Pavlo Gonchar | Light flare | Getty Images
Twitter CEO Elon Musk threatened with lawsuits Microsoft Wednesday, accusing the software giant of illegally using data from the social media company to train its artificial intelligence model.
Musk’s tweet threat came after Mashable and other posts reported This Microsoft would remove Twitter from its ad platform, which allows ad buyers to manage all of their social media accounts in one place.
“They trained illegally using Twitter data,” Musk tweeted. “Time for the trial.”
Musk, who is also CEO of You’re here and SpaceX, often tweet about plans that never materialize, and no lawsuits appear to have been filed. Twitter’s press line did not respond meaningfully to a request for comment, and a Microsoft representative declined to comment.
Musk’s threat is the latest indication that data ownership is fast becoming an uphill battleground in the rush for generative AI. Big tech companies are scrambling to develop cutting-edge AI models like OpenAI’s GPT, and data owners are looking to stop them or charge for the use of their content.
Microsoft develops its own so-called Large Language Models (LLM) and sells access to OpenAI’s models. Microsoft invested $10 billion in OpenAI last year in an unusually structured deal. Musk was a co-founder of OpenAI before leaving its board in 2018 and recently complained about the company’s shift from a not-for-profit model to a very valuable company influenced by Microsoft.
LLMs like GPT require terabytes of data for training, much of which comes from websites like Reddit, StackOverflow, and Twitter. Social media training data is valuable because it captures casual conversations back and forth.
As these new AI models move from research labs and universities to the corporate world, data owners are beginning to make demands.
For example, Reddit said earlier this week that it charge companies for access to its programming interface used to power conversations between Redditors in the AI training software. Universal Music Group also said this week that such training of artists’ music would represent “both a violation of our agreements and copyright infringement” in response to a viral video of a song that claimed to use AI to impersonate the rapper Drake.
And the Getty Images photo database continues Stable Diffusionalleging that the company copied its content to train its AI image generator.
Musk said in December that Twitter would “suspend” OpenAI’s access to its database. He also announced his intention to create his own large language model in one of his companies called TruthGPT.
SHOW: Elon Musk wanted to support OpenAI in 2018