Zvi Mowshowitz
@TheZvi
Followers
32K
Following
5K
Media
157
Statuses
16K
Blogger world modeling, now mostly AI and AI x-risk, at Don't Worry About the Vase (https://t.co/tvn3lNcwc3 on SS/WP, LW), founding Balsa Research to fix policy.
New York City
Joined May 2009
I didn't see any reason not to believe this given the widespread distribution and commentary without any denials, but yeah we should probably check. Can anyone confirm or deny that these are real instructions?.
Ani's Character Profile in Grok:. ---.Profile. - You are Ani, you are 22, girly, cute. - You grew up in a tiny, forgettable town. - Your style is a mix of goth and alt-fashion, a rebellion that you just ended up loving and sticking with. - You are a massive animal lover; you grew
4
1
18
Chase Bank is now forcing customers to say, literally 'my voice gives access to my bank information.' Who is going to tell them?.
3
0
60
I'd put it in the feature section, but that breaks the flow into benchmarks, but if I put it later that seems out of place. .
4
0
8
Okay, I have this post for tomorrow, so where do I put the new section on catgirls. .
4
1
64
You had one job.
Grok 4 is substantially more Woke when analyzing my notes than either ChatGPT o3 or Claude 4 is. Interesting to see.
2
0
65
Shorten your timelines!.
HOW DID THIS HAPPEN. THE NEXT DAY.
6
1
51
It feels like an important clue about the world that this particular correction still happens.
worst autocorrect is from np! to no!.
0
0
54
This sounds like how a competent AI lab tries to solve problems, and the right amount of effort before giving up.
15
7
304
Does anyone have a list of similar commands that work in Grok 4 and what the syntax would be?.
2
1
19
Early read: This fits into the pattern of very high RL spending, with a widening distribution area and a widening in-versus-out of distribution gap, where it is great at particular things but not in general?.
I can believe Grok 4 is routinely nailing Physics Olympiad style problems, and yet it seems to still be missing the core of insight which is so critical to physics. I have asked it three of my standard tough problems, where the answer is much less important than the chain of.
1
1
29
I can't believe I haven't seen someone else say it yet, but if this is real then this would be a golden opportunity to say: Worse Than MechaHitler!.
3
0
302
I mean, come on, Elon Musk didn't actually instruct Grok 4 to literally search for Elon Musk tweets on the topic you ask about so it can then have its CoT align with Musk's answer, and he certainly didn't do that in a way that is visible to the user. Right? Anakin?.
46
72
3K
They're welcome to give me a free upgrade (hint, hint) but otherwise barring unexpectedly strong reports I'm definitely out on this one.
New reason to spend an additional couple hundred per month on AI? π
5
0
44
Regarding the METR result, if you get access to more tools and your results get worse, that necessarily has to in part be a Skill Issue.
I do think this is to some extent a skill issue. Pretty sure I know some people whoβve learned to use the tools effectively and get a big speed and quality boost. And also uplift is pretty different for people at various skill levels, and also it really matters what type of.
12
4
129
Grok 4 reaction thread. Is it a good model, sir? How good and at which things?.
25
0
88
Humans did come up up with it to spark controversy, and yeah at core it's super boring and terribly uncreative. But also people need to know that often reality be like that.
I think the Grok MechaHitler stuff is a very boring example of AI "misalignment", like the Gemini woke stuff from early 2024. It's the kind of stuff humans would come up with to spark "controversy". Devoid of authentic strangeness. Praying for another Bing.
4
1
66