Google postpones announcement of 'GPT-4 rival'

google postpones announcement of gpt 4 rival 65735cd6c9e56 | Dang Ngoc Duy

The Information quoted two internal sources saying that Alphabet CEO Sundar Pichai requested to postpone the launch date of the large language model Gemini to next year. Previously, the company planned to announce new products this week, simultaneously at events held in California, New York and Washington.

Alphabet CEO Sundar Pichai talked about Gemini at the Google I/O 2023 event in May. Photo: YouTube/Google

Alphabet CEO Sundar Pichai talked about Gemini at Google I/O 2023 in May. Photo: YouTube/Google

Gemini is a multi-modal AI model researched and developed by the Google DeepMind department, first mentioned in May at the Google I/O 2023 event. This is considered the company’s “most serious effort” in AI fever arises. Sergey Brin, co-founder of Google, also had to return to the company to promote Gemini development after many years of “reclusion”.

Gemini is Google’s ambitious project and works similarly to the existing GPT-4 model on ChatGPT. However, sources said that, except for English, Gemini still performed poorly in other languages. “Google engineers found that Gemini lags behind GPT-4 in multilingual recognition and response capabilities, so more improvement is needed,” a source revealed.

Previously, Google said that Gemini “possesses impressive multimodal information processing capabilities unprecedented in previous models, achieving high efficiency in integrating into tools and application programming interfaces ( API), opens up future innovations”. Google also plans to create multiple versions of Gemini to put on its services such as Bard , Search and Workspace as well as a mobile version called Gecko.

According to Interesting Engineering , Gemini could be one of Google’s most important products at launch. The model will show that the company can compete and surpass OpenAI, as search trends shift from traditional platforms to chatbots.

In fact, currently no company has deployed an AI model capable of approaching OpenAI’s GPT-4. Instead, the market is flooded with products with a level equivalent to GPT-3.5. In August, two experts Dylan Patel and Daniel Nishball of research company SemiAnalysis said that Google Gemini “will break GPT-4 with five times the power even though it uses fewer GPUs”.

Last month, OpenAI launched an upgrade called GPT-4 Turbo with higher speed and more cost savings. In mid-November, CEO Sam Altman also revealed that the groundbreaking GPT-5 model will soon be introduced next year.

Bao Lam

Leave a Reply

en_USEN