OpenAI Acknowledges GPT-4’s Decreased Efficiency Due to a Specific Cause; Solution in the Works
OpenAI was in the news for most of November when CEO Sam Altman was fired and then rehired. And while it may have been a temporary bad look, OpenAI isn’t trying to lag behind its competitors, especially as Google brings out the big guns in the Gemini 1.0 AI model family. This is reflected in the fact that the company has admitted that the GPT-4 LLM using the ChatGPT Plus Chatbot has become “lazier”.
“We’ve heard all your feedback about GPT-4 being lazy! We haven’t updated the model since November 11th, and this is definitely not on purpose,” OpenAI posted on X (formerly Twitter).
It further added: “The behavior of the model can be unpredictable and we are working to fix it.”
For the uninitiated, users using GPT-4 have recently suggested that OpenAI has degraded the experience and that the model is almost “afraid” to provide answers. Some others said they need to be more specific in their prompts, which wasn’t needed before.
In its defense, OpenAI further notes that the model has not “changed in any way” since November 12th (last updated). But the problem lies in the fact that “differences in model behavior can be subtle” and that only certain prompts might be “degraded”. Despite these changes, OpenAI notes that it may take some time for users and workers to notice these patterns.
OpenAI has not promised a set deadline for the release of a new, updated model, but it will certainly follow competitors such as Google by releasing the Gemini 1.0 family.
