Ask HN: Is it just me or GPT-4's quality has significantly deteriorated lately?

Article summary

The author of the post claims that the quality of GPT-4's responses has significantly deteriorated lately, despite being faster than before. They notice that it generates more buggy code and provides less in-depth analysis. The author wonders if using a different version of GPT-4, such as Poe's, would yield better results. The post also mentions that another GPT-4 powered search engine, phind.com, has switched to using GPT-3.5 and its own model more frequently.

Main themes

• GPT-4 performance decline
• Experimentation and changes
• Novelty effect
• Alternative AI models
• Safety fine-tuning
• Lack of transparency

What commenters say

• Some users have noticed a decline in GPT-4's performance, with it providing less accurate and less detailed responses.
• The decline in performance may be due to experimentation or changes made by the developers, such as reducing compute per request.
• Others argue that the perceived decline in performance is due to the novelty effect wearing off and users having higher expectations.
• A few users have reported that they have not noticed any decline in performance and that GPT-4 still provides high-quality responses.
• Some users have switched to using alternative AI models, such as Bard, due to their improved performance.
• The quality of GPT-4's responses may have been affected by the focus on safety fine-tuning, which could have led to a decrease in its capabilities.
• The lack of transparency from the developers about any changes or experimentation makes it difficult for users to understand what is happening with the model.