- Meta CEO Mark Zuckerberg gave his thoughts on the AI data race in a new interview.
- As the AI arms race intensifies, many technology companies are vying for new data sources.
- But Zuckerberg said “feedback loops” are more important than data when it comes to training AI models.
Meta CEO Mark Zuckerberg A passionate look at Big Tech’s race for AI training data. It’s not the data that matters.
“What I think is more valuable than any prior corpus is the feedback loop,” Zuckerberg said. Interviewing on the command linea technology industry newsletter.
Feedback loops are used to retrain and improve AI models over time based on previous outputs. These algorithms can, for example, recognize when an AI model has made an error and provide data to adjust future performance.
“Getting more people to use it, seeing how they use it, and being able to improve from there will actually become more differentiated over time,” he said.
Sourcing new data for insatiable AI models to consume can theoretically make them smarter. It has now become an obsession for companies vying for AI dominance.
Companies like OpenAI, Google, Amazon, and Meta are considering it. some wild solutions. Meta, for example, was so in need of data at one point that it considered acquiring the publisher Simon & Schuster. Considering the risk of copyright litigation The New York Times reported the details.
Another solution to the problem of limited data is to create new data, which Big Tech calls “synthetic data.” Synthetic data is artificially generated and designed to mimic data generated by real-world events. Zuckerberg is into it.
“With synthetic data, I think we’re going to use it a lot more to let the model try different problems, see which path ultimately works, and then refine it,” he said.
Anthropic, the maker of the chatbot Claude, also Incorporate internally generated data into the model.and Chat GPT Manufacturer OpenAI is also considering it, but CEO Sam Altman said at a conference last May that the key would be “models that are smart enough to create good synthetic data.” He said there is.
And while Zuckerberg believes feedback loops are the key to building powerful AI models, there are risks to relying on them. Some of your mistakes, limitations, and biases can be reinforced if you are not trained on the “good data” to begin with.
Axel Springer, Business Insider’s parent company, has a global deal that allows OpenAI to train models based on its media brands’ reporting.
