OpenAI Acknowledges GPT-4's Laziness, Upcoming Fix Unveiled for Enhanced Performance

OpenAI was in the news for much of November after CEO Sam Altman was sacked and rehired. And while that may have been a short setback, OpenAI is still trying to stay caught up, especially with Google bringing the big guns with the Gemini 1.0 AI model family. It is clear from the company's admission that GPT-4 LLM, which powers the ChatGPT Plus Chatbot, has become "lazier."

"We've heard all of your complaints about GPT-4 being lazier! "We haven't updated the model since 11 November, and this isn't by design," OpenAI wrote on X (previously known as Twitter).

OpenAI Acknowledges the Problem

The statement said, "Model behaviour can be unpredictable, and we're looking into fixing it." Responding to the criticism, OpenAI has acknowledged the performance issues with GPT-4. They attribute the decline to a specific factor: a change in the training data used to fine-tune the model. While intended to improve accuracy and reduce bias, this change appears to have inadvertently led to the observed laziness.

For the uninitiated, users of GPT-4 have recently complained that OpenAI has damaged the experience and that the model is nearly "fearful" of providing answers. Others stated that they needed to be more precise with their prompts, which was previously optional.

In its defence, OpenAI claims that the model has stayed the same since 12 November (the last time it was updated). However, the problem is that "differences in model behaviour can be subtle," and only specific prompts may be "degraded." Despite these modifications, OpenAI cautions that users and employees may take time to recognise these patterns.

Timeline for the Fix

OpenAI still needs to provide a precise timeline for implementing the fix. However, they have stated that it is a high priority and are working diligently to resolve the issue. OpenAI has not specified a release date for the new, updated model, but it will undoubtedly be watching competitors such as Google with the launching of the Gemini 1.0 family.

Impact on Future AI Development

The challenges faced with GPT-4 highlight the complexities of developing and deploying large language models. OpenAI's transparency and commitment to addressing the problem offer valuable insights for the future of AI development.