Zvi Mowshowitz & Mentorship Anti-Patterns

Damn bro

Zvi came to Inkhaven as a collaborating writer for a week. He was asked to do a number of things he evidently didn’t like, such as:

• Writing in public, so residents could see what his workflow is like.

• Holding office hours.

• Critiquing drafts of blog posts for residents who sought out his advice.

This all seemed to put him in a Mood. This mood consisted of being exquisitely and exceptionally rude to a whole lot of nice and interesting people.

Why am I writing about this? First, I reserve the right to point and laugh at guys who are systematically obnoxious to the people I like. Second, I think Zvi proved to be such a vivid anti-model of competent mentorship that analyzing his behavior is illuminating in relief.

Prelude: The Other Mentors

I first want to talk about the other mentors.

I was pretty astonished at the rate of work Justis Mills, Aaron Silverbook, and Georgia Ray put in. They all wrote some nice stuff, while giving out feedback and guidance to the residents, and also while balancing other life priorities (which for Justis in particular included admiring his cutie toddler full-time). One evening I was schmoozing with Georgia in the kitchen about what it meant to “get better” as a writer, and how one can learn how to “feel” the gradient of improvement. Justis wandered into the conversation and said some very moving things about what writing meant to him, which made it clear to me that a) he got that dog in him, and b) he was chill about it.

Getting to meet some of the LessWrong posters I’ve been reading since forever was wonderful. This definitely tipped over to the “meet your heroes”-side of that particular debate. Vaniver, Abram Demski, and Tsvi Benson-Tilsen have written many wonderful things worth understanding, but it is not necessarily clear from their writing who they are as people. My conclusion having talked with each of them: they are secular sages. They are each exceptional listeners, highly equipoised, beneficent, and calming¹²³.

Scott Alexander really stood out in terms of secular sagacity. One resident showed me an email of feedback he received from Scott which was two times the word count of the original draft. Another resident characterized the Scott Alexander feedback experience as follows:

Scott will open a feedback session saying, “I’m about to say some rude, but hopefully helpful, things.” He will then slowly and exhaustively work through your piece with you. It’s impossible to come away feeling offended in any way, because he would be so calm, diligent, and encouraging throughout that every criticism (no matter how damning) felt great to hear.

Gwern provides a very different model of great mentorship. He was hilariously harsh with his criticism. He would frequently tell people that they were “not even wrong” or that they “didn’t do their homework”. He definitely wanted the residents to try harder. One thing he said about someone’s writing will always stay in my mind (no spaced repetition required):

X’s blog posts are the most LLM-slop-sounding posts I’ve ever read in my life, which I nevertheless still want to read.

This was an uncannily accurate characterization of X’s posts. It is good criticism.

The residents loved him for this. Gwern was never contemptuous, and never seemed upset or put-upon. He made himself highly available, and was willing to work with people for quite a long time in order to get them unstuck.

Enter Zvi

Zvi really put in a Herculean effort to be maximally repellent to the lowbies of Inkhaven. I personally avoided interacting with him, but out of curiosity I went around asking my fellow residents, “so what’s going on with Zvi?” After soliciting impressions from about a dozen people, I stopped; all the responses were pointing in the same direction, and merely mentioning Zvi was vibe-killing. The gestalt I came away with was that Zvi basically did the “I saw X at the grocery store” copypasta to 40 people:

Damn brother, I didn’t think my drafts were that bad!

Zvi crapped on everyone’s writing, and his feedback was not insightful. One resident said⁴,

I brought my writing to Zvi and he just sat there shitting on it the whole time. I told him, dude I already know it is bad, we have to write a blog post every day. He just didn’t tell me anything that was actually actionable. He just came off so scornful, and had this aura that his time was being wasted.

The guy saying this is a highly agentic, sharp fellow whose writing is really good⁵.

Someone else:

I got some feedback from Zvi, and it was not useful. He has shallow mental models.

Someone else:

Yeah Zvi is really quick to type people into the worth-his-time and not-worth-his-time buckets. Set expectations accordingly.

Someone else:

I heard Zvi bitching at Ben Pace about how coming to this event was a mistake and that the residents were all writing complete junk. He didn’t seem like he wanted to solve anything, and was just taking the opportunity to excoriate everyone. I wanted to knock his head off his shoulders.

Someone else:

With Gwern, there’s nothing personal about the negative feedback. Zvi just seemed like he found me, personally, contemptible.

This all got people to avoid interacting with him, which in turn reinforced his (now-accurate) feeling that his time was being wasted. I started hearing things like this:

What sort of writing advice would I go to Zvi for anyway? How to summarize three tweets?

And:

He’s just a summarizer now. His work isn’t generative.

My sole interaction with him was the following. Three residents and Zvi were circled outside a fireplace for what ostensibly was Zvi’s office hours. The residents were talking amongst themselves, as no one was actually coming to Zvi for advice anymore. I joined them and started yapping about random stuff. One of the more empathic residents piped up and said, “so does anyone have any questions for Zvi?” At this, I launched into a series of questions, which Zvi proceeded to no-sell as hard as he could:

Ah yes Zvi, I’ve been meaning to ask you. This may not be a question that is meaningfully answerable, but I’m going to give it a try anyways. I was just admiring the work Balsa did on the Jones Act Vessels. I don’t understand how you can have such high throughput on your main blog while evidently having many other irons in the fire. I’m really hitting my limit here at Inkhaven just trying to keep up with the daily posting. This has made it difficult to work on effort posts concerning technical subjects. How do you maintain high productivity while working in multiple modalities?

Somewhere during this, Zvi broke eye contact with me, leaned heavily on the sofa and squinted into the distance over his right shoulder. He was searching my question desperately for signs of intelligent life, and failed to find it. He mumbled something back at me. I think he just said “time management”.

Seeing that I failed to rise to the level of sentience, I tried again:

Okay. So I’ve been wondering about when and how to do deep work. For instance, I feel like I can read pretty much any ML paper and figure out what’s going on in it. Nevertheless, I feel like my mental models here are shallow. Meanwhile, I have this friend who does Quantum Computing at Google. He has a PhD in physics, so yes he is smart, but the thing in particular that is wonderful about him is that he has these deep intuitions about finite dimensional vector spaces. I think this is because he has seen linear algebra in so many contexts that it has given him a “sense” that defies easy characterization. He has seen linear algebra in quantum mechanics, he has seen linear algebra in algorithms courses, he has seen linear algebra in graphics programming, he has seen linear algebra in machine learning, and so on. It would take many years of serious, deep work to get those intuitions into my head. Does it make sense to step away from the daily grind in order to cultivate such things? How do you think about these sorts of tradeoffs.

Zvi thought for a second, and then said, “it sounds like your friend did more reps than you.” He then walked off.

“Yeah, fuck me, I guess,” I said to my fellow worms.

Failing Gracefully at Mentorship

A failure mode of mentoring is giving insufficient pushback and reinforcing the wrong values. Herein lies the value of the person who is willing to say “you are doing poorly”, “try harder”, and “your position doesn’t make sense”. Praise and cherish such people when they come into your life.

But sometimes people do the negging and they either don’t care about you improving, or they are themselves being stupid and giving bad advice. This could be you, and you may not even realize it! You may think you care, when you really don’t. And you can definitely give stupid advice without realizing it. Especially as you get older, you are more likely to be in this position.

Herein lies the value of being nice. It allows you to fail gracefully.

1. In the first week at Inkhaven, I struck up a conversation with Vaniver about his piece on dating and date-me docs. We proceeded to have a forty-minute conversation exchanging data and working through each other’s ideas. Someone came by to get Vaniver’s attention, and he gently said something to them to the tune of: “I will enjoy discussing this with you soon. However, for now I need to finish this conversation.” I was not being particularly interesting, but Vaniver was highly committed to getting to the heart of the matter with me.

2. In the second week, I read and enjoyed a blog post by Benson-Tilsen on bouldering being a rich source of body puzzles. The piece suggested that Brazilian Jiu Jitsu may be a borderline case of body puzzling. I strongly felt that BJJ is, in fact, peak body-puzzling, and so I was excited to talk to him about it. I went to sleep, and upon waking up I forgot that Benson-Tilsen wrote the piece, and came instead to believe that Abram Demski wrote it. At lunch that day, I was sitting next to Abram. During a lull in the conversation, I turned to him and said, “So, I do think that Brazilian Jiu Jitsu is in fact a rich source of body puzzles.” Without missing a beat, Abram replied, “Ah! How so?” He then proceeded to let me yap about BJJ for ten minutes, and all the while he sat there asking me encouraging questions. The next day, I realized to my horror that I had goofed, and rushed to apologize to Abram and explain the context for my mistake. He laughed, and said, “I thought something like that had happened. Thanks for telling me about Jiu Jitsu, it was interesting regardless.”

3. Benson-Tilsen is a bit of a pugilist on LessWrong. What doesn’t come through there is how calm and patient he is in person. He really is a joy to talk to. He has deep mental models and many well-considered frames, but when he talks to others he is mainly there to solicit information and insight. He will provide gentle pushback and will seek clarification frequently, but he does not strongly assert his frames (perhaps for Tsvi, talking is for babbling and writing is for pruning). He is also funny and modest. My first conversation with him was in the kitchen of Aumann Hall, where he said, “I’m Tsvi. Not the Zvi. I’m the other one.” I was there thinking: yeah dude! You wrote Formalizing Convergent Instrumental Goals. I know who you are. You are not the “other Tsvi”!

4. None of these quotes is precise. They are anonymized and paraphrased.

5. Also, he’s a compelling character. I’ve been thinking of him in my head as a “John McAfee”-type guy, but a millionaire instead of a billionaire, and without the Dark Triad personality structure that usually comes from being a “John McAfee”-type guy.