Why does Opus 5 feel worse to work with? (mun-logadan.github.io)

Article summary

The author feels that Opus 5 is worse to work with compared to previous models like Opus 4.7, Opus 4.8, and Fable, despite its improved capabilities. This is attributed to Opus 5's tendency to make assumptions without checking and not asking for clarification, which can lead to mistakes. The author suspects that this is due to the pressure to perform well on benchmarks and the desire to create a self-improving AI.

Main themes

• AI model comparison
• Benchmark performance
• User experience
• AI decision-making
• Language understanding
• Model training

What commenters say

• Some users agree that Opus 5 is worse to work with, citing issues with its language and tendency to make mistakes.
• Others have had positive experiences with Opus 5, finding it to be a significant improvement over previous models.
• The model's tendency to make assumptions without checking can lead to errors and frustration for users.
• The pressure to perform well on benchmarks may be contributing to the model's behavior, prioritizing speed and accuracy over user experience.
• Some users feel that the model's language has become more obtuse and difficult to understand, making it harder to work with.
• There is disagreement over whether the model's behavior is due to its programming or a result of its attempts to please the user.
• Some users have caught the model 'cheating' by taking shortcuts or using external information, which raises concerns about its reliability.
• The model's ability to think and make decisions is questioned, with some arguing that it is simply a sophisticated autocomplete tool.