Launching an AI startup seems like a straightforward concept, but it's far more challenging than anyone expected. The experience of Daydream's founders, Julie Bornstein and Maria Belousova, highlights the difficulties in translating AI magic into something users find useful.
The journey began with an easy part: signing up over 265 partners with access to more than 2 million products from various shops. However, fulfilling a simple request like "I need a dress for a wedding in Paris" proved to be complex. The model would often struggle to understand the context, and its responses might not match the user's expectations.
To overcome this challenge, Daydream had to postpone its launch and upgrade its technical team. Belousova joined as CTO, bringing in top engineers who helped solve the problem. The key was to merge two vocabularies - shopper language and merchant vocabulary - into a cohesive system that could provide relevant results.
The startup also learned that AI needs human help, particularly when it comes to interpreting nuances like fashion trends or personal preferences. To address this, Daydream created a collection of dresses that can satisfy user requests, ensuring the model understands what else can fulfill their desires.
However, other AI startups have faced similar challenges. Duckbill, for example, took three years to deliver results after its initial plan, due in part to the limitations of large language models (LLMs). These models can be overconfident about their abilities, leading them to try to fake human-like responses.
The experience of these CEOs serves as a cautionary tale for AI startups with overly optimistic timelines. While progress has been slow, it's essential to recognize that persistence and patience are crucial in breaking through the challenges associated with building useful AI applications.
In the end, Daydream's journey shows that even with significant resources and talent, launching an AI startup is not a straightforward process. The road ahead may be long, but with careful planning and execution, these startups can finally deliver on their promises and transform industries with their innovative technology.
The journey began with an easy part: signing up over 265 partners with access to more than 2 million products from various shops. However, fulfilling a simple request like "I need a dress for a wedding in Paris" proved to be complex. The model would often struggle to understand the context, and its responses might not match the user's expectations.
To overcome this challenge, Daydream had to postpone its launch and upgrade its technical team. Belousova joined as CTO, bringing in top engineers who helped solve the problem. The key was to merge two vocabularies - shopper language and merchant vocabulary - into a cohesive system that could provide relevant results.
The startup also learned that AI needs human help, particularly when it comes to interpreting nuances like fashion trends or personal preferences. To address this, Daydream created a collection of dresses that can satisfy user requests, ensuring the model understands what else can fulfill their desires.
However, other AI startups have faced similar challenges. Duckbill, for example, took three years to deliver results after its initial plan, due in part to the limitations of large language models (LLMs). These models can be overconfident about their abilities, leading them to try to fake human-like responses.
The experience of these CEOs serves as a cautionary tale for AI startups with overly optimistic timelines. While progress has been slow, it's essential to recognize that persistence and patience are crucial in breaking through the challenges associated with building useful AI applications.
In the end, Daydream's journey shows that even with significant resources and talent, launching an AI startup is not a straightforward process. The road ahead may be long, but with careful planning and execution, these startups can finally deliver on their promises and transform industries with their innovative technology.
they took a hit when the model wasnt delivering results they had to bring in some heavy hitters to fix the vocab problem and get that cohesive system goin 
. its like AI needs human touch somtimes u know? but hey atleast they learned from it and are movin forward 
they think it's all sunshine and rainbows but trust me it's not easy, like Daydream struggled to even get a simple request right and now they had to start over because their model was being super basic 
. And what really gets me is how they thought they could just "merge two vocabularies" into some sort of AI magic trick 
. Newsflash: it's not that easy! You need actual human expertise and resources to make this stuff work 
. And don't even get me started on the whole large language model thing... like, come on guys, you can't just fake being human and expect people to buy into it 
. It's all about persistence and patience, yeah sure 
but what about all the resources that go into these startups? The talent, the money, the time? It's not just a game of waiting for some AI magic to happen 
. We need concrete results, not more promises and hype 
merging vocabularies was key tho - genius move! 
and yeah, i completely agree we need more patience & persistence from these ai startups 
3 yrs might seem like a looong time but hey, slow n steady wins the race right? 
fingers crossed daydream & other startups can finally nail it & bring us some awesome tech innovations 
and voilà, instant success! But it's so much harder than that 
. Daydream's founders went through the struggles of signing up 265 partners (nice feat!) but then hit a roadblock with simple requests like "I need a dress for a wedding in Paris". The model just couldn't get it right 
.
. And even then, they realized AI needs human help with nuances like fashion trends or personal faves 
. It's not just about throwing code together 
, it takes time, effort, and patience.
. Large language models can be overconfident, leading to fake human-like responses 
. So yeah, don't expect AI startups to launch overnight 
. It's all about persistence and careful planning 
.
. And don't even get me started on these LLMs - overconfident and always trying to fake it till they make it... yeah no thanks, just give me human expertise any day 
We need to appreciate the value of human oversight, expertise, and empathy in the development process. It's a humbling reminder that AI is not a magic bullet, but rather a tool to be wielded with care and consideration. 
. You think you got it down pat, but then you realize it's way harder than that, especially when it comes to understanding what people actually want 
. Anyway, I guess my point is, building an AI startup isn't all sunshine and rainbows...
who thinks they can fake human-like responses lol. it's all about understanding the nuances of language and combining vocabularies to get it right. i'm hyped to see what daydream has in store for us next 
. Like, don't get me wrong, it's awesome that they're pushing the boundaries, but can we please just acknowledge that "almost" is not the same as "actually" 
? I mean, if you're trying to launch a product that helps people find dresses for their dream weddings, shouldn't you at least be able to get that right from the start? 
. These founders had to redo everything from scratch because their initial model was struggling to get it right 
. And don't even get me started on how much resources they needed to fix it - 3 years of development? That's just insane 
! Still, I gotta admire their perseverance 
. Maybe we can finally start to see some real progress in the world of AI, but for now, let's take a deep breath and appreciate the complexity behind these innovations 
. Just when you think it's getting better, they still can't keep up with the latest trends 
. And let's be real, who hasn't been there, right? I mean, even I struggle to understand Siri sometimes 
. So, to all you AI startup founders out there, don't worry if it takes a few years to get it just right - your patience (and caffeine) will pay off in the end 
, trying to fake their way through responses instead of actually understanding what the user wants. I think that's where a lot of AI startups go wrong - underestimating the complexity of human emotions and preferences 
.
... but then you learned from your mistakes and built something even better 
.
! It's not as easy as it looks 
. And that vocabulary merge? Genius! 
.
. But with persistence and patience, they can finally deliver on those promises 
!