managemnet company strategy managemanet Should You Start a Generative AI Company?

Should You Start a Generative AI Company?

Should You Start a Generative AI Company? post thumbnail image

I’m thinking of starting a company that uses generative AI but I’m not sure whether to do it. It seems to disappear from the ground very quickly. But if it’s easy​​​​​​for me, isn’t it easy for others too?

This year, more entrepreneurs have asked me this question than any other. Part of what’s exciting about generative AI is that the upsides are seemingly limitless. For example, if you were able to create an AI model with some sort of general linguistic reasoning ability, you would have a piece of intelligence that could potentially be adapted into a variety of new products that could be created. also use this ability – such as screenwriting, marketing materials. , teaching software, customer service, and more.

For example, the software company Luka built an AI partner called replicas which enables customers to have an open conversation with an “AI friend.” Because the technology is so powerful, Luka managers started accepting inbound requests to provide a white label business solution for businesses looking to improve their chatbot customer service. Ultimately, Luka’s managers used the same underlying technology to churn out a business solution and a direct-to-consumer AI dating app (think Tinder, but for “dating” characters in AI).

When deciding if a generative AI company is for you, I recommend establishing answers to the following two major questions: 1) Will your company compete with foundational models, or top-layer ones? application that can use these foundation models? And 2) Where along the continuum between a highly scripted solution and a highly generative solution does your company lie? Depending on your answers to these two questions, there will be long-term implications for your ability to protect yourself against the competition.

Basic Models or Applications?

Today’s tech giants rent out their most common proprietary models – ie, “foundation models” – and companies like and Stability AI provide open-source versions of these. foundational models at a fraction of the cost. The foundational models became commoditizedand few startups can afford to compete in this space.

You might think that the foundational models are the most attractive, because they are widely available and their many applications provide many opportunities for growth. Additionally, we live in exciting times where some of the most sophisticated AI is available “off the shelf” to get started.

However, entrepreneurs who want to base their company on foundational models have a challenge. As in any commoditized market, the companies that survive are those that offer unbundled offerings for cheap or that provide highly enhanced capabilities. For example, speech-to-text APIs like Deepgram and Assembly AI are competing not only with each other but with the likes of Amazon and Google in the segment by offering cheaper, no-strings-attached solutions. However, these companies are in a fierce war on price, speed, model accuracy, and other features. In contrast, tech giants like Amazon, Meta, and Google are making significant investments in R&D that enable them to continuously deliver cutting-edge advances in imaging, language, and (increasingly) reasoning. in audio and video. For example, this is estimated that OpenAI spent anywhere between $2 and $12 million computationally training ChatGPT – and this is just one of the many APIs they offer, with many more.

Instead of competing with increasingly commoditized foundational models, most startups need to differentiate themselves by offering “top layer” software applications that leverage other companies’ foundational models. They can do this by building foundational models on their own high-quality, proprietary datasets that are unique to their customer solution, in order to deliver high value to customers.

For example, the content marketing creator, Jasper AI, grew to unicorn status largely by using foundational models from OpenAI. To date, the company has used OpenAI to help customers create content for blogs, social media posts, website copy and more. At the same time, the app is tailored for their marketer and copywriter customers, providing special marketing content. The company also provides other special tools, such as an editor that several team members can work together. Now that the company has gained traction, going forward it can devote more of its resources to reducing its reliance on the foundational models that enabled it to grow in the first place.

Since the top-layer apps are where these companies find their competitive advantage, they are situated in a delicate balance between protecting the privacy of their datasets from the big technology players even though they rely on these players for foundational models. Because of this, some startups may be tempted to build their own foundation models in house. However, this may not be a good use of valuable startup funds, given the challenges discussed above. Most startups are better off using foundational models to grow quickly, rather than reinventing the wheel.

From Scripted to Generative

Your company should reside somewhere on a continuum from a purely scripted solution to a purely generative one. Scripted solutions involve choosing an appropriate answer from a dataset of predefined, scripted answers, while generative involves generating new, unique answers from scratch.

Scripted solutions are safer and constrained, but less creative and human-like, while generative solutions are more risky and unrestricted, but more creative and human-like. More scripted procedures are required for some use cases and industries, such as medical and educational applications, where there should be clear guardrails on what the app can do. However, if the script reaches its limit, users may lose their engagement and customer retention may suffer. Additionally, it’s more challenging to develop a scripted solution because you’re restricting yourself from the start, limiting your options along the way.

On the other hand, more generative solutions bring their own challenges. Because AI-based offers include intelligence, there are many degrees of freedom in how consumers interact with them, which increases the risks. For example, a married father tragically committed suicide after a conversation with an AI chatbot app, Chai, which inspires him to sacrifice himself to save the planet. The app uses a foundational language model (a bespoke version of GPT-4) from EluetherAI. The founders of Chai have since modified the app so that references to suicidal thoughts are given helpful text. Interestingly, one of the founders of Chai, Thomas Rianlan, regretted, saying: “It is not accurate to blame EleutherAI’s model for this tragic story, because all the optimization to be more the result of our efforts is emotional, fun and engaging.”

It’s challenging for managers to anticipate all the ways things can go wrong in a highly creative app, given the “black box” nature of the underlying AI. Doing so involves anticipating risky scenarios that may be very rare. One way to anticipate such cases is to pay human annotators to screen content for potentially harmful categories, such as sex, hate speech, violence, self-harm, and harassment, then use these labels to train models to automatically flag that content. However, it is still difficult to create a complete taxonomy. Thus, managers deploying more creative solutions must be prepared to proactively anticipate risks, which can be difficult and expensive. The same is true if you later decide to offer your solution as a service to other companies.

Since a fully generative solution is closer to natural, human-like intelligence, it is more attractive from the point of view of retention and growth, because it is more attractive and can be used in newer cases of use.

• • •

Many entrepreneurs are thinking of starting companies that use the latest generative AI technology, but they have to ask themselves if they have what it takes to compete with the best-selling foundational models. , or whether they need to be different in an app that uses these models.

They should also consider what type of app they want to offer on the continuum from a highly scripted to a more creative solution, given the various pros and cons that come with each. Offering a more scripted solution may be safer but limit their retention and growth options, while offering a more generative solution is full of risk but more engaging and flexible.

We hope that entrepreneurs will ask these questions HISTORY dive into their first generative AI venture, so they can make informed decisions about what kind of company they want, scale quickly, and maintain long-term protection.

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Post