Hacker News new | past | comments | ask | show | jobs | submit login
Ask HN: How to not get steamrolled by OpenAI?
13 points by charbzg 16 days ago | hide | past | favorite | 25 comments
If you're building a startup on top of the OpenAI (GPT-4) platform, how do you know your idea won't be made redundant by advances in the core GPT model itself?

For example: the recent voice/conversational AI feature included with the GPT-4o release completely steamrolls 100s of startups working on enabling and improving voice conversations with AI.

So how do you know that any idea built on top of OpenAI is future-proof and actually adds value, and IF the goal of OpenAI is to literally create AGI, doesn't this mean all software is redundant?




Stop making gimmick OpenAPI wrappers.


By creating a business that can't easily be copied or made a commodity.


Cold takes are the best takes :)


During a shovel-rush, you should mine gold

(brb adding "thought leader" to my resume)


Don't forget "career coach". Where can I subscribe to your newsletter?


I've always argued the worse businesses to start are those that rely on some type of technological advantage.

These business are extremely difficult and costly to build because you're pushing at the edges of what's possible and they're also extremely easy to disrupt because it only takes one innovative team to completely erode any advantage you once had.

The only exception to this is if you can use your technological advantage to capture a market, then once you have a dominate market position your continued survival no longer depends solely on your ability to maintain an absolute technological advantage. But even this strategy is risky long-term with companies like Intel and IBM being good examples or how even dominant market players can be disrupted if they continuously fail to keep up with the pace of innovation.

My answer to you is probably not what you want to hear, but in my opinion if you're building a startup like this you should assume you're on limited time. You should only build the startup if you're genuinely passionate about the problem you're solving, and not because you believe what you're describing is a good long-term business opportunity (because it's not).

If you want to build a businesses that isn't likely to be disrupted you should consider doing something more traditional. I highly doubt individuals running landscaping businesses are worried about technological disruption, for example.


First I would say actively care about some user problem space. Make that your mission. Not "build AI startup". Not "get rich".

Let an LLM be a tool in that mission, but "AI" is not the end goal.

Second, build solutions in that space. Find the niches in that problem space nobody is addressing, or if they're addressing, they don't care as deeply as you do. So many "products" sadly are afterthoughts. Look even at Google Search, supposedly Google's bread and butter, but we see its quality suffering. You need to care _very_ deeply about the space. If you don't care deeply, it's not worth your energy to build the business. Just go get a job in your domain.

Third, think longer term beyond even the current product. Think constantly how you can disrupt yourself to get to the next product that obsoletes your current product. Build a company culture around this idea. Focus not on the last product you built, but how you build a culture for building product to address your company's purpose.

Fourth, when you think about building product - build something _so good_ that it essentially creates a habit among your users. You obsess about the tiniest of interactions in a way that is so brainlessly easy, sticky, addictive, whatever that very few will/can obsess to the same degree, in that domain, to truly build the same level of stickiness.

Finally, as OpenAI gets bigger, and tries to do more, you will find areas where they just don't care enough about a space that you can find a niche for specialized tools. Tools that are too small for them to really obsess about the same level you could. One idea, by way of an example, might be taking voice notes for doctors. Or dictation in general...

Anyway, I'd say stop focusing on "how do I build an AI thing". Stop focusing on building the next Unicorn startup. And just ask yourself - what problem do I care enough about to truly put my skills in service of to build a product?


The only foolproof way is to have some proprietary dataset that OpenAI will never be able to get by themselves or from anyone else. That or be in a very small niche they will never bother to compete in because it’s not worth their time


Here is an analogy: Look at what Amazon did. Created a marketplace for brands then based on best selling items started Amazon Basics.

This is the way of the world. If we're building products that have marginal utility with wrappers then at one or the other point there will be a hit.

Will suggest looking for blue oceans to solve and integrating more than just a single API to build a strong userbase for which you don't have to compete at a later stage.

And by no means I'm saying this is easy. Probably the hardest thing we will ever do in our lives, but if it works out then amazing.


As coldtrait says in this thread, if you're simply building a wrapper business on top of such existing APIs, you're on thin ice. Be prepared to fall any moment. It'll be the fate of more or less all start-ups that are built in an "AI hurry".

Also, don't fall for the "AGI" hype from ClosedAI. It's a super-marketing term, like their "superalignment" project.

reply


A. Feature. Is. Not. A. Business.

Your business idea should be about solving a problem for your customers. If the way you package and service your solutions saves the customers time and adds value, it doesn't matter what tools you use. The upgrade from hammers to nailguns did not put construction companies out of business.

Even if OpenAI has a ridiculously good voice/conversation set of features, OpenAI is not going to sign a service contract with a small-to-mid size businesses. That is even if that organization had the technological ability to integrate it into their business.


> So how do you know that any idea built on top of OpenAI is future-proof and actually adds value

one idea is to not build in a way that you depend on OpenAI. You should build on top of their offering in a way that it is interchangeable with other providers, especially offline local solutions which is something OpenAI being an API first company will not have a strong offering for and where you can demonstrate your value.


Well the thing is you can do some interesting tricks they haven't thought of, one trick I am doing with these tiny language models is to train them on a handful of authors in one genre rather than training on every story ever written, a distinct style emerges. Nowadays you have these mini pc's with Rocm support (under linux) so its becoming cheap to build your own models.


Here's another example, start an AI generated music company but rather than training your model on all the music on youtube you hire local musicians to develop fully legally transparent and royalty free models used to run a sass for content producers.


Take the perspective that software is a tool to get something else done. The last couple decades, people were creating and selling software for its own sake - that era may be over. So stop focusing on the tool and start focusing on what people are doing that needs the tool in the first place. Focus your business there.


You won't out-pace OpenAI, so selling a wrapper around their API that only slightly improves upon their current work is simply a stop-gap until they release model++

> and IF the goal of OpenAI is to literally create AGI, doesn't this mean all software is redundant?

Sort of means all of everything is redundant. :^)


Don't be an 'AI company', unless you can afford to fight back the big tech incumbents. Even if you do fight and you're not making any money or sitting on VC cash, you'll bankrupt yourself.

Just build a startup around subscriptions and perishables.


Welcome to software.

This scenario has played out forever, and resurfaces every time there is a new platform.

In the early days on-disk-compression was an add on, then it became part of DOS. pretty much the same thing happened with Apple (sherlock), Windows (zip), web (netscape) and I could go on.

The underlying thread here is that when you hook on to a platform, and make a living "making thar platform better" you are in competition with the platform maker.

Since the platform maker is improving their product all the time their overlap with you happens.

In one sense what are they supposed to do? Ignore improving it because someone else is doing it?

As a rule, the newer the tech, the more obvious the improvement, the more likely you are to be Sherlocked. (Yep, this effect even has a name.)

The fact that "100s of startups" saw voice as a value-add means that it's obvious, and widely useful, so (duh) thats obvious to OpenAI too.

Let me put it another way- if you are building for "lots of users" then expect that to be done.

Assuming you don't have MS or Apple or Google or OpenAI resources, assume you shouldn't be targeting "millions" of users. If you're targeting something only a few thousand people care about then you're a lot safer. OpenAI is less likely to want to add a feature that say only helps left-handed, bald, pensioners than one that benefits "everyone".

Don't be picking the low-hanging fruit. Find something that's waaaay more niche and hard to get to.


Exactly. Don't try to build something for everyone. Try to build something that will mean everything to someone.


Don't build an AI wrapper. Use AI to build a non-AI-wrapper business like I'm doing at https://demo.fun


Yeah but what's bothering me is, we're not too far away from a version of AI where someone can ask it: "create a demo for mywebsite.com, focus on the products page, display each product and its pricing" ... and then "edit the demo you just created to remove the mouse cursor, zoom-in on the product image, etc" ... so how will your business compete against that? -- not trying to bash your idea at all, just expressing my concern


> doesn't this mean all software is redundant?

Unfortunately yes.


Join unions, government, etc. For God's sake everything that is not tech is so inefficient and moves slowly.


Learn to mine coal.


Nobody (except few niche use cases) really cares about those gpt based gimmicks to begin with, regardless which company they come from.

Most of the traffic is hype and curiosity driven.

This is the whole point of the bubble.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: