OpenAI may be launching the successor to GPT-4 by the middle of this year. Whether this will be the long-awaited GPT-5 or perhaps an improved version of GPT-4.5 Turbo remains to be seen.
The successor to OpenAI’s GPT-4 LLM -the engine behind ChatGPT- will be presented mid-year. This reports Business Insider, based on sources within the AI developer.
The new LLM would be much stronger than the last one and have new powerful features. CEOs of companies that have already seen a test version of the new LLM speak of built-in AI agents that perform tasks in an automated fashion, among other things. This might include improvements to the ChatGPT application.
The new LLM would additionally support prompts with text or images, as with GPT-4. It is also a next-token prediction model that generates the best output for the next fragment or word. This makes the model suitable for tasks such as composing sentences or writing code.
‘A special new LLM’
In response to ArsTechnica about the rumours of the arrival of the successor to GPT-4, OpenAI officially does not comment. However, spokespersons referred to a podcast in which CEO Sam Altman may hint at what is to come.
In the podcast, Altman says he does not know when GPT-5 will be released. He does announce that Open AI is introducing a “very special new LLM” this year. The CEO continues that the AI giant will release a lot of new stuff in the coming months. He indicates many other important things must be released in advance of any new version of GPT.
The podcast in question. At 01:06:13 is the excerpt.
More specifically, he indicates that OpenAI does not want to release “shock updates” to its LLMs because users, such as companies, need more time to adjust to using the OpenAI LLMs. Therefore, GPT-5 may be released in “a different way.”
This cryptic description causes uncertainty about whether OpenAI’s new LLM will actually be named GPT-5. Other options include a significant update to the latest and most powerful GPT-4 Turbo version or an incremental update such as GPT-4.5.
Training still underway
According to the rumor mill, the new OpenAI model, possibly meaning GPT-5, is still undergoing (re)training. Once the LLM training is completed, the team will first test it internally. This will be followed by tests by a select number of external users for fine-tuning and fixing any remaining issues—all this before the public release.
Also read: Confusion about training data model Sora that generates videos
17 hours ago
