Interesting. For what it's worth, I still use it regularly, and can't imagine stopping any time soon. It's a useful and fun tool for all sorts of things
Me too! I've been using it to learn a new programming language (Rust) and the ability to just ask questions that come up as I'm reading the Rust book is amazing. It's like having a personal tutor who is there 100% of the time.
I was troubleshooting an infrastructure issue just yesterday that was not documented properly and it was actually making amazing correlations that I couldn't find explained anywhere. I was pretty impressed.
How are people learning languages when it fails miserably and often in very misleading ways. If i use it for a language i have experience with the nonsense is immediately visible. When i use it in a language i know nothing about it looks plausible. But then the code is incredibly misleadingly wrong.
I'm really surprised people have stopped using it after the "hype." Most hyped things are ten layers of fluff with a grain of "good" in the middle, whereas ChatGPT actually delivers everything the hype claims and maybe more.
It is a HUGE productivity boost for me personally, and I use it for tons of personal and professional things. It may get overtaken but something better (and there is room for that), but I cannot imagine I'd choose to go back to not using an LLM in some capacity ever.
True but those average users might be using chatbots and other services with nice UIs they can understand that are making those API calls behind the scenes. So the average users are indeed using the API, just indirectly.
With so many AI products being launched these days, it could be that the user traffic is still rapidly increasing but just being spread out amongst all the interfaces becoming available.
Third month of school break, yep. I expect a surge this fall.
Neither this drop nor any future surge should be seen as a sign of failure or success. It's already one of the most successful products in history by several measures.
In California the UC system came back in the middle of August. The first few weeks are easy, I concur that demand will climb as the semester progresses.
Well, yes. It is one of the most successful products by this measure or can you name me any other mass consumer product that starts at 20$ and has those giant paying subscription numbers? (no, telco/cable doesn’t count)
Irrelevant if those paying subscription numbers are dwarfed by the money spent on compute.
That said, they just opened an enterprise product offering, with no disclosed pricing (usually this means "if you have to ask you can't afford it"), so maybe they are actually making money now?
they spend I think 21 million 700k per day) a month on compute, they earn 80 million monthly. That's a pretty nice revenue margin, their compute is also discounted since Microsoft owns like 49 percent of the company for 10 billion invested.
Ah yes i was looking for this comment. Without error it gets repeated each time news about openai losing users. Guess we’ll find out soon enough if accurate. But how everyone defending this product repeats the same few things over and over is … interesting.
I don’t think I understand your point. It’s common knowledge that ChatGPT is a tool of choice for high school and college homework. It’s also common knowledge that even ignoring this, ChatGPT is a wildly successful product that’s completely broken into mainstream consciousness.
I have found myself using it less and less because the fun of using it for goofy stuff has worn off and it still hallucinates far too much to be used for anything serious.
I paid for it for a while because I had friend who was gung-ho on it convinced me. I tried using it for coding and it just didn't save me enough time to justify using it (not so much the cost but the effort). I cancelled by subscription.
However, I recently just re-subscribed because I started using it for writing -- and this is where it has been really helpful. I'm not using it for research or to generate anything new -- instead I'm running my own writing and sometimes random thoughts through it to create actual writings.
It's somewhat nice to have to generate say a paragraph and then explain how it didn't quite get the point I was trying to make and have it regenerate a better one. I usually don't take it's content as-is but it's a great motivator and is good at restructuring sentences to make things sound better.
Initially, I subscribed to ChatGPT because an enthusiastic friend recommended it. I tried using it for coding, but it didn’t significantly improve my efficiency. So I cancelled my subscription.
Recently, however, I revisited and re-subscribed, discovering its potential in assisting my writing. Rather than using it for research or novel content generation, I've been refining my own writing and crystallizing fleeting thoughts into more coherent prose.
It's beneficial to input a draft paragraph, receive feedback, and then fine-tune it until it captures my intended message. While I rarely use the content verbatim, ChatGPT serves as a valuable motivator and is good at restructuring sentences for clarity and impact.
... OR ...
I paid for ChatGPT because a friend who really liked it talked me into it. I tried it for coding, but it didn’t save me enough time. It wasn’t about the money, but the effort. So, I cancelled it.
But I signed up again recently. Now I use it for writing, and it helps a lot. I don’t use it for research or making new stuff. I put my own writing and random ideas into it to make them better.
It's nice to give it a paragraph and then fix what it misses. I don’t always use exactly what it gives me, but it helps me change sentences to sound clearer.
Arr, me hearty, once I boarded the ChatGPT ship on the advice of a chirpy shipmate. Thought it'd help me navigate the treacherous seas of code, but nay, it didn't make sailin' faster. So I marooned it—left the subscription high 'n dry.
But listen ye, I boarded 'er again, findin' she be more of a muse than a swab. Aye, not for treasure maps or writin' the next epic of the sea, but for polishin' me own log entries and givin' structure to me scatterbrained musings.
Yarrr, it be like handin' a rough chart to a trusted navigator—ye get it back, sharper and clearer. While I don't take her words as gospel, that ChatGPT wench helps me tack in the right direction. She be a fine first mate for reworkin' yer sentences, bringin' clarity and firepower to yer words. And reckon it be quite the tool for rephrasin' yer thoughts as if ye be a pirate too, savvy?
Version 2 seems harder to process, like it's using words you wouldn't see as often as the original version. Are you asking it to make it "sound smarter" or something?
version 1:
> running my own writing and sometimes random thoughts through it to create actual writings
version 2:
> refining my own writing and crystallizing fleeting thoughts into more coherent prose
I often have to tell it to not be so flowery. But that's part of the process. I can also tell it to use more of my voice. I just keep refining it until it's what I want. Version 2 is not normally how I would keep it.
Sometimes I basically say "I want to X about Y" and spits that out in English and then I can actually figure out what my thoughts actually are.
Sometimes just seeing something phrased differently is enough to find a better way of saying something even if I don't use it all.
... OR ...
I often tell ChatGPT to keep things simple and not too fancy. That’s just a part of tweaking its answers. I also ask it to sound more like me. I keep making changes until I get what I want. Usually, the second try isn't the one I stick with.
Sometimes, I just say things like "I want to talk about X with Y" and let ChatGPT put that into words. This helps me understand my own thoughts better.
Seeing something said in a different way can give me new ideas on how to put things, even if I don’t use ChatGPT's exact words.
Well, you see, I found myself shelling out some cash for this tool for a spell, all thanks to a buddy of mine who was absolutely fervent about it. He managed to talk me into it, and I gave it a shot for my coding work. But let me tell you, it didn't quite pull its weight in terms of saving me time, not to mention the effort I had to put in. So, I waved goodbye to my subscription.
Now, fast forward a bit, I've recently jumped back on the bandwagon. Why, you ask? Well, it turns out that for my writing endeavors, this thing's been a real game-changer. I'm not using it for research or to conjure up new ideas. No, sir. I'm putting my own written pieces and even some off-the-cuff thoughts through it to craft polished, bona fide writings.
It's rather nifty, you know, to toss out a paragraph, then follow it up by pointing out where it missed the mark and watch it whip up a better version. I don't usually take its content as gospel, mind you, but it sure serves as a mighty fine motivator, and it's got a knack for rearranging sentences to give things that extra oomph they need.
Once beguiled by a friend, I subscribed to yon software for coding's sake—found it as fruitless as a bard without wit. Yet, renewed have I the subscription for writing's craft! Though imperfect, it serves as a muse, refining mine sentences as one might polish a tarnished turd. 'Tis an aid, if not an oracle.
I've actually stopped using it except via bing or phind, for writing Claude is way more creative IMHO. gpt4 I think is better for code but I use openai API via code extensions for that.
Agreed, I find Claude much more useful - I can feed in 1/3 to 1/2 of a book (PDF -> text, copy 100k tokens) and have an interactive learning session. That 10x context window is amazing with Claude, and they just announced paid plans similar to OpenAI.
An interesting sidenote, which I think most people already knew but it was cool to see played out in Claude: most non-fiction books have all their important ideas in the first half. I have only run 3-4 books through Claude like this, but so far, I haven't changed my understanding from any of the 2nd half of books.
ChatGPT is absolutely wonderful at the mundane. I tried to get it solve a complex problem and it failed miserably but it did produce all the boilerplate around the problem that was perfectly usable.
I sort of a fear a future where the sheer volume of code has increased dramatically because people are pasting code from AI rather than re-using functions and libraries.
I've already used it to generate SQL code that was too long to do correctly by hand. Usually I find that a sign of a code smell and I should look for an entirely different solution but it worked and I didn't have to write it. Ironically even ChatGPT thought the output was too long and had to be coaxed into giving me the whole thing.
Interesting, i found the exact opposite. Use it for coding a lot, hardly ever use it for writing. I like the process of digging deep into the wishing well of the mind and coming up with original writing from there
Hype cycle [0] progressing as it always does. We're past the peak of inflated expectations and headed for the trough of disillusionment. Give it another 2-3 years and we will see very practical (and likely very niche) applications. I just hope it will be more than just pretending to be a real person while trying to sell people another bunch of useless junk online.
2-3 years? It's useful now. I'm not just saying that as a techie, other people I know are using it on their own. If this is the trough, be prepared for a slope of usefulness that dwarfs the peak of excitement.
> We're past the peak of inflated expectations and headed for the trough of disillusionment.
Many techies here are still in denial and are forced to not trust it as they know it hallucinates for no reason given that it has no capability of transparent reasoning - fundamental to anything that is in deep learning.
This is why many are now realizing that the hype was just a grift to get as much VC capital as possible from investors over this ‘AGI’ narrative.
The ones upset over ChatGPT hype are just worried about their over-leveraged investments into ChatGPT-wrapper companies and are confused as to why this AI snake oil is wearing off.
>There have been numerous criticisms[8][9][10][11] of the hype cycle, prominent among which are that it is not a cycle, that the outcome does not depend on the nature of the technology itself, that it is not scientific in nature, and that it does not reflect changes over time in the speed at which technology develops. Another is that it is limited in its application, as it prioritizes economic considerations in decision-making processes.
>An analysis of Gartner Hype Cycles since 2000[11] shows that few technologies actually travel through an identifiable hype cycle, and that in practice most of the important technologies adopted since 2000 were not identified early in their adoption cycles.
Stable diffusion launched generative art but Midjourney took a lot of market share hype for the tech didn't die down, competition ramped up. We're seeing this with bing, phind, Claude, poe, Bard etc... the hype isn't hype it's mass adoption but people will adopt tools they feel work best for their flow but, large language models are going to be integrated into every facet of society in 2 years.
This could be due to the API integrations (ex: Bing/Copilot) causing people to use those instead of going directly to the openAI website.
Admittedly, the article was behind a paywall, so I am not sure if they address this possibility or not. However, from the bullet point that is free it looks like they just analyzed web traffic via similarweb.
My biggest interest is that it could be used in video games to provide dynamic procedural content. I can imagine doing a hostage negotiation in an RPG and the hostage-taker is run by ChatGPT behind the scenes.
It's definitely a new tool in my belt that I still use daily.
When I have a question, I give it a go in ChatGPT, read the answer, if it sounds reasonable, try it out. Even if it doesn't work, it might "hallucinate in the right direction" and give me some phrases that improve my Google search term.
Most of the times, I can guess how it will be wrong, so I can judge the answer quickly.
Sometimes, even just writing a prompt in plain English helps me understand my problem better and by the time I finish my prompt, I don't even send it as I found out which direction I want to go in.
I also use it for writing, though I usually have to add to my prompt to be concise and casual, as the default answer is longwinded, formal and the "AI taste" is easily detectable.
Unfortunately, I still feel like every week it gets worse, though to be honest, Google feels the same.
> and give me some phrases that improve my Google search term.
To be fair google is so lame these days that indeed even a chat bot as basic as chatgpt can help use google better. It hallucinates but in the salad of words that look cool but dont work it does occasionally mix in useful stuff.
Basically chatgpt is proving everyone that we need a better or new google.
I pay for premium still, but let me tell you this, between the self-hosted versions (that you can run with quant models) and the 'alternatives' that specialize on certain verticals (copilot for code or like claude as a generic alt.) - the competition is getting better day after day
The ones that I can run on my machine, because I have a UI/CLI and a model running actively, regardless of whether I have a connection to the internet and better privacy
I continue to pay for it because it serves as my personal assistant and saves me time that adds up over time. Want to query something obscure quickly before digging deeper? ChatGPT has your back. Want to look up some coding convention that you've forgotten the details of? ChatGPT is ready with a quick example. Want to workshop an outline or expand on an existing outline? ChatGPT is there to help.
The trough of disillusionment is here, but that doesn't mean that there aren't plenty of useful applications to make life a little easier.
How on earth is the code still held back at their 2021 data grab? I'd love to use this more but all the stuff I do is cutting edge and it's just completely lost the plot on stuff like changes in libraries over two years.
They probably need to get into the vocal assistant business. I don't know about Siri but I find Google Assistant very frustrating. A voice to voice interaction with ChatGPT along with context tracking would definitely be great.
Siri is much worse than Google Assistant. I went through a period where I was adding a bunch of smart lights and plugs to my house and test drove both. Siri did fine turning lights on and off, but it was laughably bad at anything else—both in voice recognition and ability to answer questions.
I'm sill subbed, but using it less because of the significant experienced drop of in performance for my usecases. It got so bad I'm considering cancelling.
Interestingly, every interaction I had with it has led me further and further away as it's clearly limited. People treat like it's a mind, while it's more like pachinko with words.
85 comments
[ 1.9 ms ] story [ 217 ms ] threadI'm really surprised people have stopped using it after the "hype." Most hyped things are ten layers of fluff with a grain of "good" in the middle, whereas ChatGPT actually delivers everything the hype claims and maybe more.
It is a HUGE productivity boost for me personally, and I use it for tons of personal and professional things. It may get overtaken but something better (and there is room for that), but I cannot imagine I'd choose to go back to not using an LLM in some capacity ever.
With so many AI products being launched these days, it could be that the user traffic is still rapidly increasing but just being spread out amongst all the interfaces becoming available.
tldr: an IDE/editor, technically a modified version of VScode with AI integration built in
Neither this drop nor any future surge should be seen as a sign of failure or success. It's already one of the most successful products in history by several measures.
That said, they just opened an enterprise product offering, with no disclosed pricing (usually this means "if you have to ask you can't afford it"), so maybe they are actually making money now?
However, I recently just re-subscribed because I started using it for writing -- and this is where it has been really helpful. I'm not using it for research or to generate anything new -- instead I'm running my own writing and sometimes random thoughts through it to create actual writings.
It's somewhat nice to have to generate say a paragraph and then explain how it didn't quite get the point I was trying to make and have it regenerate a better one. I usually don't take it's content as-is but it's a great motivator and is good at restructuring sentences to make things sound better.
Recently, however, I revisited and re-subscribed, discovering its potential in assisting my writing. Rather than using it for research or novel content generation, I've been refining my own writing and crystallizing fleeting thoughts into more coherent prose.
It's beneficial to input a draft paragraph, receive feedback, and then fine-tune it until it captures my intended message. While I rarely use the content verbatim, ChatGPT serves as a valuable motivator and is good at restructuring sentences for clarity and impact.
... OR ...
I paid for ChatGPT because a friend who really liked it talked me into it. I tried it for coding, but it didn’t save me enough time. It wasn’t about the money, but the effort. So, I cancelled it.
But I signed up again recently. Now I use it for writing, and it helps a lot. I don’t use it for research or making new stuff. I put my own writing and random ideas into it to make them better.
It's nice to give it a paragraph and then fix what it misses. I don’t always use exactly what it gives me, but it helps me change sentences to sound clearer.
But listen ye, I boarded 'er again, findin' she be more of a muse than a swab. Aye, not for treasure maps or writin' the next epic of the sea, but for polishin' me own log entries and givin' structure to me scatterbrained musings.
Yarrr, it be like handin' a rough chart to a trusted navigator—ye get it back, sharper and clearer. While I don't take her words as gospel, that ChatGPT wench helps me tack in the right direction. She be a fine first mate for reworkin' yer sentences, bringin' clarity and firepower to yer words. And reckon it be quite the tool for rephrasin' yer thoughts as if ye be a pirate too, savvy?
version 1:
> running my own writing and sometimes random thoughts through it to create actual writings
version 2:
> refining my own writing and crystallizing fleeting thoughts into more coherent prose
Sometimes I basically say "I want to X about Y" and spits that out in English and then I can actually figure out what my thoughts actually are.
Sometimes just seeing something phrased differently is enough to find a better way of saying something even if I don't use it all.
... OR ...
I often tell ChatGPT to keep things simple and not too fancy. That’s just a part of tweaking its answers. I also ask it to sound more like me. I keep making changes until I get what I want. Usually, the second try isn't the one I stick with.
Sometimes, I just say things like "I want to talk about X with Y" and let ChatGPT put that into words. This helps me understand my own thoughts better.
Seeing something said in a different way can give me new ideas on how to put things, even if I don’t use ChatGPT's exact words.
Now, fast forward a bit, I've recently jumped back on the bandwagon. Why, you ask? Well, it turns out that for my writing endeavors, this thing's been a real game-changer. I'm not using it for research or to conjure up new ideas. No, sir. I'm putting my own written pieces and even some off-the-cuff thoughts through it to craft polished, bona fide writings.
It's rather nifty, you know, to toss out a paragraph, then follow it up by pointing out where it missed the mark and watch it whip up a better version. I don't usually take its content as gospel, mind you, but it sure serves as a mighty fine motivator, and it's got a knack for rearranging sentences to give things that extra oomph they need.
An interesting sidenote, which I think most people already knew but it was cool to see played out in Claude: most non-fiction books have all their important ideas in the first half. I have only run 3-4 books through Claude like this, but so far, I haven't changed my understanding from any of the 2nd half of books.
There’s a bunch of stuff that I can’t remember for the life of me if:
* it’s abstracted/generalized by Rails
* it’s a common type
* it’s a pass through type
For example, writing a migration, I may involve all 3 of those
* default database types
* some paused database types that get abstracted for various database compatible
* rails concepts, like references (rather than foreign keys)
A lot of this is frustrating to track down in documentation, but OpenAI does it really well.
I sort of a fear a future where the sheer volume of code has increased dramatically because people are pasting code from AI rather than re-using functions and libraries.
I've already used it to generate SQL code that was too long to do correctly by hand. Usually I find that a sign of a code smell and I should look for an entirely different solution but it worked and I didn't have to write it. Ironically even ChatGPT thought the output was too long and had to be coaxed into giving me the whole thing.
[0] https://en.wikipedia.org/wiki/Gartner_hype_cycle
Many techies here are still in denial and are forced to not trust it as they know it hallucinates for no reason given that it has no capability of transparent reasoning - fundamental to anything that is in deep learning.
This is why many are now realizing that the hype was just a grift to get as much VC capital as possible from investors over this ‘AGI’ narrative.
The ones upset over ChatGPT hype are just worried about their over-leveraged investments into ChatGPT-wrapper companies and are confused as to why this AI snake oil is wearing off.
>There have been numerous criticisms[8][9][10][11] of the hype cycle, prominent among which are that it is not a cycle, that the outcome does not depend on the nature of the technology itself, that it is not scientific in nature, and that it does not reflect changes over time in the speed at which technology develops. Another is that it is limited in its application, as it prioritizes economic considerations in decision-making processes.
>An analysis of Gartner Hype Cycles since 2000[11] shows that few technologies actually travel through an identifiable hype cycle, and that in practice most of the important technologies adopted since 2000 were not identified early in their adoption cycles.
Admittedly, the article was behind a paywall, so I am not sure if they address this possibility or not. However, from the bullet point that is free it looks like they just analyzed web traffic via similarweb.
When I have a question, I give it a go in ChatGPT, read the answer, if it sounds reasonable, try it out. Even if it doesn't work, it might "hallucinate in the right direction" and give me some phrases that improve my Google search term.
Most of the times, I can guess how it will be wrong, so I can judge the answer quickly.
Sometimes, even just writing a prompt in plain English helps me understand my problem better and by the time I finish my prompt, I don't even send it as I found out which direction I want to go in.
I also use it for writing, though I usually have to add to my prompt to be concise and casual, as the default answer is longwinded, formal and the "AI taste" is easily detectable.
Unfortunately, I still feel like every week it gets worse, though to be honest, Google feels the same.
To be fair google is so lame these days that indeed even a chat bot as basic as chatgpt can help use google better. It hallucinates but in the salad of words that look cool but dont work it does occasionally mix in useful stuff.
Basically chatgpt is proving everyone that we need a better or new google.
The trough of disillusionment is here, but that doesn't mean that there aren't plenty of useful applications to make life a little easier.
I don’t have experience with Alexa.