Meta takes aim at GPT-4 for its next AI model
Last week, Meta made a bold move in the AI world. In the face of other companies like Google and OpenAI keeping their cutting-edge AI models secret, it decided to give away Llama 2, the code that powers its innovative new generative language model, for free. This has sent ripples through the industry as it raises the question: Should AI models be proprietary, allowing a few big companies to control the technology, or should they be open source so that anyone can create their own versions?
Meta is now reportedly taking aim at GPT-4, the most advanced chatbot AI that can be trained to mimic human speech and other behaviors, for its next model. The company’s goal is to make the new AI several times more powerful than its commercial version dubbed Llama 2 (which it currently sells for use with Microsoft’s cloud Azure services to compete with OpenAI’s Bard and Google’s own Gemini). The new system will also reportedly be able to help companies build services that produce sophisticated text, analysis, or other output.
The Wall Street Journal reports that Meta has been actively procuring Nvidia H100 AI training chips and strengthening its data centers in preparation for the training of this new large language model, which is scheduled to start in early 2024. The paper also says that the Facebook parent wants the new AI to be open source, a move that could speed up adoption.
However, there are also concerns that the company’s desire to churn out new AI with ever-increasing power is dangerous. The pace of progress in the field has prompted many prominent technologists to sign an open letter urging companies to slow down while everyone catches up and considers the implications. This warning appears to have fallen on deaf ears.
A lawsuit filed against Meta last week reveals some of the potential ramifications of the company’s plans. Specifically, the lawsuit alleges that the company has infringed on authors’ copyrights by using their works to train its models. For example, the lawsuit says that when a user prompted ChatGPT to generate writing in the style of a specific author, the bot generated a screenplay about a Chinese laborer working on the Central Pacific Railroad who “believe[s] in the power of art to keep their spirits alive.”
The complaint also accuses Meta of violating privacy laws by storing data on users’ devices, and it requests an injunction to prevent the company from creating this kind of AI in the future until a legal framework is established around how such technology should be used. The lawsuit is seeking unspecified damages. In response, a Meta spokesperson told Vox that the company “takes these concerns seriously and has put in place a number of things to support a responsible approach to building with generative AI” including measures for red-flagging data that might violate privacy laws. The company has not yet responded to a request for comment from Reuters.
