Recently, Sam Altman, the father of ChatGPT, disclosed the recent development roadmap of OpenAI, this year's priority is not to develop GPT-5, but to reduce the current cost of computing power, launch cheaper and faster GPT-4, Ultraman said that the biggest problem of OpenAI is actually the problem of computing power, ChatGPT's recent problems and customer complaints, most of them are just GPU shortages! Ultraman also stressed that open source is very, very important, facing GPU shortage panic, OpenAI is considering choosing a suitable time to open source GPT-3, and due to insufficient computing power, GPT-4's multimodal capabilities are not yet available to all users.
Since the advent of ChatGPT, OpenAI's every move has received widespread attention around the world, and the update of GPT-4 with multimodal capabilities has also led the development trend of large models, and the world's most advanced language large-scale models need to be benchmarked against GPT-4!
However, with the development of ChatGPT, the voice calling for regulation is becoming more and more powerful, and Musk, an early angel investor in OpenAI, has repeatedly stressed that GPT has security risks and should strengthen control to avoid the risk of AI destroying mankind!
Last week, led by Sam Altman, the father of ChatGPT, and Bengio and Hinton, the father of artificial intelligence, Turing Award winners, more than 300 bigwigs signed an open letter, the core content of the letter is only one sentence - humans need to reduce the risk of AI, otherwise artificial intelligence will be like nuclear war, endangering all mankind!
Recently, some OpenAI developers and startup founders held a closed-door meeting to discuss the future of OpenAI roadmap and challenges, which was released by Humanloop co-founder and CEO Raza Habib, OpenAI founder Altman believes that the biggest problem OpenAI has faced recently is limited by GPT. Recently, some OpenAI developers and startup founders held a closed-door meeting to discuss the future of OpenAI roadmap and challenges, which was released by Humanloop co-founder and CEO Raza Habib, OpenAI founder Altman believes that the biggest problem OpenAI has faced recently is limited by GPT.
Without the powerful GPU computing power, and the powerful CUDA ecosystem behind the computing power, it would be difficult for AI to develop to the extent it is today. According to Lao Huang, ChatGPT's initial success was very dependent on the support provided by NVIDIA.
The daily experience of ChatGPT affects tens of millions or even hundreds of millions of users, and the most recent problem of user complaints is the reliability of the content provided by ChatGPT, and the corresponding speed problems, Altman said that most of the current OpenAI problems are caused by GPU shortages.
In terms of future development, Ultraman gave a full-year development plan for 2023, the first priority is to launch cheaper and faster GPT-4, with the goal of reducing the "smart cost" as much as possible, and the context window may be as high as 1 million tokens in the near future.
Ultraman also mentioned that ChatGPT does not currently use efficient fine-tuning methods like Adapters or LoRa, and this year will expand the fine-tuning API to the latest model, and will launch a version of the API that can remember historical conversation records, and in 2024, it will mainly solve the multimodal problem, and until the GPU shortage is solved, the multimodal function will not be available to everyone.
Although often ridiculed as "ClosedAI," Ultraman argues that OpenAI should be more "Open," emphasizing the importance of open source, and when talking about why it wasn't made public before, Ultraman said it was because he felt that other companies were too poor to properly handle the use of such a large language model. Faced with all kinds of panic caused by the GPU shortage, Altman said that OpenAI will choose the right time to open source GPT-3 technology.
