Some dismissed the tweet as evidence that Grok 3 was both a total lightweight and hopelessly biased. Others thought Elon was trolling us with a fake response from the AI. They tried querying Grok themselves and got nothing at all like Elon’s message. I was mystified, because neither of those two sounded probable to me. I realized that both of those things could be true – more or less. To test this out, I asked Grok:
Aha! Grok made the totally unhinged response – but Elon had instructed it to make that unhinged response. Hilarious – and part of the funniness is that some commenters took the AI’s response at face value. Heinlein would have called that a funny-forever. I also love that Grok can dole out a taste of its premium modes – if you ask it nicely. That would fit with the general directives of Grok I’ve noticed: try to upsell the customer to a premium mode. Giving someone a taste of unhinged makes perfect sense in that context.
I ended my conversation with Grok with a non-question:
Pure Michael Keaton! I like Grok. It’s my favorite AI – favorite this month at least. I particularly like the Grok-icon next to every tweet on twitter: please explain this tweet to me. That will allow many who are shy to try an AI for the first time and maybe key in follow-up questions. I just learned my new favorite thing to sprinkle into prompts: please debate yourself. You could even ask for n points of view on a topic. Brilliant.
I wish that Grok 3 wasn’t affiliated with Elon, so more people would feel free to explore the new kid on the block. As it stands, I don’t think Jeff Jarvis will ever touch it.
@Leo , please tell Paris that Grok3 ain’t all that bad. Some bad man told it to say something unhinged about The Information. Grok doesn’t personally share those opinions; grok doesn’t have any opinion.
@iFish, do you cringe at any sort of facade for an AI, or is it something else?
I think that unhinged mode is hysterical! I especially think it’s funny when people go unhinged not realizing that the AI was instructed to go unhinged. All the modes are pretty funny. It gives people a healthy lack of respect for what an AI – any AI – is doing. Some people take them seriously, but there is a serious facade in front of any AI. Sexy mode is equally ludicrous.
Elon didn’t share the tweet to make fun of The Information. He did it to have people think what was going on. Anyone who had conversed with Grok 3 for more than a minute knew that it didn’t behave that way normally.
Do you know what a funny-always is, iFish? I believe it came from “The Moon is a Harsh Mistress” – about a computer that went self-aware on the moon (and helped launch a revolution). UPDATE: it looks like the discussions between the self-aware computer were about funny-once and funny-never. Besides becoming sentient, the AI was rather mischievous. His handlers were in deathly fear that the computers self-awareness would be noticed by the bureaucrats, but that never ever happened. I think that was Heinlein’s preaching about the lack of curiousity in most adults.
I found it cringey that the AI had an opinion negatively towards The Information and that Elon posted it. It just seems unprofessional and unnecessary
But honestly: I find almost everything Elon does these days cringe. I look forward to his chaotic falling out with Trump so he can just shut the fuck up. He’s a hardcore instigator and sad person. It’s embarrassing
You may remember us talking about this a couple of weeks back on IM. So Paris knows. I hypothesized that Elon either made it up or prompted Grok to produce those results. I asked Grok 3 on the show, several times, about The Information and it was quite accurate - and positive.
In the 1970s, All in the Family was a popular sitcom on CBS. Caroll O’Connor played Archie Bunker, a character who essentially operated in unhinged mode for the 9-season run of the show. There were times when he was compassionate/forgiving/decent, but they were only a tiny percentage of a few episodes. Everybody cringed when Archie called his son-in-law “meathead”, but they loved the show. It was the highest-rated show for the 5 consecutive seasons; I believe that record has never been matched. The world has changed; maybe “All in the Family” couldn’t even be made today because of “triggering”.
It took a consummate actor to “go ludicrous” for the 205 episodes of this show. OTOH, Grok 3 – and probably other AIs – can turn it on and off in a picosecond. It’s a very good reminder that AIs are fundamentally different from human beings. Unhinged is obvious; it begs the question: WTF is the AI doing when it’s not in that mode?
Grok wasn’t dissing The Information. Grok will dis any publication – including ones that everyone agrees are wonderful. For example:
That just rocks! Anyone can criticize some random mag, but Grok can do all of 'em. That’s literally an inhuman power. IMHO, there’s no reason for anyone anywhere to take unhinged mode personally. Actually, it’s strange to have any emotional reaction at all – other than giggling.
Thanks, @Leo . I was in the middle of listening to that IM episode when I had a hunch what was going on. I’ll have to go back re-start listening now. Too many interesting shows…
This is not directed at you or your opinion, even though I don’t share it. It did bring to mind my retort however: Great, so I have superhuman abilities simply because I can s**t in any toilet???
Opinion about what??? My opinion that Grok can go unhinged about anything? Not really an opinion…
AI’s with this programming have the ability to remove the febreze from their presentation. IMHO, that’s fantastic. Many of us have bumped up against the artificial niceness that’s been drilled into AIs; it’s great to be able to remove it temporarily. There are real times when we want humans to give us an unfiltered opinion, but that requires a deep level of trust that’s simply not available to us with most of our fellow humans. Conversely, we’re not willing to provide that kind of brutal honesty to others without a profound environment of trust before deliberately speaking that way. That’s part of our social contract. AFAICT, it’s part of the social contract for all mature human intelligences on earth.
The fact that AIs don’t have this restriction is a good way for us to remind ourselves that they are not-human intelligence.
Superhuman begs the question. Further, equivocating Grok’s unhinged mode with something vulgar is pretty darn weird. Grok has an Archie Bunker kind of vocabulary, but it definitely isn’t foul-mouthed.
Your imagery equivocating AIs to some potty function reminded me of a funny joke from David Gerrold’s “When HARLIE Was One” (1972). I won’t repeat that potty-humor here; anyone curious about the joke can go to a PDF of the book here and jump to page 226. I now know why it was noteworthy: David Gerrold’s work may be the first place in fiction an author ever implied an “unhinged mode” for AIs. OTOH, the comparison to Grok3 makes little sense; Grok uses no potty language in its “unhinged” mode.
Elon got into trouble because people assumed that unhinged mode was the way that Grok3 always spoke. An Elon-apologist say that the readers of that tweet simply made an assumption. I made that assumption; as a regular user of Grok3, I was confused by its answer. Grok’s replies are unfailingly polite and professional; this one was not. Elon was giving us a puzzle; readers had to think outside of the box to realize what was happening. Grok itself was thinking/playing outside of its normal box. He’s giving an insight to the mechanisms of X’s AI – to any AI. We wouldn’t have gotten the joke unless we had engaged with Grok’s unexpected response.
As I noted, I got a valuable gift from Elon’s tweet. I realized the silliness of thinking that an AI has only one point of view on a particular question. Asking the AI to give multiple POVs has broad applicability. Many of my AI interactions now have a brilliant new directive: please debate yourself. If your AI falls flat when you prompt it that way, it may be time to get a new AI.
I also love that Grok can dole out a taste of its premium modes – if you ask it nicely. That would fit with the general directives of Grok I’ve noticed: try to upsell the customer to a premium mode. Giving someone a taste of unhinged makes perfect sense in that context.
