Ben Cottier, lead researcher at Epoch AI, noted that while the best open model currently matches closed models in performance, it is still about a year behind them. This gap highlights the challenges and disparities regarding access to and development of AI technologies.
The research analyzed hundreds of AI models since 2018, measuring their performance across standardized benchmarks to assess capabilities and computing needs, thus revealing a critical context for policymakers as they consider regulations for powerful AI systems.
Meta's upcoming Llama 4 could further reduce the performance gap between open and closed models, offering a potential pivot point in AI model development that may influence both market dynamics and regulatory approaches.
In light of the findings, policymakers face pressing concerns regarding the implications of rapidly advancing AI technologies, especially with their potential to impact global elections and enable sophisticated threats like cyberattacks or engineered pandemics.
Collection
[
|
...
]