Title: OpenAI’s GPT-5: The Next Breakthrough in AI?
In a recent podcast, speculation arose that OpenAI’s Code Interpreter may actually be built on GPT-4.5, causing some to believe that the company is intentionally keeping this information quiet. Just months ago, AI experts called for a halt to OpenAI’s AI experiments after the release of GPT-4, fearing that the company’s rapid advancements were becoming too difficult to keep up with. However, the tides may now be turning against OpenAI as competition from other startups and companies looms large.
One such competitor, Anthropic, recently introduced Claude-2, which in many cases outperforms OpenAI’s GPT-4. Social media platforms have been buzzing with users showcasing the capabilities of Claude-2, particularly its ability to write code and interpret documents. Moreover, Anthropic’s offering boasts a large context length of 100K, surpassing GPT-4’s 32K, making it both more affordable and suitable for a wider range of applications.
While OpenAI has long been the dominant force in the AI landscape, many others have been trying to replicate their success. However, it seems that OpenAI has decided to let its competitors catch up. Following the pause letter from influential figures such as Elon Musk and Steve Wozniak, OpenAI’s CEO, Sam Altman, announced that the company would delay training the successor to GPT-4 for the time being. Altman cited the need for the company to focus on other projects before diving into building the new model.
Altman revealed during a visit to India that it took over six months to develop and release GPT-4, indicating that the company was already training the model while they were preparing to release ChatGPT based on GPT-3.5. After OpenAI released ChatGPT in November 2022, it took an additional five months for GPT-4 to be unveiled. Since then, OpenAI has only made incremental updates to its technology, with nothing groundbreaking to showcase.
The approach of incremental software updates is a shared philosophy between Altman and OpenAI co-founder Greg Brockman. Altman explained during the Lex Fridman podcast that despite the potential disagreement from some, releasing AI models incrementally helps maintain control over AI. Similarly, in April, Brockman tweeted about the pace of AI progress, calling attention to the possibilities of creating newer, incrementally better AIs rather than infrequent major upgrades.
Certain claims speculate that the next iteration of GPT will be superintelligent, with OpenAI suggesting in a blog post on AI model alignment that this level of intelligence may be achieved within the next four years. However, the given timeline does not align with the need for OpenAI to compete with the increasing capabilities of rival AI models. If OpenAI intends to keep up with its competitors, it may need to consider lifting the halt on training GPT-5 and proceed with building it immediately.
Interestingly, the introduction of Code Interpreter aligns with the significant leap in capabilities observed between GPT-3.5 and GPT-4, leading some to argue that Code Interpreter is, in fact, GPT-4.5. It’s possible that OpenAI is currently training GPT-4 with different modalities, similar to their approach in transitioning from GPT-3 to GPT-3.5. The introduction of plugins, an API, and multimodality all serve as hints towards the eventual training of GPT-5.
As OpenAI navigates the rapidly evolving AI landscape, it will be interesting to see how the company adapts to the ever-increasing competition. With the possibility of GPT-5 on the horizon, AI enthusiasts and industry experts eagerly await OpenAI’s next move.
