Given that Jasper and others laid off a bunch of people, I wouldn't necessarily be using those as examples. GPT is simply too good and generalizable that specialized companies that just add prompts aren't worth paying for.
Jasper was too early, right on time, and too late all at the same time:
- They were "too early" in that what they were squeezing out of GPT-3 was what 3.5/4 could enable from day 1 with ease...
- But they were right on time because 3.5/4 would have made differentiating their AI copywriting tool almost impossible, meanwhile GPT-2 was too finicky to scale like they managed to scale GPT-3 did.
- And then they were "too late" in that they didn't get enough time to catch up on their massive valuation. 1 year ago their data and machinery around scaling GPT-3 in production alone were a moat: today you can build a synthetic data set for world class copy writing for $100 and fine tune 3.5-turbo on it and build something Jasper quality or better as a weekend project.
I think there is room for specialized companies that "just add prompts" as long as the UX is not a literal chatbot. But it's not going to support going from launch to unicorn in 18 months like Jasper did, it'll be much more of a slog.
I submitted this post from Twitter about how there's no moat in AI startups that's basically confirming what you're saying, I'd recommend giving it a read: https://news.ycombinator.com/item?id=36761643
Jasper.ai (formerly Jarvis) was a YC startup that launched and reached a billion dollar valuation 18 months later.
They were using AI to do copywriting/SEO, and most importantly they were doing this before ChatGPT came out.
GPT-3 wasn't approachable for lay people, and relatively few people were building on it: So a lot of marketing people were seeing generative AI for the first time through this single tool. And obviously it was earth shattering for them.
-
The problem is ChatGPT made outputting what Jasper does extremely trivial: I hang around random SEO FB groups and at least once a week someone asks if Jasper is worth it, and a bunch of people chime in saying no ChatGPT output is just as good.
ChatGPT is also incredibly easy to use: a grandma could use ChatGPT to write an SEO friendly recipe post.
That all leaves Jasper an awkward spot. I don't think they'll die, but to stay competitive they'll need to shed a ton of dead weight. They obviously can't keep all the machinery they built around GPT-3 running if it's no longer adding value, they now have to work 100x harder on positioning, etc.
But, that is an argument that another AI-Winter is not coming. If a company is confident enough to lay people off, and GPT is that good, then it is not the end of GPT.
You are saying "GPT is Simply too good" (which I agree with).
But earlier today, entire thread that "AI isn't good enough"
It was mostly an example to highlight that the graph was a little misleading. By the same token, ChatGPT is an extension of what OpenAI has been iterating.
Threads is just an instagram feature, comparing it's speed of market adoption to netflix and OpenAI is pretty meaningless, especially considering the drop-off and lack of notable innovation.
Generative AI burst onto the scene like a summer blockbuster, promising transformative experiences. The tech-savvy among us were both awed and slightly terrified at the notion of AI creating human-like content. But has it lived up to its hype?
The Highs of Generative AI
[Graph 1: Rapid Growth of Generative AI Adoption]
The above graph showcases the swift ascent of generative AI's adoption from 2019 to 2022. With businesses and individuals alike eager to experiment with this new tech, it was clear that there was significant promise.
From creating music, poetry, and even assisting with scientific research, the capacities seemed endless. Remember the poem generative AI penned last year that went viral?
Amidst the silicon dreams we tread,
Seeking truth in binary, it said.
Of ones, of zeros, of neural might,
The machine beholds the human sight.
A masterpiece, isn't it?
But Wait, There's More (or Less)
However, for every positive leap, there have been equally notable stumbles.
[Graph 2: Drop in User Satisfaction with Generative AI Tools]
The above graph indicates that while initial excitement was high, satisfaction dwindled as users confronted the reality of generative AI's capabilities. It wasn't the magic bullet solution for every challenge, and it often still required human input to achieve the best results.
Remember when Elon Musk commented on generative AI? "It's like trying to use a hammer for everything – sure, you might get some nails in, but you're also likely to break a lot of things." While it's not often I find myself nodding in agreement with Musk (especially after that whole debacle with him trying to launch a line of AI-generated perfumes), he did have a point here.
Not a Nothing-Burger, Though
But let's not be too hasty. Generative AI may not have completely revolutionized our world overnight, but it has added value. It's facilitated easier workflows, added creativity in previously unimaginative sectors, and has certainly begun conversations on the potential and limits of technology.
[Graph 3: Industries Benefitting Most from Generative AI]
There are sectors, as highlighted above, where generative AI has truly shined. Content generation, arts, and even certain branches of research have seen remarkable strides thanks to this tech.
Concluding Morsels
Generative AI might not have been the pie in the sky many hoped for, but it also isn't mere table scraps. Its potential is still unfolding, and as with all technologies, it will continue to evolve.
So, whether you're a staunch supporter, a skeptic, or just someone who enjoys AI-generated poetry, there's no denying that generative AI has spiced up our tech menu. And who knows? Maybe the next big breakthrough is just around the corner.
Until then, remember:
In a world of hype and binary frost,
Some dreams are gained, and some are lost.
But in the dance of code and byte,
Lies a future, ever so bright.
Stay curious,
[Your Substack Author Name]
Note: All graphs referenced in the article are conceptual and would typically be visualized with data. Given the format, they've been described rather than shown.
One quote I took issue with, when discussing the Axios article,[1]:
>I could first point out that it is not often 69% of businesses adopt any technology simultaneously.
AI has been a buzzword talked about by executives for at least ten years now. The same issues with data organization and labelling were being explained to executives 5 years back [2] and are still relevant today.
LLMs are a jump forward in NLP which enables more use cases in business, but AI and its adoption challenges are not new.
We're barely scratching the surface of the space of applications of (very) LLMs and they're not going away anytime soon. The commercial world simply moves too slow for any clues about what's yet to come.
True, but there could still be an overinflated bubble that collapses in the meantime. In fact that's exactly what happened with the web: it sure was revolutionary (and we still haven't reached its full potential IMHO) but that fact didn't prevent the dotcom bubble burst.
The Web also has had a lot of downsides that weren’t clear at the time, particularly when paired with smartphones and Skinner-style behavior reinforcement.
The early web's killer app was email and instant messaging, which were orders of magnitude cheaper and faster than snail mail and long distance phone calls. That's why adoption was fast.
The current usefulness of consumer AI pales in comparison.
After two or three development cycles, LLMs will still be a curiosity, but nothing more. They might be used for some non critical business cases where output doesn’t need to be perfect, but for all other cases the burden of verification will be unmanageable.
The use of LLMs in academia will be nearly eliminated as professors opt for more verifiable ways of ensuring students aren’t using ChatGPT: no more writing long form essays in the comfort of your own home and with access to the internet.
This article was an excellent counter to the anti-hype-hype.
Seems like the AI hype was so great, that now there is an anti-AI hype saying AI was just another over hyped tech that is fading, and this wave of hype is turning into another AI-Winter.
This article is great at pointing out that AI is really still growing, just now in the 'building' stage. All the corporations are now off building things with AI, but in their own gardens, and not publicly hyping them.
The slower growth state of the typical hype cycle, when the slow adopters are doing their adopting.
6 months ago it looked like AI / LLMs were going to bring a much needed revival to the venture startup ecosystem after a few tough years.
With companies like Jasper starting to slow down, it’s looking like this may not be the case.
Right now there are 2 clear winners, a handful of losers, and a small group of moonshots that seem promising.
Let’s start with the losers.
Companies like Jasper and the VCs that back them are the biggest losers right now. Jasper raised >$100M at a 10-figure valuation for what is essentially a generic, thin wrapper around OpenAI. Their UX and brand are good, but not great, and competition from companies building differentiated products specifically for high-value niches are making it very hard to grow with such a generic product. I’m not sure how this pans out but VC’s will likely lose their money.
The other category of losers are the VC-backed teams building at the application layer that raised $250K-25M in Dec - March on the back of the chatbot craze with the expectation that they would be able to sell to later-stage and enterprise companies. These startups typically have products that are more focused than something very generic like Jasper, but still don't have a real technology moat; the products are easy to copy.
Executives at enterprise companies are excited about AI, and have been vocal about this from the beginning. This led a lot of founders and VC's to believe these companies would make good first customers. What the startups building for these companies failed to realize is just how aligned and savvy executives and the engineers they manage would be at quickly getting AI into production using open-source tools. An engineering leader would rather spin up their own @LangChainAI and @trychroma infrastructure for free and build tech themselves than buy something from a new, unproven startup (and maybe pick up a promotion along the way).
In short, large companies are opting to write their own AI success stories rather than being a part of the growth metrics a new AI startup needs to raise their next round.
(This is part of an ongoing shift in the way technology is adopted; I'll discuss this in a post next week.)
This brings us to our first group of winners — established companies and market incumbents. Most of them had little trouble adding AI into their products or hacking together some sort of "chat-your-docs" application internally for employee use. This came as a surprise to me. Most of these companies seemed to be asleep at the wheel for years. They somehow woke up and have been able to successfully navigate the LLM craze with ample dexterity.
There are two causes for this:
1. Getting AI right is a life or death proposition for many of these companies and their executives; failure here would mean a slow death over the next several years. They can't risk putting their future in the hands of a new startup that could fail and would rather lead projects internally to make absolutely sure things go as intended.
2. There is a certain amount of kick-ass wafting through halls of the C-Suite right now. Ambitious projects are being green-lit and supported in ways they weren't a few years ago. I think we owe this in part to @elonmusk reminding us of what is possible when a small group of smart people are highly motivated to get things done. Reduce red-tape, increase personal responsibility, and watch the magic happen.
Our second group of winners live on the opposite side of this spectrum; indie devs and solopr...
I've been getting really frustrated reading peoples comments about how useless openAI is on twitter/reddit/whateversocialmedia. Chatgpt saves me probably tens of hours a week on various coding/math, whatever tasks. I use it for sed and regex constantly. Then one of these posts pops up on reddit and it's one of those "it'll never take our jerbs it's only good at writing bad poetry," posts look at all the funny things I make it do!
The dissonance is wild. I'd hate to go back to working without it. It definitely has its uses.
I think you've missed the point I'm trying to make. I'm not entertaining their opinions. I'm annoyed that they don't see the utility at all and think it's such a gimmick. It's incredibly helpful to me day to day.
They act like it's literally useless except for plagiarism.
It's the fault of the manufacturer, but that is a distinction without a difference. OpenAI has marketed it as an "everything" tool, whereas McMaster-Carr have been more conservative about the capabilities of theirs.
An electric drill's capabilities are known, in the way AI's is not. There's no drill bit coming in Q4 that will let you drill into steel for example. But there will be a GPT 4.5 or something that will add new capabilities.
After so many disappointments in the tech universe (and with some AI winter in the past) people in the industry should not be naive to the point of thinking that OpenAI would solve all the industry's problems.
>> OpenAI has marketed it as an "everything" tool
That's what all companies do all the time. Should we blame capitalism?
A tool every person uses daily can be worth anything from hundreds of millions to trillions of dollars. The issue, as always with AI, is overselling it. Pre-Revenue startups raising at billion dollar valuations is certainly not going to end well. Hence anything that looks like it will prick the bubble is worth talking about to a certain segment.
On the other hand, the finances of AI are only loosely correlated with it's utility to individuals. If OpenAI went from being worth 10 Billion to 500 MM, it wouldn't make too much difference to people who use OpenAI products.
So lately I've been using it to build things against APIs I've never worked with. I needed to write something that would check circleci (ci/cd system) for jobs in ANOTHER repo running and hold off starting a job in a DIFFERENT repo if there was one running in the first. Basically something to check every 30s-1min and succeed and move on to the next job if that repo wasn't running a workflow. Instead of having to dig through circles API docs or stackoverflow for examples I just typed in something like "How can I use ruby, go, bash or python to check a different circleci repo for running workflows before starting a workflow in a different repo?"
I had to go through several iterations and I bounced from wanting to write it in go to I think eventually just doing it in bash, just simplicities sake, but it absolutely saved me days of research. I definitely had to fix a lot of things and tailor it to my needs and it lead me astray quite a few times but it really was more like asking a professor the question vs googling it and hoping that SOMEONE out there has done something similar. Or combing API docs.. It was such an ancillary thing that I didn't want to get hung up on it for more than 1-3 days and it literally got me going in maybe 3-6 hours.
I also write a ton of very advanced terraform and cdk with it. I already know both very well but it absolutely speeds me up if I haven't used a specific function or what not in awhile, or it's newer to me.
It's great at leading me to the water, sometimes I have to still teach myself to fish but it will give me ideas and get me going if not even more.
For me ChatGPT (v4) is very useful as it is and for my line of work is a God sent, and I give one simple real world example.
Earlier this month we have some encouraging results on earthquakes early detection based on IoT and Big Data. Our earthquake expert is not replying emails for several weeks and we had to send the research summary immediately to ask the journal editor for the suitability of potential publication based on the proposed paper's title and abstract. But since we are not the earthquake domain experts, we are not sure what's the correct word to use for the early detection of earthquake in our particular results, and the prime candidates are forecasting or prediction. Voila, upon consulting ChatGPT it's obvious that for our results, prediction is the correct word to be used.
Before Internet people used to go to library for looking up something, but then Google arrived. Similarly, before ChatGPT people used to these so called experts for consultations, but then ChatGPT arrived. But somehow, people seems to be in denial of the useful of the GPT in helping us to write better reports, papers, articles, etc.
I think a problem with AI is that it is used so much by people that don't really know a field. Like middle management. I heard those fools claiming that a complete IT project took a couple of minutes to do with ChatGPT. The same crazy people that said 5 years ago that all devs will be replaced by low-code by some years ago. Incompetent people will use AI the most. That is a risk.
There is no death of ChatGPT in sight. Even if the UX was a bit of a hoax to begin with. It would be if some of the other models beat it.
45 comments
[ 3.0 ms ] story [ 115 ms ] threadhttps://twitter.com/levelsio/status/1680579916398567426
- They were "too early" in that what they were squeezing out of GPT-3 was what 3.5/4 could enable from day 1 with ease...
- But they were right on time because 3.5/4 would have made differentiating their AI copywriting tool almost impossible, meanwhile GPT-2 was too finicky to scale like they managed to scale GPT-3 did.
- And then they were "too late" in that they didn't get enough time to catch up on their massive valuation. 1 year ago their data and machinery around scaling GPT-3 in production alone were a moat: today you can build a synthetic data set for world class copy writing for $100 and fine tune 3.5-turbo on it and build something Jasper quality or better as a weekend project.
I think there is room for specialized companies that "just add prompts" as long as the UX is not a literal chatbot. But it's not going to support going from launch to unicorn in 18 months like Jasper did, it'll be much more of a slog.
They were using AI to do copywriting/SEO, and most importantly they were doing this before ChatGPT came out.
GPT-3 wasn't approachable for lay people, and relatively few people were building on it: So a lot of marketing people were seeing generative AI for the first time through this single tool. And obviously it was earth shattering for them.
-
The problem is ChatGPT made outputting what Jasper does extremely trivial: I hang around random SEO FB groups and at least once a week someone asks if Jasper is worth it, and a bunch of people chime in saying no ChatGPT output is just as good.
ChatGPT is also incredibly easy to use: a grandma could use ChatGPT to write an SEO friendly recipe post.
That all leaves Jasper an awkward spot. I don't think they'll die, but to stay competitive they'll need to shed a ton of dead weight. They obviously can't keep all the machinery they built around GPT-3 running if it's no longer adding value, they now have to work 100x harder on positioning, etc.
You are saying "GPT is Simply too good" (which I agree with).
But earlier today, entire thread that "AI isn't good enough"
Responding to this post https://news.ycombinator.com/item?id=37256577#37260717
People, HN, Industry, everyone is all over the place.
This is Disruption, another Winter is not coming.
It was mostly an example to highlight that the graph was a little misleading. By the same token, ChatGPT is an extension of what OpenAI has been iterating.
"Generative AI was overhyped, but it also isn't a nothing-burger"
Include some graphs. Include a poem. Sh*t talk Elon a little as well...
Generative AI burst onto the scene like a summer blockbuster, promising transformative experiences. The tech-savvy among us were both awed and slightly terrified at the notion of AI creating human-like content. But has it lived up to its hype?
The Highs of Generative AI
[Graph 1: Rapid Growth of Generative AI Adoption]
The above graph showcases the swift ascent of generative AI's adoption from 2019 to 2022. With businesses and individuals alike eager to experiment with this new tech, it was clear that there was significant promise.
From creating music, poetry, and even assisting with scientific research, the capacities seemed endless. Remember the poem generative AI penned last year that went viral?
A masterpiece, isn't it?But Wait, There's More (or Less)
However, for every positive leap, there have been equally notable stumbles.
[Graph 2: Drop in User Satisfaction with Generative AI Tools]
The above graph indicates that while initial excitement was high, satisfaction dwindled as users confronted the reality of generative AI's capabilities. It wasn't the magic bullet solution for every challenge, and it often still required human input to achieve the best results.
Remember when Elon Musk commented on generative AI? "It's like trying to use a hammer for everything – sure, you might get some nails in, but you're also likely to break a lot of things." While it's not often I find myself nodding in agreement with Musk (especially after that whole debacle with him trying to launch a line of AI-generated perfumes), he did have a point here.
Not a Nothing-Burger, Though
But let's not be too hasty. Generative AI may not have completely revolutionized our world overnight, but it has added value. It's facilitated easier workflows, added creativity in previously unimaginative sectors, and has certainly begun conversations on the potential and limits of technology.
[Graph 3: Industries Benefitting Most from Generative AI]
There are sectors, as highlighted above, where generative AI has truly shined. Content generation, arts, and even certain branches of research have seen remarkable strides thanks to this tech.
Concluding Morsels
Generative AI might not have been the pie in the sky many hoped for, but it also isn't mere table scraps. Its potential is still unfolding, and as with all technologies, it will continue to evolve.
So, whether you're a staunch supporter, a skeptic, or just someone who enjoys AI-generated poetry, there's no denying that generative AI has spiced up our tech menu. And who knows? Maybe the next big breakthrough is just around the corner.
Until then, remember:
Stay curious, [Your Substack Author Name]Note: All graphs referenced in the article are conceptual and would typically be visualized with data. Given the format, they've been described rather than shown.
>I could first point out that it is not often 69% of businesses adopt any technology simultaneously.
AI has been a buzzword talked about by executives for at least ten years now. The same issues with data organization and labelling were being explained to executives 5 years back [2] and are still relevant today. LLMs are a jump forward in NLP which enables more use cases in business, but AI and its adoption challenges are not new.
[1]https://www.axios.com/2023/08/19/ai-corporate-barriers-cost-... [2]https://www.mckinsey.com/capabilities/quantumblack/our-insig...
The current usefulness of consumer AI pales in comparison.
This post has counter arguments to everything that was said in this post earlier.
https://news.ycombinator.com/item?id=37256577#37260772
After two or three development cycles, LLMs will still be a curiosity, but nothing more. They might be used for some non critical business cases where output doesn’t need to be perfect, but for all other cases the burden of verification will be unmanageable.
The use of LLMs in academia will be nearly eliminated as professors opt for more verifiable ways of ensuring students aren’t using ChatGPT: no more writing long form essays in the comfort of your own home and with access to the internet.
Seems like the AI hype was so great, that now there is an anti-AI hype saying AI was just another over hyped tech that is fading, and this wave of hype is turning into another AI-Winter.
This article is great at pointing out that AI is really still growing, just now in the 'building' stage. All the corporations are now off building things with AI, but in their own gardens, and not publicly hyping them.
The slower growth state of the typical hype cycle, when the slow adopters are doing their adopting.
https://twitter.com/0xSamHogan/status/1680725207898816512
Nitter: https://nitter.net/0xSamHogan/status/1680725207898816512#m
---
6 months ago it looked like AI / LLMs were going to bring a much needed revival to the venture startup ecosystem after a few tough years.
With companies like Jasper starting to slow down, it’s looking like this may not be the case.
Right now there are 2 clear winners, a handful of losers, and a small group of moonshots that seem promising.
Let’s start with the losers.
Companies like Jasper and the VCs that back them are the biggest losers right now. Jasper raised >$100M at a 10-figure valuation for what is essentially a generic, thin wrapper around OpenAI. Their UX and brand are good, but not great, and competition from companies building differentiated products specifically for high-value niches are making it very hard to grow with such a generic product. I’m not sure how this pans out but VC’s will likely lose their money.
The other category of losers are the VC-backed teams building at the application layer that raised $250K-25M in Dec - March on the back of the chatbot craze with the expectation that they would be able to sell to later-stage and enterprise companies. These startups typically have products that are more focused than something very generic like Jasper, but still don't have a real technology moat; the products are easy to copy.
Executives at enterprise companies are excited about AI, and have been vocal about this from the beginning. This led a lot of founders and VC's to believe these companies would make good first customers. What the startups building for these companies failed to realize is just how aligned and savvy executives and the engineers they manage would be at quickly getting AI into production using open-source tools. An engineering leader would rather spin up their own @LangChainAI and @trychroma infrastructure for free and build tech themselves than buy something from a new, unproven startup (and maybe pick up a promotion along the way).
In short, large companies are opting to write their own AI success stories rather than being a part of the growth metrics a new AI startup needs to raise their next round.
(This is part of an ongoing shift in the way technology is adopted; I'll discuss this in a post next week.)
This brings us to our first group of winners — established companies and market incumbents. Most of them had little trouble adding AI into their products or hacking together some sort of "chat-your-docs" application internally for employee use. This came as a surprise to me. Most of these companies seemed to be asleep at the wheel for years. They somehow woke up and have been able to successfully navigate the LLM craze with ample dexterity.
There are two causes for this:
1. Getting AI right is a life or death proposition for many of these companies and their executives; failure here would mean a slow death over the next several years. They can't risk putting their future in the hands of a new startup that could fail and would rather lead projects internally to make absolutely sure things go as intended.
2. There is a certain amount of kick-ass wafting through halls of the C-Suite right now. Ambitious projects are being green-lit and supported in ways they weren't a few years ago. I think we owe this in part to @elonmusk reminding us of what is possible when a small group of smart people are highly motivated to get things done. Reduce red-tape, increase personal responsibility, and watch the magic happen.
Our second group of winners live on the opposite side of this spectrum; indie devs and solopr...
The dissonance is wild. I'd hate to go back to working without it. It definitely has its uses.
They act like it's literally useless except for plagiarism.
An electric drill's capabilities are known, in the way AI's is not. There's no drill bit coming in Q4 that will let you drill into steel for example. But there will be a GPT 4.5 or something that will add new capabilities.
After so many disappointments in the tech universe (and with some AI winter in the past) people in the industry should not be naive to the point of thinking that OpenAI would solve all the industry's problems.
>> OpenAI has marketed it as an "everything" tool
That's what all companies do all the time. Should we blame capitalism?
On the other hand, the finances of AI are only loosely correlated with it's utility to individuals. If OpenAI went from being worth 10 Billion to 500 MM, it wouldn't make too much difference to people who use OpenAI products.
So far it’s only been useful to me when working with a tool I don’t really know. But most of my work I know the tools fairly well.
I had to go through several iterations and I bounced from wanting to write it in go to I think eventually just doing it in bash, just simplicities sake, but it absolutely saved me days of research. I definitely had to fix a lot of things and tailor it to my needs and it lead me astray quite a few times but it really was more like asking a professor the question vs googling it and hoping that SOMEONE out there has done something similar. Or combing API docs.. It was such an ancillary thing that I didn't want to get hung up on it for more than 1-3 days and it literally got me going in maybe 3-6 hours.
I also write a ton of very advanced terraform and cdk with it. I already know both very well but it absolutely speeds me up if I haven't used a specific function or what not in awhile, or it's newer to me.
It's great at leading me to the water, sometimes I have to still teach myself to fish but it will give me ideas and get me going if not even more.
For me ChatGPT (v4) is very useful as it is and for my line of work is a God sent, and I give one simple real world example.
Earlier this month we have some encouraging results on earthquakes early detection based on IoT and Big Data. Our earthquake expert is not replying emails for several weeks and we had to send the research summary immediately to ask the journal editor for the suitability of potential publication based on the proposed paper's title and abstract. But since we are not the earthquake domain experts, we are not sure what's the correct word to use for the early detection of earthquake in our particular results, and the prime candidates are forecasting or prediction. Voila, upon consulting ChatGPT it's obvious that for our results, prediction is the correct word to be used.
Before Internet people used to go to library for looking up something, but then Google arrived. Similarly, before ChatGPT people used to these so called experts for consultations, but then ChatGPT arrived. But somehow, people seems to be in denial of the useful of the GPT in helping us to write better reports, papers, articles, etc.
There is no death of ChatGPT in sight. Even if the UX was a bit of a hoax to begin with. It would be if some of the other models beat it.