During the World Governments Summit in Dubai in February, OpenAI’s CEO shared insights into the upcoming GPT-5, highlighting its anticipated intelligence leap compared to its predecessors.
Describing the significance, Mr. Altman emphasized the inherent versatility of these models, stating that even a slight improvement in intelligence enhances overall performance across various tasks.
Mr. Altman hinted at the expanded capabilities of GPT-5, indicating its reliance on more extensive datasets for training. The strategy involves utilizing publicly available internet data sets and proprietary ones from organizations, including comprehensive written or conversational materials.
In a conversation on Bill Gates’ Unconfuse Me podcast in mid-January, Mr. Altman underscored the importance of multimodal functionality, encompassing speech input/output, images, and potentially, video. He anticipated significant advancements in reasoning abilities as well as enhanced reliability, aiming for consistently optimal responses.
GPT-5 Advancements Over GPT-4
How will GPT-5 surpass its predecessor, GPT-4 Turbo, released to subscribers in November? Notably, it boasts an expanded knowledge base, updated up to April 2023 compared to GPT-4’s knowledge cutoff of 2021. Additionally, it can process longer inputs, up to 128,000 tokens or roughly the length of a 300-page book. It demonstrates improved adherence to instructions and can seamlessly switch between tools, such as the Dall-E 3 image generator and Bing search engine, based on user queries.
The chatbot has undergone real-life exam testing, exhibiting promising results. It showed competence in exams such as the chartered financial analyst (CFA) exam, scored impressively in the SAT reading and writing section, and excelled in the 2020 USA Biology Olympiad semifinal exam.
Release Date and Speculations
As of now, OpenAI has not disclosed a launch date for GPT-5, leaving its availability uncertain. The company’s internal upheaval late last year, which saw Mr. Altman briefly ousted before returning, raises questions about the development timeline’s potential impact.
While some users reported encounters with a purported GPT-4.5 Turbo model, this was later confirmed as an error. OpenAI’s recent release pattern suggests a trend of shorter intervals between updates, with more frequent updates like the “turbo” version becoming common.
Will June witness the debut of GPT-5, following the historical trend, or will OpenAI unveil it at the developer day conference, possibly in November? Only time will reveal the answer.
