Campbell Brown has dedicated her career to pursuing reliable information, starting as a prominent TV journalist and later serving as Facebook's inaugural - and sole - news executive. Today, as AI transforms information consumption, she warns of history repeating unless action is taken. This time, she is taking charge.

At her startup, Forum AI - which she covered in a recent TechCrunch discussion with Tim Fernholz at a StrictlyVC event in San Francisco - the focus is assessing foundation models on "high-stakes topics" like geopolitics, mental health, finance, and hiring. These are areas "lacking simple yes-or-no solutions, filled with nuance and complexity."

The approach involves assembling top experts to design benchmarks, then developing AI judges to score models efficiently. For geopolitics, Brown enlisted Niall Ferguson, Fareed Zakaria, ex-Secretary of State Tony Blinken, former House Speaker Kevin McCarthy, and Anne Neuberger, ex-Obama cybersecurity head. Forum AI aims for AI judges matching human experts at about 90% agreement, a mark they have achieved.

Brown links Forum AI's start, 17 months ago in New York, to a pivotal instant. "I was still at Meta when ChatGPT launched publicly," she said, "and soon after, I saw it becoming the main channel for all information - yet it was subpar." The stakes felt personal for her family. "If we don't address this, my kids will lack critical thinking," she thought.

Her biggest gripe: no one prioritized accuracy. She notes foundation model firms emphasize "coding and math," sidelining tougher news and info challenges. Yet she insists difficulty cannot justify neglect.

Forum AI's early tests of top models revealed stark shortcomings. Brown highlighted Gemini citing Chinese Communist Party sites "irrelevant to the story," plus widespread left-leaning bias. Other issues include overlooked context, ignored viewpoints, and misrepresented arguments. "Progress is needed," she said, "but simple tweaks could yield big gains."

Having witnessed Facebook's missteps during her tenure, Brown knows the risks of poor optimization. "We fell short on many initiatives," she told Fernholz. Her fact-checking effort is gone. The key takeaway - ignored by social platforms - is that engagement-driven design harms society and erodes knowledge.

She hopes AI avoids this trap. "The path is uncertain," she noted; firms might chase user preferences or prioritize "truth, honesty, and facts." The truth-focused ideal may seem optimistic, but enterprise users - in credit, lending, insurance, hiring - demand accuracy to dodge risks and "insist on correctness."

Forum AI targets this enterprise need for revenue, though converting interest into steady income is tough amid reliance on superficial audits and generic benchmarks Brown deems insufficient.

The compliance field, she calls, "a farce." New York City's pioneering AI hiring bias law saw the state comptroller uncover over half of audits missing violations. True checks demand specialists tackling not just standards but rare cases "that spell trouble unexpectedly." This requires effort. "Generalists won't suffice."

Brown - whose firm secured $3 million last fall from Lerer Hippeau - astutely highlights the gap between AI hype and user reality. "Tech giants proclaim world-altering tech, job losses, cancer cures," she said. "But everyday chatbot users get junk and errors."

AI trust is dismal, and her view is that doubt is often warranted. "Silicon Valley is debating one topic, while consumers are focused on something completely different."