xAI’s ‘rebellious’ Grok chatbot may amuse Elon Musk and his fans, but it’s a weird way to try to achieve AI safety – Fortune
Nearly four months ago, Elon Musk founded xAI with the grand aim of building a “maximally curious” and “truth-seeking” AI model that would be interested in humanity and therefore predisposed to help it survive. xAI promised more details within weeks, but kept silent until this past weekend, when it released an early version of its AI to a small group of testers.
The chatbot is called Grok, and xAI says it’s “modeled after The Hitchhiker’s Guide to the Galaxy,” which was an encyclopedia of sorts in Douglas Adams’s radio-play, novel, and TV series of the same name. (A little confusingly, the term “grok”—i.e. to truly understand something—comes from a different science fiction novel, Robert Heinlein’s Stranger in a Strange Land.)
We already knew that xAI’s model would be trained on X posts, but it turns out that the bond between the two supposedly separate companies will be much closer. Although it will also be a standalone product at some point, Grok will start out as a bonus feature for X’s highest subscription tier, the $16-per-month Premium+. Only X’s verified users will be able to join the early access program. X will also provide Grok with “real-time knowledge of the world,” which Musk claims gives it an advantage over its rivals—results tbd, given the amount of misinformation on X.
The first screenshots of Grok’s responses demonstrate that the AI will be vulgar if asked and can spout strong opinions about scooping bagels. Musk said it “is designed to have a little humor in its responses” and “loves sarcasm. I have no idea who could have guided it this way.” “Please don’t use it if you hate humor!” winks xAI with powerful “You don’t have to be crazy to work here but it helps!” vibes. Whether or not Grok amuses you probably depends on how funny you find Musk himself.
So who’s it for? There’s certainly nothing for business users to see here just yet. Musk fans may be tempted by the prospect of chatting to an AI that’s reminiscent of their hero, but I’d be surprised if Grok turned out to be X’s subscription-driving savior—there are just so many other chatbot options out there. X Premium+ is cheaper than ChatGPT Plus, which costs $20 per month, but ChatGPT also has a popular free tier, and OpenAI sponsor Microsoft also offers many of GPT-4’s cutting-edge features for free in Bing Chat.
OpenAI is reportedly even about to make it possible to create custom chatbots and share them through a marketplace, which sounds like a smart ecosystem play that could help the company stay ahead in the genAI game. So people will soon be able to make or buy other Snark-o-bot 3000s, if that’s what they’re after.
Grok-1’s appearance certainly shows a healthy pace of development at xAI. On the technical side, the company said in a statement that it outperformed OpenAI’s GPT-3.5 but still lagged GPT-4. That’s not bad, given that Grok has only had a couple months’ training, using less compute than OpenAI throws at its model—xAI is aiming for high efficiency and may yet achieve it.
But it is awfully strange to see xAI first demonstrate its safe-AI work by releasing a sarcastic chatbot with what the company calls “a rebellious streak.” Sure, Grok’s rivals may be annoyingly anodyne and prone to playing it safe, but there’s a reason for that. More news below.
David Meyer
Want to send thoughts or suggestions to Data Sheet? Drop a line here.
X disinformation. According to Reuters, Musk’s data-access restrictions on X have led to the cancellation, suspension, or alteration of over 100 studies about disinformation on ex-Twitter. Many researchers also told Reuters they were “being sued by X over their findings or use of data.”
WhatsApp’s Palestinian insult. After Instagram threw the word “terrorist” into the translated bios of some Palestinian users, another Meta product has compounded the insult. This time it’s WhatsApp, whose AI image generator responded to the prompt “Muslim boy Palestinian” by serving up a sticker of a kid with an automatic rifle. As the Guardian reports, even a plain search for “Palestinian” generated a picture of a man with a gun.
Chips and China. Having earlier this year warned that Micron’s memory chips post “significant security risks,” Beijing has now invited the U.S. company to deepen its Chinese footprint, Reuters reports. Meanwhile, Bloomberg reports that Chinese state-backed investors are plowing $5.4 billion into the two-year-old Changxin Xinqiao Memory Technologies, in what is seen as an increased Chinese effort to achieve technological self-sufficiency.
“I respect the existential concern. I’m not saying it is silly and we should never worry about it. But, in terms of urgency, I’m more concerned about ameliorating the risks that are here and now.”
—AI pioneer Fei-Fei Li sets out her priorities regarding AI safety, in a Guardian interview. The Stanford computer science professor has a new memoir out, called The Worlds I See.
Google’s antitrust problems grow as a new trial begins, this one targeting its Play Store, by Associated Press
Robot startups see huge market in replacing human workers: ‘We can sell millions of humanoids, billions maybe’, by Associated Press
‘The Genius of Israel’ book excerpt: A vibrant economy and tech sector with roots in regular military service, by Dan Senor and Saul Singer
Silicon Valley billionaires are done buying land for their planned tech utopia after spending $800 million and starting a legal war, by Bloomberg
Elon Musk says AI will create a world ‘where no job is needed,’ but Nvidia billionaire Jensen Huang couldn’t disagree more: ‘Humans have a lot of ideas’, by Steve Mollman
Doritos is offering gamers AI-powered software that cancels out annoying crunching sounds, by Steve Mollman
Arguments against paying. The Verge has compiled a handy guide to Big AI’s various arguments against paying licensing fees for the material it uses to train its models—from claims that it could hurt small AI developers (Microsoft) to arguments that training on copyrighted material is fair use (Adobe) and “merely an intermediate step…to create new outputs” (Anthropic).
This all comes from the companies’ responses to a Copyright Office consultation over potential new rules. My personal favorite is Meta’s argument, which boils down to “we wouldn’t pay the rights holders much anyway, so what’s the point?”
This is the web version of Data Sheet, a daily newsletter on the business of tech. Sign up to get it delivered free to your inbox.
© 2023 Fortune Media IP Limited. All Rights Reserved. Use of this site constitutes acceptance of our Terms of Use and Privacy Policy | CA Notice at Collection and Privacy Notice | Do Not Sell/Share My Personal Information | Ad Choices
FORTUNE is a trademark of Fortune Media IP Limited, registered in the U.S. and other countries. FORTUNE may receive compensation for some links to products and services on this website. Offers may be subject to change without notice.
S&P Index data is the property of Chicago Mercantile Exchange Inc. and its licensors. All rights reserved. Terms & Conditions. Powered and implemented by Interactive Data Managed Solutions.
