There is a company called Vals AI, backed by Andreessen Horowitz, that wants to be the gold standard for testing how good AI models actually are. The pitch is neutrality: independent benchmarks, rigorous methodology, no thumb on the scale. This arrives at the precise moment when the word "benchmark" has been so thoroughly gamed by AI companies that it functions as marketing vocabulary rather than scientific measurement. And into that vacuum steps a firm funded by one of the most ideologically legible venture firms in technology. The irony is not subtle.
Who Defines the Gold Standard?
Benchmarking is not a neutral act. What you test determines what gets optimized, and what gets optimized determines what gets built. A 2022 paper in Proceedings of the ACM on Human-Computer Interaction by Raji et al., titled "AI and the Everything in the Whole Wide World Benchmark," argued that AI benchmarks systematically embed the values and assumptions of their creators, producing a feedback loop where the test and the system co-evolve toward each other rather than toward real-world usefulness. Vals is trying to break that loop, which is genuinely valuable work. But being backed by a16z, whose portfolio is dense with the AI companies being benchmarked, puts the independence claim under strain before the first test is run. TurboFund's live investor signals track exactly this kind of infrastructure bet, where the real power move is owning the measurement layer, not the product.
The Inigo Philbrick Problem
The art world has been here before. When Inigo Philbrick was running his fraud, the market's valuation infrastructure, the auction estimates, the private sales, the advisory opinions, were all products of parties with something to gain from the numbers they produced. AI benchmarking in 2026 looks structurally similar: a market where the entities being evaluated have deep financial relationships with the entities doing the evaluating. The scandal has not happened yet. But the architecture for one is fully in place. Neutrality is not a feature you declare. It is a condition you earn, slowly, in public, under adversarial scrutiny. Vals has the funding. It does not yet have the track record.