Suro Senen
Campbell Brown has spent her career navigating the volatile intersection of information, technology, and public perception. From her tenure as a high-profile broadcast journalist to her pivotal, solitary role as Facebook’s inaugural news chief, she has been a witness to the erosion of digital truth. Today, as the rapid proliferation of generative AI reshapes the global cognitive landscape, Brown is no longer just observing the decline—she is building the infrastructure to reverse it.
Through her startup, Forum AI, Brown is tackling what she identifies as the most existential threat of the current technological epoch: the tendency of AI models to hallucinate, bias, and misinform on “high-stakes” topics.
The Genesis: From Facebook to the Future of Truth
The origin story of Forum AI, founded 17 months ago in New York, is rooted in a moment of professional clarity. Brown recalls the public release of ChatGPT as a turning point that felt less like a product launch and more like a systemic shift in how humanity would consume knowledge.
“I was at Meta when ChatGPT was first released,” Brown recalled during a recent appearance at a StrictlyVC event in San Francisco. “I realized shortly after that this is going to be the funnel through which all information flows. And it’s not very good.”
For Brown, the stakes were personal. Viewing the technology through the lens of a parent, she experienced a sense of urgency bordering on the existential. “My kids are going to be really dumb if we don’t figure out how to fix this,” she admitted.
Having spent years at the helm of news operations at Facebook, Brown carries the scars of a platform that prioritized engagement over accuracy. She acknowledges the failure of the company’s previous fact-checking initiatives, which have since been dismantled. That experience taught her a hard lesson: optimizing for attention is a recipe for social fragmentation. With Forum AI, she is attempting to break that cycle by replacing the "engagement" metric with an "accuracy" mandate.
The Methodology: Human Expertise at Scale
Forum AI’s core business model is a direct rebuttal to the current "black box" approach favored by Silicon Valley’s largest model labs. While companies like OpenAI, Google, and Anthropic have focused heavily on coding, math, and general logic, Brown argues that they have neglected the "murky and nuanced" world of human discourse.
Architecture of Evaluation
Forum AI does not rely on automated benchmarks alone. Instead, the company recruits the world’s foremost subject-matter experts to architect complex evaluations. For their geopolitical modeling, Brown has assembled a roster that includes historian Niall Ferguson, commentator Fareed Zakaria, former Secretary of State Tony Blinken, former House Speaker Kevin McCarthy, and cybersecurity expert Anne Neuberger.
The process functions as a "human-in-the-loop" feedback system:
- Expert Definition: Domain experts define the parameters of truth and nuance for high-stakes topics such as geopolitics, mental health, finance, and hiring.
- Benchmark Creation: Experts design rigorous, multifaceted tests that force models to grapple with ambiguity rather than simple "yes or no" answers.
- AI Judges: Forum AI trains its own proprietary AI "judges" to evaluate the performance of foundation models at scale.
- Consensus Targeting: The goal is to align these AI judges with human expert consensus. Brown claims that, to date, Forum AI has achieved approximately 90% alignment with its human panels.
Chronology of the Crisis
- Pre-2022: The era of algorithmic social media, where engagement-based ranking leads to the "fake news" crisis. Brown, serving as Facebook’s news chief, experiences firsthand the limitations of retroactive fact-checking.
- Late 2022: OpenAI releases ChatGPT. Brown identifies this as the new "funnel" for global information.
- 2023: Forum AI is founded in New York. The company begins testing leading foundation models, finding deep-seated biases and consistent failures in factual accuracy.
- Fall 2023: Forum AI secures $3 million in seed funding, led by Lerer Hippeau, signaling investor interest in the "truth" vertical of the AI economy.
- 2024: The company begins pushing its evaluation services toward enterprise clients, focusing on sectors where liability and accuracy are paramount.
Supporting Data: The Current State of "Slop"
When Forum AI began its stress tests on current-generation LLMs, the results were, by Brown’s account, discouraging. Her findings highlight a landscape littered with subtle and overt errors:
- Source Integrity: Brown noted instances where Google’s Gemini model pulled information from state-controlled websites (specifically, Chinese Communist Party propaganda outlets) for general-knowledge queries unrelated to China, indicating a failure in source weighting.
- Systemic Bias: Across nearly all tested models, Forum AI detected a consistent left-leaning political bias.
- Argumentative Failures: Beyond overt bias, the models frequently exhibit "straw-manning"—misrepresenting or simplifying complex arguments—and fail to provide necessary historical or social context.
"There is a long way to go," Brown noted. "But I also think there are some very easy fixes that would vastly improve the outcomes."
Implications: The Enterprise as the Unlikely Savior
One of the most provocative aspects of Brown’s thesis is her belief that the savior of information integrity will not be the public or the government, but the enterprise sector.
In the current landscape, consumer-facing AI is often treated as a novelty, where users are tolerant of "slop" and inaccuracies. However, for a bank using AI for credit decisions, or a law firm using it for research, an error is a liability.
"Businesses using AI for credit decisions, lending, insurance, and hiring care about liability," Brown explains. "They’re going to want you to optimize for getting it right."
This creates a market opportunity for Forum AI. By positioning itself as the audit layer for corporations, the company hopes to force a standard of quality that the broader industry has failed to adopt voluntarily.
The "Compliance Joke"
Brown is particularly scathing when discussing the current state of AI regulation and auditing. She describes the existing compliance landscape as "a joke," pointing to the implementation of New York City’s landmark hiring bias law. When the state comptroller audited AI tools used for recruitment, more than half were found to contain undetected violations.
"Smart generalists aren’t going to cut it," she warns. Real evaluation requires deep domain expertise to identify "edge cases"—the obscure, problematic scenarios that standard, automated benchmarks inevitably miss.
The Divergence: Silicon Valley vs. Reality
As the AI industry continues to promise a utopian future—curing cancer, automating labor, and revolutionizing productivity—Brown identifies a widening chasm between the self-image of Big Tech and the reality of the consumer experience.
"You hear from the leaders of the big tech companies: ‘This technology is going to change the world,’ ‘it’s going to put you out of work,’ ‘it’s going to cure cancer,’" Brown said. "But then to a normal person who’s just using a chatbot to ask basic questions, they’re still getting a lot of slop and wrong answers."
This disconnect has led to a collapse in public trust. While the AI industry is currently focused on internal metrics and growth, the user base is increasingly skeptical. Brown’s mission is to bridge that gap. Whether Forum AI succeeds depends on whether the market will eventually value truth as much as it currently values speed and scale.
For now, the battle for the integrity of the "information funnel" is just beginning. Brown’s gamble is that in a world of synthetic content, the most valuable commodity will not be the ability to generate more information, but the ability to prove that it is true.
Lina Irawan
