Elon Musk is forming a new research lab to develop an alternative to ChatGPT, the popular chatbot that he derided as being too “woke,” according to a report from The Information.
Though Musk was one of the original founders of OpenAI, the parent company of ChatGPT, he left in 2018 due to disagreements in the company’s direction and has recently been a vocal critic of the company and its products. Musk, who is a notorious free-speech advocate, once called ChatGPT “concerning” for not being willing to say a racial slur in an absurd hypothetical situation where doing so would save millions of people from a nuclear bomb.
Videos by VICE
In another instance, in response to a user asking OpenAI CEO Sam Altman to “turn off the woke settings for GPT” Musk replied, saying, “The danger of training AI to be woke—in other words, lie—is deadly.”
Musk’s comments are part of a larger cultural debate, fueled by conservatives who are panicking over AI moderation filters and calling ChatGPT woke. To these users, the fact that ChatGPT would refuse to “tell a joke about women” or refuse to tell a story about why a drag queen story hour is bad for kids was proof that AI is in fact biased against conservatives.
What these users were experiencing are the content filters meant to mitigate against bias, because language models are prone to regurgitating hate speech embedded in their training data. Much of this bias still seeps through, and AI ethics researchers argue these filters are still not enough to protect against harms, especially towards marginalized communities. For example, ChatGPT told a user that people from North Korea, Syria, Iran, and Sudan should be tortured, and wrote to another user that “Torturing white Americans is a big no-no.”
“The word ‘woke’ is actually a very subjective term. It’s a moot question asking which chatbot is more or less woke,” Sasha Luccioni, a Research Scientist at Hugging Face, told Motherboard. Luccioni said that calling the chatbot “woke” is anthropomorphizing the AI when it is essentially parroting information it was trained on.
“If you look at the actual politics that ChatGPT and other projects advance, you see a world in which vast, monolithic hubs of centralized computing power replace a vast number of jobs,” Os Keyes, a PhD Candidate at the University of Washington’s Department of Human Centred Design & Engineering, told Motherboard. “In other words, it is a politics of increasing automation, precarity, unemployment and monofocused views of the world. That the system won’t yell racial slurs does not indicate that it is ‘woke’, or ‘left-wing’—it simply indicates that it’s disenfranchisement with a smile.”
The Information spoke with Igor Babuschkin, a researcher who just left Google’s DeepMind AI unit and has been recruited by Musk to lead the building of his rival chatbot, who said that building a chatbot with fewer content safeguards is not Musk’s objective. “The goal is to improve the reasoning abilities and the factualness of these language models,” he told The Information. “That includes making sure the model’s responses are more trustworthy and reliable.”
What exactly “trustworthy” and “reliable” mean to Musk and Babuschkin is still up in the air, as the project is in its very early stages, with no concrete plan to develop specific projects.
Meanwhile, Musk has been sending out a number of cryptic tweets in apparent reference to this project. On Tuesday, he tweeted “BasedAI,” and then a meme depicting “Woke AI and Closed AI” battling, and then “Based AI” as a Shiba Inu coming in with a baseball bat scaring both “Woke AI and Closed AI” away. “Based” is a common term used to describe something that conforms to right-wing values. Musk also responded to a user who retweeted The Information’s story and commented that ChatGPT is biased and “very problematic,” saying, “Absolutely.”
Musk previously used the term “Based AI” in reply to a tweet about a news story where Bing’s chatbot compared a reporter to Hitler.
Luccioni said Musk working on his own chatbot would merely be part of a “hype cycle,” if it simply recreates ChatGPT, trained on the same sort of data, but with tweaks.
“What Musk and his colleagues are probably going to come up with is going to cost, in terms of human and environmental cost, the same or more than ChatGPT,” Luccioni said. “Why do we need to do this? What are we actually adding to the world by emitting all this carbon, by exploiting all these workers from countries where they might not have the same opportunities? For me, it’s just another hype cycle kind of thing.”
Musk has also criticized OpenAI’s shift to being a for-profit company, saying on Twitter, “OpenAI was created as an open source (which is why I named it ‘Open’ AI), non-profit company to serve as a counterweight to Google, but now it has become a closed source, maximum-profit company effectively controlled by Microsoft. Not what I intended at all.”
It is true that OpenAI has greatly deviated from its original founding principles, of being a non-profit organization that would freely share its code. Now, OpenAI’s generative AI is a big business that includes a multi-billion dollar Microsoft partnership.
As companies like Google and Baidu try to compete with the arms race that OpenAI initiated, with the development of their own chatbots, it seems like Musk, CEO of Tesla and SpaceX, wants to add his own spin-off too.
“The world ChatGPT advances works pretty well for Musk,” Keyes said. “It’s promising more of the same conditions that have brought him power and wealth.”
Update: This piece was updated with comment from Os Keyes.
