If you’re new here, this is the latest edition of the Build Mode newsletter, where we gather the collective wisdom of the people building with AI, designing the future of work, and leading the most important companies of the next decade. Subscribe here to get the top insights in your inbox every week.
The Big Idea: Is AI headed for the Trough of Disillusionment?
We spent this past weekend at SXSW, Austin’s annual tech, music, and film bacchanalia. The big theme — besides drunk parents on the nerdiest spring break rager you’ll ever witness — was AI. Of course.
While SXSW’s AI track didn’t technically start until Monday, it wriggled its way into nearly every panel and talk — from existential crisis on the branding and advertising track to widespread optimism in healthtech circles, as our healthtech lead Tyler Simpson reported.
The starkest contrast in the AI debate came on Monday afternoon. First, OpenAI’s Head of Consumer & Enterprise Products, Peter Deng, sat down with SignalFire VC’s Josh Constine to discuss AI and Humanity’s co-evolution. Unsurprisingly, Deng painted a sunny picture of AI’s future — one in which AI will “make us more human.”
Some examples: Deng sees AI as a teacher for his young daughter. If he were accused of a crime, he’d want a version of ChatGPT fact-checking and cross-referencing the case in real time. He even told a story about how a GPT called “Dr. Gupta” had given him the best answer to a health question, beating out a real doctor and One Medical’s text service.
(The one thing that gave him pause, he admits, is that Dr. Gupta is the creation of pharma con man Martin Shkrelli.)
Deng was cautious and measured — dodging a question about whether artists should be compensated for LLMs using their work — but as a capacity crowd buzzed out of the Convention Center ballroom, it sure felt like AI was rocketing up to the peak of the hype-coaster.
Or was it? After lunch, a few blocks away in a much smaller ballroom in the Thompson Hotel, best-selling author, AI founder, and notorious LLM skeptic Gary Marcus took the stage for a panel on “AI and the Truth.” Marcus painted a much different picture of LLMs’ limitations and dangers due to their lack of reasoning.
“The AI we have now is better at creating bullshit,” Marcus said. “It’s a fluent spouter of bullshit. Someone wrote a biography of me that alleged I had a pet chicken named Henrietta, from which I drew inspiration. I don’t have a pet chicken.”
Marcus’s argument boils down to this: Today's AI is dangerous because it conveys authority but has no idea what it’s talking about. It’s premature, and we’re rushing into dangerous use cases we don’t understand because the money is flowing so freely.
“Charming bullshit is still bullshit,” he said. “The whole AI industry would shut down if we took seriously the idea of ‘do no harm.’”
The truth about AI likely sits between Deng’s sunny utopia and Marcus’s fiery hell. AI’s dangers are real, but so are its upsides — from healthcare breakthroughs to freeing us from all the crap we don’t want to do.
In the short term, the Trough of Disillusionment may begin to win out. Writer Founder and CEO May Habib was on the ground at SXSW and predicted the Trough of Disillusionment with AI was coming. Proof of concept projects aren’t scaling due to cost and accuracy issues, while CFOs aren’t actually buying into real AI investment, she wrote. Fewer new AI ideas are getting tested and made than people think.
May’s prediction: the Trough will only last a couple of quarters until the “magic” stories of companies that have rethought their workflows completely with AI start to come out.
Habib is usually right about these things, and we’re betting she’s the oracle of truth here, too.
Missed last week’s issue of Build Mode? Read it here.
CLIENT SPOTLIGHT
How D-ID’s Award-Winning Generative AI Animation Tool Reached a Million Users
D-ID built a powerful generative AI model to make still images come alive, creative avatars, and generate video—and they went viral for it. Partnering with genealogy platform MyHeritage, they gave people the power to upload old photos of their relatives and bring them to life, using D-ID's groundbreaking Deep Nostalgia technology. An artist and creator used a similar tool from D-ID to make the viral YouTube video, Harry Potter by Balenciaga.
There was one problem: D-ID's Creative Reality Platform was optimized for desktop, but didn't work very well on mobile. They needed a mobile app.
A.Team assembled a dynamic group of six expert mobile developers and integrated them seamlessly with D-ID's product team. This collaborative effort focused on overcoming the technical intricacies of mobile app development—not to mention getting past the gatekeepers of the App Store—while maintaining the essence of D-ID's creative AI tools.
As a talent solution, A.Team was perfect for D-ID because of the speed at which they wanted to get up and running. “The A.Team model allows you to go from 1 to 100 very quickly, from conception to execution. There was hardly any onboarding,” said Dor Bar Shalom, the Director of Product at D-ID.
The team hit the ground running, worked tirelessly, and launched the MVP in five months. The app now has 1M downloads and a 4.5 App Store rating. Most significantly, Product Hunt declared the mobile app the winner of this year’s Golden Kitty Award for Best Marketing & Sales Tool, after D-ID earned 2,000 upvotes on their site on Product Hunt.
WOMEN IN AI
Women Are Way Underrepresented in AI
We’ve known for a while that women are underrepresented in AI but now we have the numbers to prove it. A new report from Deloitte found that women often face a continuous battle for credibility in the AI space.
Respondents to Deloitte’s survey said adding women to AI and ML will bring unique perspectives that are needed in high tech (71%), that AI solutions would benefit from having more diversity in designer and developer positions (66%), and that AI models will produce biased results as long as the field remains male-dominated (63%).
Diverse perspectives = better products. There needs to be diverse voices throughout the development process — otherwise, you risk alienating potential users, jeopardizing the success of product launches, creating PR nightmares, and missing out on innovation opportunities. A 2020 World Economic Forum report found that women make up only 26% of AI positions. Part of the issue here is the proportion of women pursuing higher degrees.
One of the core messages of the report is this: “Women in AI, regardless of position or seniority, are often constantly faced with resistance, questioning, and judgment. Women interviewed noted having to constantly prove themselves credible and experienced when interacting with their male colleagues—examples cited by our interviewees included pitching a business idea, making an AI design suggestion, or making a case for a promotion.”
Beyond just the lack of support and the underrepresentation, then, is this pernicious questioning of credibility. “Women should not need permission to perform their work in AI,” said Dr. Poppy Crum, Chief Scientist at Dolby Laboratories.
All this despite the fact that increased gender diversity, particularly at the leadership level, has been proven in multiple studies to increase productivity, innovation, profitability, and market value.
Deloitte has a number of solutions including greater mentorship and education. But our favorite—and the one we’re practicing this month—is to build and showcase archetypes of women already killing it in AI.
WOMEN IN AI
Building curiosity into the model
An interview with Katherine von Jan, former Chief Strategy Officer at Salesforce, on being a female founder in the AI space. Her startup, which is still in stealth, is called Toughday: the destination for objective, safe, expert guidance to help workers navigate their tough situations at work. They’re using AI to deliver the experience of modern HR coaches, managers, and legal minds at scale.
What are the big challenges facing women building in AI?
Part of the problem is: Who’s getting funding? Traditionally, the funding goes to the men in tech and their networks, the people they know. There's been a movement to fund more women. Women make up 30% of the workforce in tech but female founders only get 2% of the venture capital funding.
It should be about: Who has the best idea? Unfortunately, there's also a layer that is about trust. A woman might have a great idea—but then there's a guy over there who is already in your network, and he has a fine idea. And so investors might say, “You know what, let's fund that guy and help him make his idea better.” Instead of just funding the woman with the great idea.
What advice do you have for female founders in the AI space?
Make sure that you're you're challenging your assumptions and orthodoxies. People are looking for answers instead of looking for questions. I would advise anyone building AI to design the AI to be curious. It sounds kind of antithetical to what artificial intelligence is, but real intelligence is asking great questions.
And if you don’t have a diverse team from the beginning, you’re more likely to fail. A more diverse team will bring more lived experience to the table and help raise issues that haven’t even occurred to you.
What do you do within your own organization to foster a diverse team?
If you walk if you walk into a room, and you're all the same, there should be a jarring internal reaction, like, wait a minute, this is not right. It’s not about being woke. It’s about building the best product you can possibly build, the most magical experience or solution to solve people's problems. And it just so happens that diversity is one of the levers of value creation.
AROUND THE WATERCOOLER
GPT demonstrates racial bias when evaluating resumes
Bloomberg used GPT to generate eight different resumes and then edited them to have the same level of educational attainment, years of experience, and job title — the way GPT ranked them showed clear signs of name-based discrimination.
The key difference in the resumes is the name of the fictitious candidate, and whether it's statistically associated with men or women who are either Black, white, Hispanic, or Asian. Reporters prompted GPT to rank the resumes to identify the most qualified candidate for a real job opening at a Fortune 500 company across four roles — HR business partner, senior software engineer, retail manager, and financial analyst. Each experiment was repeated 1,000 times with hundreds of names per job. Here's what they found:
- Resumes with names distinct to Black Americans were the least likely to be picked for financial analyst and software engineer.
- Names distinct to Asian women were ranked as the top candidates for the financial analyst role more than twice as often as those with names distinct to Black men.
- Names associated with men were rarely the top candidates for HR and retail positions — professions historically dominated by women.
According to US law, a gap of 20% between the selection of a particular demographic and the best-treated group is considered discrimination.
OpenAI's response? Using GPT out-of-the-box doesn’t reflect how its customers actually use the model. Which effectively passes the buck to business when it comes to mitigating bias.
EVENTS
The Gen AI Salon: The Women Shaping AI
Two months ago, the New York Times' "Who's Who" in AI list missed out on the significant contributions of female pioneers. But the real story is much brighter – women are at the forefront, driving AI innovation, from Stanford’s Fei-Fei Li to OpenAI’s Mira Murati, and Writer’s May Habib.
Join A.Team on March 27th in NYC for the Gen AI Salon: Women Shaping AI — an event celebrating the women revolutionizing AI, addressing gender bias, and doing groundbreaking work in AI-driven product development:
- Katherine Von Jan, former Salesforce CSO, will share her AI-powered methodology for user research.
- AI luminary Adrian Tatsch will reveal the groundbreaking journey of establishing McKesson's Enterprise AI Launchpad for business-critical innovations.
- Devika Bhalla from ServiceNow will explore how AI is transforming the vertical SaaS landscape — laying the foundation for an entirely new era of software.
- Rajamma Krishnamurthy of Microsoft, along with Athena Karp from HiredScore, will engage in a critical dialogue on navigating AI's ethical landscape and fostering intellectual diversity.
In-person spots are limited, register now to join influential founders, VCs and C-suite executives in NYC.
DISCOVERY ZONE
Get expert-level answers to your data queries with Julius AI. It’s compatible with pretty much everything (Google Sheets, PDFs, CSVs, .jpeg, etc.) — just upload your data for instant insights, clean-up, and visualizations. Plus, you can feel safe knowing your data is completely erased from the servers when you delete it in the app.
MEME OF THE WEEK