AI Made Friendly HERE

John Pasmore Tries to Make AI Less Biased. Just Don’t Call It BlackGPT

On a recent cold, rainy afternoon in downtown Brooklyn, I met John Pasmore, the CEO of AI startup Latimer.

He’s a Black founder who is trying to accomplish something that seems almost impossible these days: Making AI less biased. Just don’t call it “BlackGPT.” He’s not a fan of the moniker, which has somehow attached itself to his project.

After years of preaching about AI ethics, the tech industry has piled into a race to build the most powerful AI models. Making sure these products reflect the diversity of different cultures has taken a back seat to energy-sucking training runs, obsession with benchmarks, and endless pitches about AGI and the coming productivity revolution.

If all this seems overwhelming, Pasmore doesn’t show it. For our interview, he chatted freely while tucked away on a couch in a bar at a Brooklyn coworking space, sipping Maker’s Mark neat.

In fact, the glaring shortcomings of AI models, and a relative lack of other founders working on these problems, attracted Pasmore.

AI has well-documented blindspots when it comes to Black culture. Outputs are biased because models are trained on data that is mostly white, male, and Western. Pasmore realized he might be able to make a better alternative, and founded Latimer in May 2023 to try.

“I just didn’t think that people on Twitter asking one of the giant tech companies to fix Black history was going to work out,” Pasmore told me. “I didn’t think they would be responsive to that, and that it was something we could do ourselves.”

Navigating two worlds

Pasmore is well positioned for this moment, with a background that gives him unique perspectives on AI’s promise and its failings.

He grew up in a majority-white, middle-class neighborhood on Long Island. His father came to the US from Jamaica and owned a gas station. His mother, from Trinidad, worked as a nurse for PanAm Airlines when it offered in-house medical care to airline staff.

“Based on my background and childhood I can kind of exist in and navigate both worlds fairly easily,” Pasmore said.

In his 20s, he moved to Texas and worked at Chemical Bank, which ended up as part of JPMorgan Chase. Rather than climb the rungs of corporate finance, Pasmore decided to go in another direction.

“I jumped off the corporate path fairly early on,” he recalled. “My first business, Intrigue Magazine, was kind of a youth-culture, city-guide publication.”

From Intrigue to AI

Russell Simmons

Russell Simmons

AP Photo/Dima Gavrysh

Intrigue was eventually sold for an undisclosed amount. By 1995, Pasmore and hip-hop mogul Russell Simmons were building media brand OneWorld. From there, Pasmore joined Paris-based Trace TV as an EVP.

But it wasn’t long before the founder bug came biting, and Pasmore launched in 2007. That venture-backed video travel platform was sold to Next 1 Interactive in 2013 for $10 million.

Then came advisory roles at companies like Strides AI, where he worked to diversify corporate boards. He also put in a seven-year stint as managing partner at messaging platform Yestime.

He always felt he’d missed out on being a technical founder and became increasingly interested in AI. In 2019, he went back to school and earned a BA in computer science from Columbia University with further training in machine learning from Cornell.

Along the way, he also returned to his finance roots and joined private equity firm TRS Capital as a partner in 2021.

Esther backs Latimer

Esther Dyson

Esther Dyson

Esther Dyson

Pasmore saw that the conversation around AI, and the need for inclusive tools, had reached a new inflection point.

By this stage, Pasmore had intimate experience with a common conundrum among Black tech entrepreneurs: How to catch the attention of venture capitalists and raise money for your idea.

“The expectation on the VC front is that you’re able to go out to your friends and family and raise a bunch of money,” Pasmore told BI. “That is a hurdle if your set of friends and family is not in a position to do that.”

So, Pasmore tapped his network. He’d known prominent angel investor Esther Dyson for more than a decade and pitched her on his vision for Latimer.

She quickly became Latimer’s first investor because she saw the startup’s potential to address AI’s troubling tendency to ignore Black history and culture.

“AI won’t solve all of these problems, but it will do what we as humans ask it to do. Use Latimer, ask it to find the history that was hidden,” Dyson told BI. “An AI hallucination is a hypothesis and the only way to prove one or disprove one is to go search out the facts.”

Dyson, daughter of famed scientist Freeman Dyson, has backed successful startups, including Flickr and Square, while chairing the influential Electronic Frontier Foundation.

“The value of AI isn’t the nodes, it’s in the edges and the relationship between the kinds of data,” she said. “The big business model is going to be quality assurance for the information supply chain.”

A different approach to AI training data

Legitimizing information sources that form the basis of training data for AI models will become increasingly important, and it’s something Pasmore has spent a lot of time thinking about.

Rather than web crawlers sucking up information from the internet, which makes it hard to give compensation, consent, control, and credit, Latimer’s data strategy is different.

It centers on partnerships with historically Black colleges and universities, and licensing deals with Black media companies and other similar content producers.

The lead content czar for Latimer’s database is Temple University’s Molefi Kete Asante, who created the school’s first African American studies Ph.D. program.

“Black and Brown publishers want to both financially benefit from having produced this content, but also don’t want the content to be so siloed that you can only get accurate information from one place,” Pasmore said.

Latimer has also partnered with Morgan State University to develop a method for detecting bias in AI-generated text. This method is baked into a suite of business offerings Latimer is set to launch soon. In February, the company announced it was launching an API that can detect bias in text.

How Latimer’s tech works

Now, just under a year after Latimer started, Pasmore and his team of eight employees are busy building out its specialized datasets and integrating this with OpenAI GPT models and alternatives such as open-source model provider Mistral AI.

Latimer works by tapping OpenAI’s GPT-3.5 for its baseline capabilities. Before Latimer answers queries, it pulls in information from a trove of data on Black culture and history through a process known as retrieval-augmented generation, or RAG.

Latimer is among a growing list of companies building new tools atop foundational AI models using RAG, including search-engine startup Perplexity AI.

Pasmore’s startup has been differentiated from rivals by being labeled the “BlackGPT.”

While catchy, Pasmore is not a fan of the nickname. That’s partly because it’s too close to ChatBlackGPT, a product developed by Erin Reddick to highlight Black culture and history.

Of course, if AI models actually reflected diversity accurately, there’d be no need for such a name. Pasmore wants to make this a reality someday.

Correction: March 28, 2024 — An earlier version of this story misstated a relationship between Intrigue and OneWorld. These businesses are separate entities.

Originally Appeared Here

You May Also Like

About the Author:

Early Bird