Twitter owner Elon Musk plans to develop his own generative artificial intelligence (AI) model that will be integrated into Twitter. However, up until now it is not clear what exactly it will be used for.
Two Twitter sources have told Insider that Musk recently purchased 10,000 of the powerful GPUs needed to handle large language models.
Musk's AI project, which is still in its early stages, will be built around LLMs, or large language models, according to people familiar with the situation inside Twitter. And Twitter has a huge amount of data that can be used to teach LLMs. For example, OpenAI, which created ChatGPT, previously had access to Twitter data for educational purposes, but last December Musk announced that they had already closed OpenAI's access to their data.
Adding to the picture of Musk's plans to develop his own AI is that two well-known AI developers, Igor Babushkin and Manuel Kreis of DeepMind (a subsidiary of Alphabet) moved to Twitter in early March.
The newspaper notes that it is not yet known for what purpose Twitter plans to use AI. It is possible that it will be used to improve the search in the social network, because it was the search that Musk complained about a lot. He even hired security expert George Hotz for three months to improve search on Twitter (Hotz, however, left the company a month later). Another area where AI systems may be in demand is advertising technology.
Iinterestingly, Elon Musk has his own story related to AI: it was he and Sam Altman who founded the OpenAI company that developed ChatGPT. Musk left the company in 2018. This year, Musk began criticizing OpenAI and its development, ChatGPT, and in March joined a letter to the Future of Life Institute calling for it to stop teaching AI systems more powerful than GPT-4.
On April 11, Alibaba presented its equivalent of ChatGPT, which it plans to integrate into all its platforms. It's a large language model called Tongyi Qianwen, which the company will integrate into its Tmall Genie smart speakers and workplace messaging platform DingTalk for now. This AI neural network was trained on massive data to be able to provide stunning responses to user prompts.