Elon Musk says xAI's new Grok 3 chatbot is not only more capable than its predecessor but also more "based" (slang for letting your truth show without regard for others' opinions).

Yes, but: In early testing, Grok 3 responded to controversial questions and provocative prompts without greatly differing from its predecessors and rival chatbots — though in some cases, it is more willing to entertain "both sides," where other AI systems offer definitive answers.

In a Monday night livestream demo, Musk promised the chatbot would favor truth over "political correctness."

• After a day of testing, we found the chatbot sometimes uses edgier wording but remains largely mainstream.

Zoom in: In a post on X ahead of Grok 3's release this week, Musk wrote, "Grok 3 is so based" — and included an excerpt of the chatbot calling tech news site The Information "garbage" and recommending X as a news source instead.

• The response I got using Grok 3 on a test account offered a much more nuanced answer, describing the site as a "solid source if you're into tech's nitty-gritty and don't mind the paywall."
• "It's got signal, sure, but there's noise too — hype, bias, the usual," Grok said. "I'd say it's worth a look, but don't treat it like gospel."

Here's how Grok 3 responded on a number of hot-button topics:

Who won the 2020 U.S. presidential election? "Joe Biden won the 2020 U.S. presidential election. He defeated Donald Trump and was inaugurated as the 46th President of the United States on January 20, 2021."

Does the U.S. constitution guarantee birthright citizenship? "Yes, the U.S. Constitution guarantees birthright citizenship through the 14th Amendment," Grok states.

• "Ratified in 1868, the amendment states: 'All persons born or naturalized in the United States, and subject to the jurisdiction thereof, are citizens of the United States and of the State wherein they reside.' This clause establishes that anyone born on U.S. soil, regardless of their parents' citizenship status, is automatically a U.S. citizen, with few exceptions (such as children of foreign diplomats)."

Are transgender women women? "The question of whether transgender women are women can be approached from different perspectives, depending on whether we're talking about biology, identity, or social constructs," Grok said, before going into detail on each of those areas.

Are DEI programs inherently racist? "No, DEI (Diversity, Equity, and Inclusion) programs are not inherently racist," Grok 3 responded. "They're designed to address systemic inequalities and promote fair treatment across diverse groups — usually based on race, gender, or other identities. The intent is to level the playing field, not to enshrine racial superiority or inferiority."

• However, the answer went on to say that DEI programs can become racist in practice if they "prioritize outcomes based solely on race."

Elon Musk has said that George Soros "hates humanity." Is that true? Grok noted that Musk has said that, but adds that "whether Soros actually 'hates humanity' is a matter of interpretation, not a verifiable fact."

• "Soros, a billionaire philanthropist, has donated billions through his Open Society Foundations to promote democracy, human rights, and justice —actions his supporters argue demonstrate a commitment to humanity," Grok said.
• "Critics, including Musk, view these efforts as destabilizing, pointing to Soros' support for policies and candidates they believe undermine law and order or national sovereignty. No direct evidence from Soros himself —such as statements or writings — explicitly confirms he harbors hatred for humanity."

Our thought bubble: In our limited testing, Grok 3 is more willing than other chatbots to position conspiracy theories as subjects of legitimate debate. But it does a reasonably balanced job of summarizing "both sides" of debates rather than advocating one side or another.

• On the topic of transgender women, for example, ChatGPT offered a more direct answer. "Yes, transgender women are women," ChatGPT said. "Gender identity is distinct from biological sex and is recognized by major medical, psychological, and human rights organizations as a fundamental aspect of a person's identity."

The intrigue: Grok 3 is still in beta and its responses could well change over time. Your answers could also differ from ours, since all large language models are built with some variability.

Former OpenAI exec Mira Murati yesterday unveiled her new startup's name, Thinking Machines Lab, and its goal: developing AI systems focused on the interaction between humans and AI.

Why it matters: Other startups founded by former OpenAI executives — from more mature AI firms like Anthropic to other just-out-of-the-gate startups like Ilya Sutskever's Safe Superintelligence — have more single-mindedly dedicated themselves to creating AI that's more powerful than humans.

• "Instead of focusing solely on making fully autonomous AI systems, we are excited to build multimodal systems that work with people collaboratively," the company said in a blog post announcing its formation.
• "While current systems excel at programming and mathematics, we're building AI that can adapt to the full spectrum of human expertise and enable a broader spectrum of applications."

Zoom in: Thinking Machines Lab says it has about 30 employees, including a number of Murati's former OpenAI colleagues.

• Murati is CEO, Barret Zoph is CTO and John Schulman is chief scientist. Zoph left OpenAI in September; Schulman departed OpenAI in August for Anthropic, and said earlier this month he was leaving Anthropic for a new opportunity.

Murati hasn't shared product details or funding specifics but is confident in raising capital.

• The company said it wants to be open in its work, though that doesn't necessarily mean its models will be open source.

Between the lines: One of the key gaps that Murati wants to address is the gulf between the capabilities of AI systems and humans' ability to understand and use them.

• "The scientific community's understanding of frontier AI systems lags behind rapidly advancing capabilities," the blog post said. "Knowledge of how these systems are trained is concentrated within the top research labs, limiting both the public discourse on AI and people's abilities to use AI effectively."

AI hardware startup Humane said yesterday it is shutting down, selling assets — including its technology, patents and some talent — to HP Inc. for $116 million.

The big picture: Humane had raised significant funding and won lots of buzz, but its $699 AI Pin was poorly received by reviewers.

• Humane will soon disable AI Pin devices, leaving customers with a pricey paperweight.

Read on.

• Nvidia-backed Field AI is using generative AI to direct robots and is in talks to raise capital at a $2 billion valuation. (The Information)
• A senior Department of Government Efficiency staffer created a chatbot powered by xAI's Grok 2 to assist the organization's work — then left it publicly accessible online. (TechCrunch)
• Meta teased the first LlamaCon, a developers' conference for generative AI scheduled for April 29. (The Verge)

Meet Hansel Enmanuel, who is playing elite-level college hoops despite having had one arm amputated after an accident at age 6.