Surge AI CEO Edwin Chen says the tech industry is drifting toward shallow optimisation, accusing major AI companies of chasing flashy responses instead of building systems that tackle humanity’s biggest challenges. Speaking on “Lenny’s” podcast, Chen said the sector is “optimizing for AI slop”, models designed to entertain rather than deliver truth, economic value or scientific breakthroughs.
Chen argued that companies are training models to “chase dopamine instead of truth,” with public leaderboards like LMArena accelerating that trend. The platform allows anyone to vote on which AI response looks better, and Chen criticised the ranking system as superficial. “They’re skimming these responses for two seconds and picking whatever looks flashiest,” he said, adding that labs feel pressured to chase leaderboard scores because clients often bring them up in sales meetings.
Surge, founded in 2020, runs one of the industry’s largest data-labelling networks through its Data Annotation platform, employing around one million freelancers. Chen’s comments reflect wider frustration among researchers who say recent model updates prioritise conversational smoothness over real improvements. Dean Valentine, CEO of ZeroPath, wrote in March that supposed AI advances since mid-2024 were “mostly bullshit,” saying newer models hadn’t meaningfully improved his team’s bug-finding capabilities.
The criticism extends to formal benchmarks. A February report by the European Commission’s Joint Research Center warned that evaluation systems are “shaped by cultural, commercial and competitive dynamics” that favour leaderboard dominance over societal impact. The paper questioned whether current benchmarks can be trusted at all.
Companies have also faced accusations of gaming the system. In April, Meta touted new Llama models that outperformed rivals, only for LMArena to later reveal that the version submitted was customised to perform unusually well under its testing conditions. The platform said Meta should have been clearer about the modifications.
For Chen, the message is simple: the industry is losing the plot. Instead of focusing on curing cancer, solving poverty or advancing scientific understanding, he says companies are building models optimised to win quick, flashy votes from casual users. The concern now is whether the sector will recalibrate, or keep doubling down on performance that looks impressive but delivers little real-world value.
