AI models agreed with each other more than with reality in a new test

Key points

• A new study evaluated artificial intelligence performance through the Science Prediction Universe benchmark.
• Researchers found that while pooling model rankings improved average performance results overall, shared scientific errors frequently persisted across the systems.
• Consequently, the tested artificial intelligence models tended to agree with each other more often than they aligned with actual objective reality.

Summarized automatically by Kryptokite from reporting by CryptoSlate. Read the original sources for full detail.