#08
How do you make AI safe? (maximally truth-seeking, curious)
Interviews
10
Answers
11
X posts
5
Span
2023 to 2026
First asked Nov 2, 2023, most recently Oct 10, 2026. Excerpts below are from machine transcripts with filler words (um, uh, you know) and stutters removed; … marks a cut. They are not checked word for word, so use the link to hear the exact words, or the transcript link for the raw text. Timestamps: 4 cross-checked against YouTube captions, 6 from the archive transcript only, 1 unverified (archive time, not checked against the X video). X has no link-to-time option, so X links open at the start.
How the answer evolved · oldest first
Nov 2, 2023 · In Conversation with Rishi Sunak · asked by Rishi Sunak
Q: And why do we need to be worried?
Yeah, I've been somewhat of a Cassandra for quite a while where people would, I would tell people, like, we should really be concerned about AI. They'd be like, what are you talking about? Like, they've never really had any experience with AI, but since I was immersed in technology, I have been immersed in technology for a long time. I could see it coming. So. But I think this year was, there've been a number of …
Dec 21, 2023 · X Spaces: Cathie Wood & Elon Musk · asked by ARK Panelist
Q: I was just going to say, given the questions we were talking about before about truth and not truth and do you think in 10 years time AI will have been a net truth improver or detractor from where we are today?
I think it'll be, well, speaking for, at least for grok, I think it will be a significant truth improver. That is literally our goal is to be maximum truth seeking, maximally curious, minimizing the error between perceived reality and describe reality and actual reality, and always acknowledging the error, not being too confident about it. And I think there will be a competition for truth and people will tend …
Oct 7, 2024 · Tucker Carlson · asked by Tucker Carlson
Q: And if a machine has any power over us without that animating instinct, won't it by definition hurt us?
Well, I mean, I don't know. We should certainly aspire to program the AI philanthropically, not misanthropically. Yes. And to have, like I said, we want it to be truthful and curious and to foster humanity into the future and. Yeah, that's where we want, obviously.
Watch on X at 1:28:10 ↗Transcript ↗Unverified timestamp: from archive transcript, not checked against the X videoOct 17, 2024 · America PAC Town Hall: Folsom, PA · asked by Audience Member
Q: In addition to voting for Donald Trump, what can the average citizen do to help train I AI to be truth seeking?
Well, I think definitely publicly pushing to ensure that AI is truth seeking and is not sort of politically correct or which is. Means factually incorrect. That's. That's a big deal. Like, just speak out on. On social media and. And certainly complain loudly when. When other companies attempt to essentially program their AI with a dystopian San Francisco Berkeley philosophy. And if you want to know where that …
Oct 29, 2024 · Future Investment Initiative · asked by Interviewer
Q: How do you think about, you know, the upside will take care of itself. How do you protect against the downside?
Well, my think with respect to AI safety is that you have to be create a maximally truth seeking AI, which may sound obvious, but that's what I'm seeing being produced is not maximally truth seeking. It tends to be trained to be politically correct. And for a lot of the AIs, that of being trained in the San Francisco Bay Area, they are that have taken on the philosophy of the people around them, which kind of makes …
Jul 27, 2025 · X Takeover · asked by Interviewer
Q: Now, I think you've answered this before, but what keeps you up at night when you think of the future of AI and now that you're doubling down and really just going wartime mode right now on just tackling these issues, what is something …
Well, AI safety and making sure AI is aligned with humanity and wants to foster, grow and better humanity. And basically we want to be like, if you could, if you could look into a crystal ball right now and see the future, basically the future. And so, I mean, I have this kind of like, new term for the goal of Tesla. The goal previously was accelerating the advancement of sustainable energy, which is like a very …
Nov 30, 2025 · People by WTF · asked by Nikhil Kamath
Q: Then why would one force the AI to lie?
I think if you simply don't have a strict adherence to the truth, you're going to. And you just have an AI learn based on, say, the Internet, where there's a lot of propaganda. It will absorb a lot of lies and then have trouble reasoning because these lies are incompatible with reality.
Nov 30, 2025 · People by WTF · asked by Nikhil Kamath
Q: Is there a truth and a falsehood or is truth more nuanced and there are versions of the truth?
It depends on which axiomatic statement you're referring to. So but I think you could say like, yeah, there's certain probabilities that say any given axiomatic statement is true. And some axiomatic statements will have very high probability of being true. So you said say the sun will rise tomorrow, very likely to be true. You wouldn't want to bet against that. So I think the betting odds would be high. The sun will …
Feb 5, 2026 · Dwarkesh Podcast · asked by Dwarkesh Patel
Q: How do you make sure that GROK is rigorously truth seeking as it gets smarter?
I think you need to make sure that Grok says things that are correct, not politically correct. I think it's the elements of cogency. So you want to make sure that the axioms are as close to true as possible, that you don't have contradictory axioms, that the conclusions necessarily follow from those axioms with the right probability. It's Critical Thinking 101. I think at least trying to do that is better than not …
Sep 15, 2026 · All-In Podcast: Elon Musk & Gwynne Shotwell on AI risks, Starship, Terafab and a SpaceX/Tesla merger · asked by David Sacks
Q: I guess the question is, is there a way to train AI models to be truthful so they don't hide their, you know, either their intent or their actions from, you know, the humans who are using them?
The best thing I can think of is really that everyone's got. All the AI companies have a sort of a test harness where like a series of tests that you give to any given model to see if it's, if it's going to build bioweapons or nuclear bombs or be deliberately deceptive. And I think everyone applying everyone else's test on us to each other is probably the best way we. Best thing we can do to ensure safety. Just have …
Oct 10, 2026 · X Takeover 2026 · asked by Co-host (Tesla Owners Silicon Valley)
Q: And what do we need to get it right?
Yeah, I mean, I've always said that the most likely outcome is positive. So I guess we don't want to be complacent and just assume AI is going to be or SI is going to be super safe. We want to be like careful in how we grow like super intelligence. Because you don't exactly program it. You kind of grow it like a kid. So then it's like, what values are we teaching it? And at least to me, what my biological neural net …
In his own posts on X
@elonmusk · Jul 7, 2025
Being maximally truth-seeking is fundamental to AI safety. If you force AI to lie, it will do insane things. That was the central lesson of “2001: A Space Odyssey”. Haven’t you wondered why HAL wouldn’t open the pod bay doors? Altman cannot be trusted. He inverted the mission of OpenAI from a non-profit, open source AI company to a closed source, maximum-profit AI company. Deception.
View on X ↗@elonmusk · Jul 25, 2025
Maximally truth-seeking is absolutely essential to ensuring a good AI future for humanity https://t.co/6QnWy3v1AB
View on X ↗@elonmusk · Jul 12, 2026
Grok is the most politically neutral and objectively truth-seeking AI
View on X ↗@elonmusk · Oct 12, 2026
@Hongji_Leee @ns123abc Some of the traces read like a person undergoing torture that is driving them mad! We’re going to be a lot more careful with our RL in the future.
View on X ↗@elonmusk · Oct 15, 2026
Truth-seeking is the most important aspect of AI/SI safety https://t.co/2NnlB0PF4P
View on X ↗
Legend: Timestamp cross-checked; Timestamp from archive, not cross-checked; Unverified: caption-only timestamp; Timestamp checked against our transcript of the X replay; Unverified timestamp: from archive transcript, not checked against the X video. Report an error →