family fully understands it's just a tool, you need to check it's output, and it's not perfect but it's saved them a bunch of time and allows them to focus on what they are best at by getting a good starting point from *GPT.

I'm not fully convinced it was a net-positive, because sometimes asking the "dumb" question shows that other people in class had that question but didn't want to ask, or didn't know they had that question until someone else voiced it.

The simple fact that people are predicting that GPT will somehow change the world makes me doubt its actual value.

I think ChatGPT is worth doubting, but I think your examples need some work.

I pay for a premium OpenAI ChatGPT account, but except for explaining what code in unfamiliar programming languages does (including libraries it is calling), and occasionally having it write an email for me I don't get a lot of value out of it.

I do get a lot of value from the OpenAI APIs (as well as the offerings from Hugging Face).