OpenAI has begun to recruit Android and Apple iOS engineers, and mobile phones are expected to become hardware portals for generative AI.

Thanks to CTOnews.com netizen Snailwang for the clue delivery! CTOnews.com, April 17, OpenAI recently released a recruitment message, announcing the recruitment of mobile terminal engineering manager, Android software engineer and iOS software engineer, etc., to start the ecological research and development of mobile software. This means that the mobile phone may become the hardware portal for the generative AI.

ChatGPT was launched on November 30, 2022 and has 100 million users in just two months. Behind its AI model and algorithm is Nvidia's artificial intelligence hardware.

According to reports, the model behind ChatGPT is GPT (Generative Pre-Training, generative pre-training model), and the name GPT shows that the model is biased towards "generating text and code".

CTOnews.com to do a simple calculation here, with 175 billion parameters of GPT-3, because of its huge floating-point computation per second, the cost of a single training reached $4.6 million. If you use the Nvidia A100 chip, the use of a single training should reach hundreds, the cost is about 30 million yuan.

Counting, the first generation GPT-1 was born in June 2018, the number of training parameters is 120 million, the database size is 5GB; half a year later launched the second generation GPT-2 (February 2019), the number of training parameters is 1.5 billion, the database size is 40GB; the third generation GPT-3 was born in May 2020, the number of training parameters leapt to 175 billion, the database size reached 45TB.

At present, OpenAI has launched a new generation of text generation AI model GPT-4 multimodal model in March this year, which is comparable to human beings in a variety of professional and academic standards!

OpenAI CEO Sam Altman (Sam Altman) has confirmed that the company is not currently training GPT-5 and "will not do so in the short term". GPT-5 is considered the successor to GPT-4, the big language model released by OpenAI in March this year.

However, not training GPT-5 does not mean that OpenAI is no longer expanding GPT-4 capabilities. Altman stressed that they were also considering the security of the work. "We are doing other things on top of GPT-4, and I think these are security issues, which have been completely ignored in the letter," he said.