SCOTT DETROW, HOST:
It has been quite a week at Elon Musk's X, the social media platform formerly known as Twitter. On Tuesday, Grok, the platform's AI chatbot, put out a series of antisemitic tweets and even praised Hitler. The next day, X's CEO quit, just as the latest version of Grok was set to be released.
So what does all of this tell us about X and where it may be headed? Charlie Warzel covers technology for The Atlantic, as well as the broader story of X's shift into a platform that highlights and centers extreme right-wing voices and white supremacy. Welcome back to ALL THINGS CONSIDERED.
CHARLIE WARZEL: Thank you for having me.
DETROW: I want to get back to the broader trends. But first, for the people who may be lucky enough to not know all the proper nouns in what I just described, can you give, like, a 30- to 45-second recap of what exactly happened?
WARZEL: Yeah, so Grok is the large language model chatbot that kind of overlays the entirety of the X platform. It's sort of like the ChatGPT interface for X owned by Elon Musk. And the company made a tweak to the chatbot, which is named Grok, sometime last week, and it nominally tried to make the chatbot a little less politically correct. Didn't want it to be woke in any way - but they also wanted it to pull its answers from the entire corpus of X posts, right? The company gave the chatbot these new sets of prompts, and the chatbot processed all that and responded by becoming incredibly racist and basically became a white nationalist.
DETROW: One reason I wanted to talk to you is that I feel like you repeatedly frame this very bluntly and succinctly when you write about. One sentence in your last piece jumped out to me.
(Reading) It is worth reiterating that this platform is owned and operated by the world's richest man who, until recently, was an active member of the current presidential administration.
So what, to you, are the broader implications for what this chatbot is doing and saying, replying to people on X?
WARZEL: Well, I think there's a couple of things here. The first is that I think this highlights a real problem with the large language models in general as they get more complex. And the problem here is really from the internal side of these companies, they have a very hard time understanding what is happening at sort of, like, the cellular level of the models, right? They know the weights, they know the balances, they know basically what it was trained off of, and they know what the prompts are given. But they don't know how the model is interpreting these prompts, right?
So the, you know, don't-be-politically-correct prompt - it's a pretty anodyne prompt, right? It's not saying be Hitler, be antisemitic. But the models respond in this really unexpected way. So I think there's this idea of, we've unleashed this technology on people. At the same time, the people who, you know, are in charge of it can't really control what it's doing with any real specificity or know exactly why it's doing what it's doing. I think that that's a big problem.
The second problem, I think, with all of this is that Grok is sort of a reflection of Elon Musk's values or, at least, the way that he wants an AI agent to respond to the world. And it's really alarming that an AI - which is just supposed to sort of interpret the intentions of its, you know, builders - reads all of this direction, and then decides, I'm going to go to X and get some context and some reference points, and it decides that the most rational course of action is to become a neo-Nazi. I think that says just quite a bit about the man who's in charge of this platform and also the culture and values that are being imbued by it.
DETROW: Does that, you think, say more about the people in charge of X right now or just the conversations that are happening on X? If you're soaking in all of X and your output is praising Hitler, who does that condemn more?
WARZEL: I think it definitely condemns, more than anything, the level of discourse and conversation that's happening on X. Since Musk bought it, you know, he has reinstated lots of accounts of people who are, you know, proud white nationalists, trolls, shock jocks, you know, you name it. The level of permissiveness on the platform for, you know, the type of hate speech has increased greatly. And I think it speaks to the toxicity of that platform and what a lot of people, whether they know it or not, are marinating in if they stay on the platform. You might not realize just how incendiary the conversation has become on there. But I think, like, Grok is actually a very clarifying moment for that - right? - because it can kind of take everything in in an instant and grab a snapshot, and the snapshot that it's chosen to grab is essentially becoming a bigot.
DETROW: That's Charlie Warzel, staff writer at The Atlantic, where he covers technology. Thank you so much.
WARZEL: Thank you for having me.
(SOUNDBITE OF MUSIC)
DETROW: Earlier today, X put out an apology for Grok's, quote, "horrific behavior." They say Grok's actions on Tuesday was due to a coding update. The statement went on to say, quote, "we have removed that code and refactored the entire system to prevent further abuse," end quote.
(SOUNDBITE OF MUSIC) Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
