The Great Middle Layer AI Gold Rush of 2023

In our last post, France discussed what a Middle Layer AI Model is and why they build upon the power of Foundation Layer AI Models to create platforms that are capable of doing so much more. In this post, I’ll discuss what it takes to build a Middle Layer AI company and why boodleAI is well past the starting line.

The race is on to create the Middle Layer AI models that will reshape the internet.

With the overnight success of ChatGPT, the generative AI user adoption tipping point is finally upon us, setting the stage for or a Middle Layer AI gold rush primed for companies that are ready.

Sam Altman, the Founder and CEO of OpenAI, spoke about where the opportunities lay in today’s AI world saying:

“I think it’ll be this middle layer. The startups will train their own models, just not from the beginning. They will take base models that are hugely trained with a gigantic amount of compute and data and then they will train on top of those to create the model for each vertical. And those startups, in some sense, are training their own models, just not from scratch. They’re doing the one percent of training that really matters for whatever this use case is going to be. Those startups I think they will be hugely successful and very differentiated…”

What does it take to create a true Middle Layer AI company?

We may have made it sound like creating a middle layer model is as simple as adding data to a pre-trained model such as GPT-4. 

Let’s be clear: it’s not.

Yes, some platforms have quickly integrated out-of-the-box GPT functionality in an effort to get ahead of the competition. (Example: “Our AI assistant can draft a personalized message for you!”) These early integrations — while useful and a great introduction to generative AI — are using GPT’s capabilities in a way that any company can. They replicate the native ability of ChatGPT to respond to good user prompts. (See the “Be your own AI assistant” Bonus Section below on how to do this.) This type of GPT integration will soon be commonplace.

But it’s not Middle Layer AI.

As noted by Oliver Molander, to build a “middle-layer” AI that does a better job of solving the problems of a specific vertical or organization, a company must have the following:

1. Unique data set
2. Powerful AI/ML and data infrastructure
3. Rare talent to train models

The above enables a company to build a middle layer AI that can generate responses that ChatGPT can’t and that are distinct from those of other middle layer AIs.

So, is boodleAI up to the task of middle layer greatness?

You better believe it! As it turns out, boodleAI has all three ingredients:

1. Unique data set: A proprietary dataset of 35 billion predictive insights about 250 million adult Americans. 
2. Powerful AI/ML and data infrastructure: A proven predictive analytics engine and identity resolution engine consisting of 300,000 lines of code across 140 repositories, both validated by use by hundreds of customers
3. Rare talent to train models: An in house, onshore development and data teams that created the above data and infrastructure.

Over the next few posts, we’ll deep dive into each of these ingredients to explore how they are guiding our middle layer AI journey.

Bonus: Be your own AI assistant

An increasingly common integration of GPT-3.5 or GPT-4 is message personalization within other platforms. This is accomplished by having the user specify the desired message (length, intent, audience, author, tone, etc.), which is then turned it into a prompt that GPT, through an API, uses to generate and return the desired message.

You can do the same thing yourself through a carefully worded prompt to ChatGPT. For example, let’s say you wanted to draft a fundraising direct mail letter from a development director to a potential major gift donor with a light-hearted tone that is no more than 250 words long and incorporates an ask for $1000 in support of this year’s annual fund. Here’s the prompt you can enter into ChatGPT right now:

Act as a nonprofit fundraising assistant. Generate a 250 word letter from the development director to a potential major gift donor asking for $1000 in support of this year’s annual fund. Draft the letter in a light hearted tone.

Here’s the result:

ChatGPT of course generates a letter that meets all the requirements and does so in seconds – much faster than what it would take a human to draft a letter from scratch. Incredibly useful and the reason why generative AI is going to be everywhere soon.

However, the user still had to specify the author, the length, the tone, the ask amount, and the specifics of the letter. Also, the letter did not incorporate anything specific about the recipient beyond what was included in the prompt.

But imagine this …

If a LLM incorporated an organization’s donation history and previous donor appeals, the LLM could recommend the most effective fundraising strategy, and then generate the required appeal:

This is the future we want to bring to the present with boodleGPT.

References

AI for the Next Era | Greylock

Generative AI: How To Spot Real From The Hype | Moveworks

ChatGPT, LLMs, and Foundation models — a closer look into the hype and implications for startups | by Oliver Molander | Better Programming

The generative AI revolution has begun—how did we get here? | Ars Technica