The Race for Innovation: OpenAI's Urgency to Launch GPT-5

A recent podcaster argued that Code Interpreter is actually built on GPT-4.5, suggesting that OpenAI is being coy about it.

Days after the release of GPT-4 in March, AI experts rallied against OpenAI, urging a pause in giant AI experiments, including training models beyond the company's latest release. These experts might have been concerned about the potential risks of AI or simply wanted to slow down the progress of the Microsoft-backed AI company, which was becoming increasingly hard for them to keep up with. Interestingly, the tide may now be turning against OpenAI.

OpenAI is facing strong competition. While the founders believe that no one can replicate what the company has built, as they mentioned during their visit to Israel, numerous startups and companies have the potential to lure away OpenAI's customers. A notable example is Anthropic, which recently introduced Claude-2.

Read: Claude-2 vs GPT-4

When comparing the capabilities of GPT-4 and Claude-2, Anthropic's offerings often outperform OpenAI's. Twitter is buzzing with users leveraging Claude-2's language model capabilities to write code and make sense of uploaded documents. The large context length of 100K, compared to GPT-4's 32K, along with a more affordable price, makes Claude-2 appealing to many companies and their specific use cases.

OpenAI wasted enough time already

The company, with GPT, had managed to outshine other companies and had everyone scrambling to replicate its success. However, it seems that OpenAI decided to let them catch up.

Following the pause letter from industry giants like Elon Musk and Steve Wozniak, Altman, OpenAI's CEO, announced that the successor to GPT-4 would not be trained "for some time." Altman cited the company's need to focus on other tasks before starting to develop the new model. That was back in April. Fast forward to June, Altman stated that OpenAI had not yet begun training GPT-5 and would instead concentrate on exploring new ideas. Even Musk has jumped into the competition with his own AI company, xAI, to challenge OpenAI.

"When we finished GPT-4, it took us more than six months until we were ready to release it," Altman shared with the Economic Times during his visit to India. This suggests that the company had already started training GPT-4 while they were about to release ChatGPT, based on GPT-3.5.

After releasing ChatGPT, based on GPT-3.5, in November 2022, it took OpenAI an additional five months to release GPT-4. Since then, OpenAI has introduced incremental updates to their technology, but nothing groundbreaking.

Interestingly, the incremental approach to software updates is something both founders, Altman and Greg Brockman, agree on. During the Lex Fridman podcast, Altman defended the company's approach, acknowledging that not everyone might agree with it but believing that incremental releases of AI models help maintain control over AI.

Similarly, in April, Brockman tweeted his views on the pace of AI progress, suggesting that creating a continuum of incrementally better AI models could enhance safety, unlike the company's previous approach of infrequent major upgrades.

GPT-4.5 is already here!

Speculations are rife about the existence of GPT-4.5, with claims that the next iteration of GPT could achieve superintelligence. OpenAI's recent blog on AI model alignment hinted at the potential of reaching superintelligence within four years. However, the company's timeline doesn't align with this claim. To compete with the advancing capabilities of rival AI models, OpenAI might need to reconsider the pause on training GPT-5 and expedite its development.

Furthermore, the introduction of Code Interpreter stands out as a significant advancement comparable to the leap from GPT-3.5 to GPT-4. In a thought-provoking Latent Space podcast, experts such as Simon Willison, Alex Volkov, Aravind Srinivas, and Alex Graveley posit that Code Interpreter could be the elusive GPT-4.5.

OpenAI may have avoided using this terminology to avoid potential backlash after the pause letter controversy.

It is plausible that OpenAI is currently exploring diverse modalities to train GPT-4, akin to their approach in transitioning from GPT-3 to GPT-3.5. The introduction of plugins, API, and multimodality suggests preparations for GPT-5 training. Furthermore, with Musk's xAI entering the scene, OpenAI must act swiftly to maintain its competitive edge in the field.