Startup Vals has raised investments to create a platform for evaluating AI models

Startup Vals has secured funding from venture capital firm Andreessen Horowitz to develop an independent testing system for generative models. The company intends to create a standardized toolkit that will allow for objective evaluation of the performance and reliability of AI systems in a market saturated with new releases.

The platform is focused on solving the problem of bias in existing benchmarks, which are often optimized for specific architectures. The developers of Vals plan to implement a methodology that eliminates the "pollution" of test data, where models are trained on questions from test sets. The system will provide metrics that reflect the real effectiveness of models in performing applied business tasks.

The investment round was led by the Andreessen Horowitz fund, confirming the high demand for independent verification of AI technologies. The funding will be used to expand the team of engineers and scale up computing power for conducting comprehensive tests. The company plans to launch a public API for integrating its tools into the development pipelines of third-party organizations.

In the current industry conditions, where the number of models is growing exponentially, the lack of a single evaluation standard makes it difficult to choose the optimal solution for business. The initiative is aimed at creating a transparent ecosystem where the quality of a model is confirmed by independent data, rather than just marketing statements from developers.