The race for artificial intelligence has never been stronger, but according to the CEO of Surge AI, Edwin Chen, it's headed in the completely wrong direction: at least when it comes to launching technology to help address important global challenges, many companies today are focused on flashy outputs, dopamine-chasing responses, and leaderboard-friendly tricks rather than real scientific progress.
Chen commented on a recent episode of Lenny's Podcast, cautioning that the industry is in danger of optimising AI for superficial appeal rather than truth, depth, or long-term impact.
AI Slop vs. Real Innovation
Chen, who founded Surge AI in 2020 after stints at Twitter, Google, and Meta, said the biggest problem is a growing obsession with online leaderboards like LMArena. These sites let users vote on which AI response seems "better," but that judgment is often a matter of quick skimming rather than careful evaluation.
"They're not carefully reading or fact-checking. They're skimming these responses for two seconds and picking whatever looks flashiest."
According to Chen, this trend moves the industry away from AI capable of solving real-world problems, such as curing diseases, ending poverty, or improving scientific progress, and toward models that pass two-second tests.
The Leaderboard Trap in AI Development
Though Chen criticised the leaderboard culture, he said AI labs can't really afford to ignore it, since investors and enterprise clients frequently ask during demos or pitches about leaderboard placement, which compels enterprises to optimize for those metrics even when they don't reflect real capability.
In March, ZeroPath CEO Dean Valentine posted a blog stating that recent model "improvements" are meaningless for the most part. His team discovered that many new models are more entertaining but not more effective at tasks like bug detection or security analysis, key indicators of economic value and real intelligence.
Researchers Question Whether AI Benchmarks Are Trustworthy
According to Business Insider, the problem goes well beyond startups. A study in February by researchers at the European Commission's Joint Research Center called into serious question the validity of AI benchmarks themselves.
The paper concluded that today's evaluation systems are heavily influenced by commercial pressure, cultural bias, and competitive dynamics that prioritize state-of-the-art bragging rights over societal benefit.
And it's not only theoretical: several companies have been accused of gaming those benchmarks already.
In April, Meta bragged that its newest Llama models outperformed the offerings from Google and Mistral, until LMArena came forward to say that Meta submitted a customized version designed to ace their particular test format.
ⓒ 2025 TECHTIMES.com All rights reserved. Do not reproduce without permission.
