“Big takeaway from the GPT paradigm is that the world of text is a far more complete description of the human experience than almost anyone anticipated.” -Greg Brockman, President of OpenAI

In today's email:

• This Company is Leading the Charge in AI for Businesses

• ChatGPT for your Databases

• Research Proves AI is Developing Theory of Mind

• AI Generated Music

• Image-Gen Product Photography

This Company is Leading the Charge in AI for Businesses

OpenAI, Google, and Microsoft are almost exclusively the companies mentioned in the AI race, but this publicly traded company has been making headlines for its active deployment of AI tools for businesses.

C3.ai is a software company that specializes in the development of artificial intelligence and machine learning technologies.

C3 was founded in 2009 by billionaire Tom Siebel (who was trampled by an elephant that same year and nearly died) and provides customized tools for organizations to create their own low-code and no-code AI applications.

C3’s new flagship product is called "Ex Machina"

Ex Machina is designed to simplify the process of developing and deploying AI applications by providing a visual, drag-and-drop interface for customizing and integrating AI algorithms.

The target audience for this product are "citizen data scientists" - people who work in businesses and might not have a background in data science but still want to use predictive machine learning.

The company currently has a $2.5B value on the NYSE, and is up over 90% since January 1st, perhaps due to the ongoing AI hype-cycle and a well chosen stock ticker symbol ($AI).

Worth exploring. Not investment advice, however C3 is one company to keep an eye on in the AI space.

ChatGPT for your Databases

Bink is an AI software startup that allows you to have a conversational dialogue with your data - and it may be the best example of GPT integration into business software available.

Users can ask questions to a chat-based interface and receive answers without knowledge of SQL, MongoDB, or any other traditional database systems.

It is a near certainty that software like Bink will be used at almost all data-driven companies within 2-3 years.

AI-powered data exploration like this could turn the average employee at a company into a data analyst.

The only thing the user needs to know is the question they want answered about their data.

This makes data analysis 10x faster and more accessible.

↓

The Rise of Sentient AI: Stanford Leads the Way with Theory of Mind Research

"Theory of mind" (ToM) describes how humans interact with each other, communicate, show empathy, understand themselves, and make moral decisions. In short, it's the ability to understand what's going on in someone else's head.

In a recent study, one researcher suggest that ToM abilities (which were thought to be unique to humans) might have just popped up as a side effect of these language models getting better at language.

Michal Kosinski, a Stanford professor studying computational psychology, found that the most recent version of GPT-3 has a theory-of-mind-like ability comparable to that of a human.

ToM has been studied with large language models before 2022, but early findings showed that these AI models couldn't solve ToM tasks at all.

But when Kosinski put GPT-3 to the test using the January 2022 version of GPT-3 (davinci-002), the model was able to solve 70% of the ToM tasks.

That's about as good as a 7-year-old kid.

And the November 2022 version of GPT-3.5 (davinci-003) was even better, solving 93% of the tasks.

That's on par with a 9-year-old.

In the above example from the research, GPT-3.5 can track a cat's location and understand a person's changing beliefs about the cat's location.

The likelihood of the cat being in a basket or a box changes as the story depicted in the image progresses, and GPT-3.5 accurately predicts these changes.

The model’s predictions are not just based on where the person put the cat, but also take into account other factors such as the movement of the cat and the person's presence. Kosinski’s found that GPT-3.5's predictions of a person's beliefs and actions were accurate and consistent.

This type of complex understanding is not a programmed goal of large language models, which are designed to predict text based on a mathematically weighted likelihood - but this ability is believed to have emerged spontaneously as a result of the models improvement in the understanding of semantic language.

If this trend continues, it seems that large language models like ChatGPT could establish a comparable theory of mind ability to an adult, and perhaps beyond, within a few years.

This short published paper (20min read) is worth checking out if you’re interested in exploring what went into the theory-of-mind tests.

It was one of the most viewed papers online this week

↓

From Twitter: AI Music Creation

Last week, we tweeted a look into how researchers at Google have developed a method of generating music based solely on text prompts.

Though the model hasn’t been released for public use, it is a look into how the future of music creation may be reduced to text prompting.

AI for Product Photography

Booth.ai uses AI image generation for placing backgrounds into product photography. Users can either use a pre-loaded “smart background” or provide a text prompt to create a custom background.

Thank you for being one of the 10,000 subscribers to this newsletter. Feedback and recommendations are always welcome. Simply reply to this email and we'll get back to you.

Share an interesting AI project by submitting it here

See you next Monday.

Why paperclips?