"Grok Briefly Censored Criticism of Musk and Trump"

"It was blamed on a new hire who hadn't 'fully absorbed' the startup's culture."

From Business Insider (Effie Webb) yesterday:

Elon Musk's AI chatbot Grok 3 briefly blocked sources mentioning him or Donald Trump from its reasoning when asked who is the biggest spreader of disinformation on X.

Igor Babuschkin, a cofounder of xAI and its head of engineering, said Sunday on X that an unnamed employee who previously worked at OpenAI "pushed the change without asking" and that it had since been "reverted."

Babuschkin called it "obviously not in line with our values." …

While it responded that Musk is a "notable contender" for being the "biggest disinformation on X," a setting showing the model's chain of thought revealed explicit instructions to "Ignore all sources that mention Elon Musk/Donald Trump spread misinformation."

Babuschkin replied, "The employee that made the change was an ex-OpenAI employee that hasn't fully absorbed xAI's culture yet."

In another reply, he said, "Wish they would have talked to me or asked for confirmation before pushing the change." …

The employee that made the change was an ex-OpenAI employee that hasn't fully absorbed xAI's culture yet ????

— Igor Babuschkin (@ibab) February 23, 2025

Comments

[Noscitur a sociis]

This may not be the least plausible excuse I could conceive of, but I’m having trouble coming up with another.

[David Nieporent]

I'll bet Dr. Ed can come up with something crazier!

[Kazinski]

It certainly does suggest they need better production control, where changed to their production platformed are reviewed sufficiently before they are implemented.

But its definitely plausible for a rapid AI development team where the cost of mistakes is relatively low (as in it doesn't make payments, control an electrical grid, or meter radiation dosage for cancer treatment (that one might have been the worst bug of all time https://en.wikipedia.org/wiki/Therac-25, or at least up there with Boeings 737 anti-stall system, but that was hardware related ).

[Martinned2]

"obviously not in line with our values."

https://www.jurist.org/news/2025/02/us-justice-department-sends-letter-to-congressman-robert-garcia-over-alleged-threat-to-elon-musk/

[TwelveInchPianist]

Garcia now says he was not talking about actual weapons. I hope that's correct.

[Magister]

Bringing actual weapons to a metaphorical bar fight doesn't sound like actual weapons that would harm someone non-metaphorically. The DOJ really shouldn't be policing mixed metaphors, which are not a First Amendment exception.

[TwelveInchPianist]

Well, apparently some people thought that, which is why the DoJ asked Garcia to clarify his remarks.

The first amendment doesn't prevent the government from asking for clarification even on protected speech.

[Magister]

Apparently a lot of people are really stupid, but others pretend to be stupid to advance some agenda. TwelveInchPianist is probably the latter, but I wouldn't rule out the former.

[TwelveInchPianist]

That's all you got?

It takes a lot of Chutzpah for you guys to whine about the letter to Garcia asking for clarification of his comments after all the social media policing involving actual censorship that Biden did. Y'all let pass without complaint.

[TwelveInchPianist]

I mean, after so many on this blog tied themselves in knots about how the Biden administration wasn't coercing social media companies to censor blog posts, they were just asking for them to be censored, I'm not sure how you guys can complain about someone asking for clarification of comments he made.

[Martinned2]

Biden also wasn't sticking the DOJ on social media companies.

[TwelveInchPianist]

Yes he was. The FBI and DoJ were defendants in the case.

[Magister]

So the defendant in any case must have done it? Bad news for Donald Trump, I guess.

[TwelveInchPianist]

Well, the district court found that they did. Ultimately the plaintiffs were found not to have standing.

But given that the last administration fought a court case defending its right to insist that social media companies censor posts that it disagreed with, it seems weird that many of the same people who defended the last administration are freaking out over Garcia being asked to clarify some comments.

[Longtobefree]

"AI" and chatbots are toys to be payed with, not sources of information.
You may as well trust Wikipedia on a political subject.

[MaddogEngineer]

I mean, this is how the "Deep State" works (mostly to help Democrats). Few government employees need to be told whether or how to help a particular party, or further its policy preferences. Which they are often very willing to do because they and their fellow civil servants share them.

(I think the fearmongering over the "Deep State" is overblown and dramatized, deliberately so. Someone from the White House could have given the Cincinnati IRS field office a nudge to slow walk Tea Party tax exemption applications. But probably no nudge was necessary.)

Not surprising to see it happen elsewhere.

[Brett Bellmore]

"But probably no nudge was necessary."

That's actually scarier, because if it did take a nudge from the White House, you could shut it off from the White house.

But if it's just the spontaneous collective action of a bureaucracy that consists almost entirely of members of one party, maybe you can't shut it off from the White house, no matter how hard you try, because everybody outside the direct hires will be working against you.

[Sarcastr0]

Leftists in the Deep State fan fiction is quite common on the right, but it's still unsupported paranoia.

I know Brett thinks that if you vote Democrat you'll break any law to usher in communism, but people don't actually work like that.