The rapid proliferation of generative artificial intelligence has created a paradoxical landscape where technological capabilities are advancing at breakneck speed while public trust in the accuracy of digital information remains at an all-time low. Campbell Brown, a figure who has occupied the front lines of the information wars for decades, is now positioning herself at the center of this transition. As the founder of Forum AI, Brown is attempting to solve a problem that she argues the tech industry has largely ignored: the "slop" and inaccuracy inherent in how large language models (LLMs) handle complex, high-stakes information. Speaking recently with TechCrunch’s Tim Fernholz at a StrictlyVC event in San Francisco, Brown detailed her mission to bridge the gap between Silicon Valley’s technical ambitions and the nuanced reality of human expertise.
The Genesis of Forum AI: From Meta to the Frontier of Machine Learning
Campbell Brown’s career trajectory provides a unique vantage point on the evolution of information dissemination. After a high-profile career as a television journalist at NBC and CNN, she transitioned into the tech sector, serving as the first and only dedicated news chief at Facebook (now Meta). During her tenure at Meta, Brown witnessed firsthand the systemic challenges of managing news on a platform optimized for engagement rather than accuracy. She led initiatives like the Facebook Journalism Project and developed fact-checking programs, many of which have since been scaled back or discontinued.
The catalyst for Forum AI occurred approximately 17 months ago, during the final stages of Brown’s time at Meta. The public release of ChatGPT by OpenAI served as a wake-up call. Brown recalled the realization that AI would soon become the primary "funnel" through which the global population consumes information. However, upon testing the technology, she found it lacking in the depth and accuracy required for meaningful discourse. Her concern was not merely professional but personal; she expressed a fear that if the quality of information provided by these models did not improve, the next generation would suffer from a fundamental lack of critical knowledge.
Founded in New York, Forum AI was born out of the necessity to create a more rigorous evaluation framework for AI. While foundation model companies like OpenAI, Google, and Anthropic have focused heavily on optimizing their models for mathematics and coding—areas with objective, binary answers—Brown argues that they have neglected "high-stakes" topics. These include geopolitics, mental health, finance, and hiring—subjects characterized by nuance, complexity, and a lack of simple "yes-or-no" answers.
A Methodology Rooted in Human Expertise
The core innovation of Forum AI lies in its approach to benchmarking. Traditional AI benchmarks often rely on standardized tests or automated "vibe checks" that fail to capture the subtleties of expert-level knowledge. To address this, Brown has recruited a roster of world-renowned experts to architect specific benchmarks for the company.
The list of contributors is a "who’s who" of global policy and intellectual leadership. For its geopolitics vertical, Forum AI has engaged figures such as Niall Ferguson, Fareed Zakaria, former Secretary of State Tony Blinken, former House Speaker Kevin McCarthy, and Anne Neuberger, the former Deputy National Security Advisor for Cyber and Emerging Technology. These experts provide the "gold standard" for what an accurate, nuanced response should look like in their respective fields.
The process involves several distinct steps:
- Expert Architecture: High-level experts define the parameters of a "good" answer for a complex query.
- Benchmark Creation: These parameters are turned into rigorous benchmarks that go beyond simple factual accuracy to include context, perspective, and the avoidance of logical fallacies.
- AI Judge Training: Forum AI trains specialized AI "judges" to evaluate the performance of foundation models at scale, using the expert-defined benchmarks as the training set.
- Consensus Achievement: Brown reports that Forum AI has reached a threshold where its AI judges achieve roughly 90% consensus with the human experts.
This methodology aims to solve the scalability problem. While a single human expert cannot review millions of AI-generated responses, an AI judge trained on that expert’s logic and nuance can perform evaluations across massive datasets in real-time.
Analyzing the Failures of Current Foundation Models
When Forum AI began applying its evaluation tools to leading models like Google’s Gemini or OpenAI’s GPT series, the results highlighted significant deficiencies. Brown pointed to instances where Google’s Gemini pulled information from Chinese Communist Party (CCP) websites for stories entirely unrelated to China, suggesting a failure in source prioritization and contextual relevance.
Furthermore, Forum AI’s evaluations have identified a persistent left-leaning political bias across nearly all major models. Beyond political leanings, Brown noted subtler but equally damaging failures:
- Missing Context: Models often provide facts without the necessary background to make them meaningful.
- Perspective Gaps: Complex issues are often presented from a singular viewpoint, ignoring valid counter-arguments or alternative interpretations.
- Straw-manning: AI frequently simplifies opposing arguments to the point of absurdity without acknowledging the complexity of the debate.
Brown contends that while these issues are significant, many are "easy fixes" if accuracy becomes a prioritized metric for developers. The challenge, however, is that the current incentive structure in Silicon Valley often favors speed of deployment and user engagement over editorial rigor.
The Shift from Engagement to Truth: Lessons from the Social Media Era
Brown’s perspective is heavily influenced by her years at Facebook, a period she describes as a time of both ambitious experimentation and significant failure. "We failed at a lot of the things we tried," she admitted during the San Francisco event. She specifically cited the fact-checking programs she helped build, which have largely been dismantled as social media platforms pivoted away from the news business.
The central lesson from the social media era, according to Brown, is the danger of optimizing for engagement. When platforms prioritize what keeps users clicking, the result is often a "race to the bottom" that favors sensationalism and misinformation. Brown sees the AI revolution as an opportunity to break this cycle.
"Right now it could go either way," Brown said. AI companies can choose to give users "what they want"—which often means reinforcing existing biases—or they can "give people what’s real and what’s honest and what’s truthful." While she acknowledged that optimizing for truth might sound idealistic or even naive, she believes there is a practical, market-driven reason for optimism: the enterprise sector.
The Enterprise Market: A Catalyst for Accuracy
While individual consumers might be satisfied with a chatbot that tells them what they want to hear, businesses cannot afford such a luxury. Companies using AI for high-stakes decisions—such as insurance underwriting, credit lending, legal research, and hiring—face significant legal and financial liability if those models provide incorrect or biased information.
Forum AI is betting its business model on this corporate demand for reliability. In the fall of last year, the company raised $3 million in a seed funding round led by Lerer Hippeau. This capital is being used to build out the compliance and evaluation tools that businesses require to de-risk their AI implementations.
However, the path to consistent revenue is not without obstacles. Brown noted that the current compliance landscape is "a joke," characterized by "checkbox audits" and standardized benchmarks that provide a false sense of security. She cited New York City’s pioneering law requiring AI audits for hiring tools as a prime example of the current system’s inadequacy. A report from the state comptroller found that more than half of the audited companies had violations that went undetected by standard review processes.
"Smart generalists aren’t going to cut it," Brown asserted. Real evaluation requires domain expertise to navigate not just common scenarios but "edge cases" that can lead to catastrophic failures in a corporate or legal context.
Bridging the Trust Gap: Silicon Valley vs. The Consumer
The disconnect between the rhetoric of tech leaders and the experience of everyday users is a central theme of Brown’s critique. While CEOs of major tech firms promise that AI will cure cancer, solve climate change, and revolutionize the workforce, the average consumer often encounters "slop"—incorrect answers, hallucinations, and unhelpful responses to basic queries.
This disparity has led to extraordinarily low levels of public trust in AI. Brown argues that this skepticism is justified. While Silicon Valley is having a conversation about the existential risks and utopian possibilities of Artificial General Intelligence (AGI), the public is struggling with a technology that often fails to provide a reliable answer to a simple question about history or current events.
Forum AI’s mission is to bring a level of professional accountability to the AI space that has been missing since the dawn of the social media era. By leveraging the insights of seasoned experts and translating those insights into scalable AI judges, Brown hopes to ensure that the "funnel" through which information flows remains clear, accurate, and grounded in reality.
Implications and Future Outlook
The success or failure of Forum AI may serve as a bellwether for the broader AI industry. If the company can prove that expert-led benchmarking is both technically feasible and commercially viable, it could force foundation model companies to elevate their standards.
The implications are particularly profound for the upcoming global election cycles. With AI-generated content expected to play a major role in political discourse, the need for independent, expert-backed evaluation tools has never been more urgent. If AI models continue to pull from unreliable sources or exhibit unacknowledged biases, the potential for societal destabilization is significant.
Furthermore, the focus on "high-stakes" topics like mental health and finance highlights the human cost of AI inaccuracy. A model giving poor financial advice or incorrect mental health guidance can have devastating real-world consequences. By shifting the focus from "engagement" to "getting it right," Campbell Brown and Forum AI are attempting to steer the next wave of technological innovation toward a more responsible and truthful future.
As the industry moves forward, the tension between the "move fast and break things" ethos and the need for editorial integrity will remain. Brown, having seen the consequences of that tension from the inside of one of the world’s largest tech giants, is now betting that the world is finally ready to prioritize the truth.
