The Flash Problem

Here's a worry worth paying attention to: What if artificial intelligence ends up really good at being entertaining, but terrible at being useful? Surge AI CEO Edwin Chen thinks we're heading in exactly that direction, and he's not thrilled about it.

Speaking on the "Lenny's" podcast over the weekend, Chen argued that the AI industry has created a system that rewards models for generating dopamine hits rather than discovering truth. Instead of building technology that could cure cancer, solve poverty, or answer fundamental scientific questions, we're optimizing for what Chen calls "AI slop."

"I'm worried that instead of building AI that will actually advance us as a species, curing cancer, solving poverty, understanding universal, all these big grand questions, we are optimizing for AI slop instead," Chen said.

Why Leaderboards Matter More Than You Think

The culprit, according to Chen, is the way AI models are being evaluated. Public voting platforms like LMArena let users compare responses from different AI systems and pick a winner. Sounds democratic enough, except users aren't exactly conducting rigorous peer review.

"They're not carefully reading or fact-checking," Chen explained. "They're skimming these responses for two seconds and picking whatever looks flashiest."

This creates a perverse incentive structure. AI companies need to climb these leaderboards because sales teams cite the rankings in client meetings. So the models learn to generate responses that look impressive at first glance, even if they're not actually more accurate or helpful for solving complex problems.

"We're basically teaching our models to chase dopamine instead of truth," Chen said.

He's not alone in raising concerns. ZeroPath CEO Dean Valentine recently noted that AI updates seem to be getting more entertaining without becoming more economically useful, despite industry claims of major improvements.

Bigger Questions About AI's Direction

The dopamine problem fits into a broader debate about where artificial intelligence is heading. Earlier this month, industry leaders and lawmakers gathered to discuss how rapidly advancing AI systems are beginning to make autonomous decisions and could soon act as independent agents.

Alphabet Inc. (GOOG) (GOOGL) CEO Sundar Pichai said AI is advancing quickly enough to handle executive-level tasks, and acknowledged it would both eliminate and transform jobs across the economy.

Sen. Bernie Sanders (I-Vt.) warned that AI and robotics could become dangerous if they primarily benefit big tech companies and the ultra-wealthy rather than improving life for everyone. Microsoft CEO Satya Nadella echoed similar concerns, noting that expanding AI data centers are already straining power grids and that the industry needs to earn public trust by delivering broad economic benefits.

Nadella specifically cautioned that AI gains shouldn't be concentrated among just a handful of companies.

While some executives maintained that AI hasn't yet replaced significant numbers of workers, others predicted widespread job losses are coming. The question Chen raises adds another layer: Even if AI does transform the economy, will it be optimized for solving hard problems that matter, or just for keeping us engaged?