Reporter’s Notebook: What The Big AI Model Companies Shared At HumanX

More than 3,000 attendees, myself included, descended into Las Vegas last week for the first HumanX AI conference.

The running theme throughout the three-day conference was trust — namely, how to build trusted outcomes from a powerful but inherently probabilistic technology.

Keep in mind, this was the same week that Apple pulled back on launching its AI features due to accuracy concerns. And at HumanX the statement that “only 6% of AI projects make it to production” blared across screens — the statistic, from an AWS study, serving as a blunt reminder of the experimentation and challenge in deploying AI.

Nonetheless, more than $100 billion was invested in the AI sector in 2024, an increase of 80% from 2023, per the joint HumanX and Crunchbase AI report compiled for the conference.

The conference mixed panels and product launches on multiple stages in a large expo hall with offshoots of Q&A sessions and product demos in smaller rooms. Throughout the voluminous spaces were lounge seating, pods, and an app to meet and make connections.

There was a lot of content, but perhaps what piqued my interest most were the perspectives, just over two years since the launch of OpenAI’s ChatGPT made generative AI mainstream, from some of the leading model companies.

OpenAI’s moat

OpenAI, valued at $157 billion, still had to answer the question from Kate Rooney at CNBC, “What is your moat?”

Kevin Weil, the company’s chief product officer for the past 10 months, noted that the days of a 12-month lead are gone, with the current reality of a three- to six-month lead still “really valuable.”

And whereas in past development cycles, “a database was a database,” Weil summed up the zeitgeist with “every two months, there’s some new model, [that] can do something that computers have never been able to do.”

Still, OpenAI’s numbers are huge. Weil stated that 3 million developers use the API, more than 400 million people use ChatGPT every week, and more than 2 million businesses use its enterprise products.

Anthropic on Claude Code

A highlight of the conference was the conversation between Alex Heath, deputy editor at The Verge, and Mike Krieger, the CPO of Anthropic, about building a model company and its plan to build apps. Claude Code, launched weeks ago, reached 100,000 users within a week.

Krieger said he reached out to Anthropic’s leading code API customers in advance of launching, as it put it in direct competition with its customers, Anysphere, maker of Cursor, Windsurf from Codeium and GitHub’s Copilot.

He pushed for the need to have first-party products in the market, as you “just can’t get that kind of feedback if you’re only an API provider,” he said.

This first-party learning will go directly into the model, “providing a level playing field, being transparent, and then feeling it out.”

“Hopefully we’ll all be able to navigate the occasionally closer adjacencies,” said Krieger.

On a more philosophical note, he acknowledged that he joined Anthropic for the role it could play in “guiding the future of human-AI interaction.”

“If it’s just chat boxes and chat bots a year from now, we’ll all have failed,” Krieger said.

Mistral, open source and smaller models

France-based Mistral AI differs from Anthropic and OpenAI as an open-source model builder to create decentralized AI, so as not to be dominated by a few companies. Arthur Mensch, Mistral’s CEO and co-founder, said there is huge demand for open source for those with data governance requirements and sovereign needs.

“What we bring on top of our open source models is a platform for deployment, for agent creation, for management of data, for management of feedback that can be deployed in a fully isolated manner,” he said.

With its leading smaller models, Mistral is active in robotic applications. “Having a small vision-to-action model deployed on specific hardware is going to be extremely important in the coming years, and we’re bringing the software stack for that.”

The company partners with Helsing on its drones and is working with robotics companies in the Bay Area.

Mistral started as an enterprise company. However, once you have the APIs, you are close to having a product, said Mensch. Mistral’s consumer product, Le Chat, was launched last month.

Next conference

Next year HumanX is headed to San Francisco, as the majority of AI investment is concentrated in the Bay. And with the projection that nearly 30% of companies presenting at HumanX are likely acquisition targets, the conference could look very different a year from now.

Illustration: Dom Guzman