Discover more from Ben Mann Monthly
Ben Mann Monthly June 2022
Red-eyes, antidepressant update, major milestones, lazy carrot cake, altruistic ambition
An index for my memory
A menu of topics for my next conversation with you
A faster way to share what I’m excited about without the barrier of writing a complete blog entry on it
A skimmable way to spread content I found valuable
Experiments and experiences
Brought my clarinet to a housewarming and spontaneously started a jam session with an accordion, a viola, a guitar, and a piano. We did some covers and some standard jazz chord progressions. So fun! I didn't know their names or anything about them until after 😅
Note to self: never take a red eye from SF to NYC. It's not worth being tired for a few days after from only getting 5 hours of sleep. Maybe it's worth it if you can go sleep somewhere immediately afterwards. But it's so cheap and it feels like you're saving time by sleeping in the plane! When we have rocket transit or supersonic jets I'll be happier.
Followup on my depression post
In January 2020 I posted My depression and the magic of antidepressants. ~3k reads later, when I tell people I'm taking antidepressants, I usually refer them to the story to hear all the details. I've now been taking antidepressants for almost two years. Once I decided I was in it for the long haul, I worked with my psychiatrist to try six different options: tianeptine, escitalopram, venlafexine, sertraline, bupropion, and fluoxetine. I tried each for at least 6 weeks except for venlafaxine, which might have given me bloody diarrhea for two weeks. Each time I hit the 6 week mark, I met with her to discuss how it was going - symptoms, side effects, mood journal stats, etc. When I found something I liked, I tried it for a few months. Sometimes it didn't work well and I was sad for a few weeks. It was worth the short term pain of experimentation. I settled on fluoxetine, which I've been taking for about a year. For me, it has the least brain fog, sleep disruption, sexual effects, dry mouth, and plenty good mood and anxiolytic effects. A few times, I've tried tapering off to see if maybe I don't need it anymore, but I do. Within a week I'm back to how I was before I began experimenting with antidepressants. It's not good. Taking a pill every day and avoiding grapefruit juice is easy and worth it. I've also found that once I mention that I'm on antidepressants, an astounding number of people say, "me, too!" This stuff works, folks. Others tell me they tried an antidepressant many years ago and didn't like it due to side effects, lack of success, or something else. Yep, some of the ones I tried didn't do anything for me. I was lucky that the first one I tried (escitalopram) made me feel great on day 0. Don't give up! Work with your doctor to keep trying. If you find something that works, it'll change your life.
Note on life events
A friend mentioned that in past newsletters I've mentioned momentous life events such as raising a round, getting married (no wedding party), or expecting a child in passing. I'm intensely proud of these events, but I purposely mentioned them in a low key way to avoid drawing attention. I've considered writing posts on eg the desired properties of wedding rings (like my cheap $20 choice), or how we chose to have a child, but I haven't made time for it yet. Talk to me in realtime and I'll be happy to share my experiences! Otherwise, patience, I’ll write about it in due time.
5 point Likert ratings for “I would recommend this content to a friend”, sorted
Cribsheet 5/5 for expecting parents
Sequel to Expecting Better. Once your baby is born, what does the research say matters?
Breast feeding vs formula - really up to the mother. For some women it's very painful to breast feed. For some it's easy. Emotional connection of breast feeding can be important.
My mom read it and said the biggest difference between it and the books around when she was learning this stuff is that the wellbeing of the parents now matters too. A sleep deprived, suffering parent isn't going to provide as good care.
There's a lot of "if you don't do X, you're a monster." Usually the data doesn't support this, except sometimes.
Don't sleep with your baby. Put them in an empty crib, tie them up with a swaddle.
Day care vs nanny - research shows nanny early and day care later leads to the best outcomes, but it's not very different and it's more important what works for your family.
Sleep training works. It's fine if you're in the room too.
When transitioning to solid food, kids need to be exposed multiple times before they'll try a food. Be patient, keep trying.
Parents are going to get sick from their kids constantly.
Late walking is fine.
Be patient with potty training. Research says it's hard to accelerate date of final training.
Reading to your kid is super important. Make it interactive. TV is neutral or bad, regardless of whether it's educational, but if you need a break as a parent it can be a good distraction.
Documentary of the Russian opposition party to Putin, with excellent commentary direct from the source. What a character!
Gets poisoned, recovers in Berlin, then goes back to Russia and immediately gets arrested and thrown in prison.
Putting on my skeptic's hat for a moment, this seems like exactly the kind of thing the right would accuse the left of doing in the US. It'd be easy to fabricate the phone call they recorded with his supposed poisoner. But given Putin & Russia, I'm inclined to believe it.
I basically skipped all the steps here by throwing the ingredients in a bowl in random order, mixing, and baking in a 13x9 pan. Used a cuisinart to shred the carrots; toasted chopped walnuts instead of pecans; melted butter in microwave; replaced the whole wheat flour with equal weight whole oats. My brown sugar was rocky, but I threw it in anyway without breaking the rocks. Skipped the frosting. Turned out great!
If an apocalypse suddenly struck humanity, such as a pandemic that killed 99% of people, how would we most quickly recover? What would we need to know and what tech would we use?
Metals are easy, just melt the skeletons of the giant skyscrapers. Concrete also easy.
energy is hard. The easy dense energy sources like oil have mostly been extracted. We'd probably want to jump to wind and concentrated solar power quickly.
Some simple tech such as microscopes took a very long time to discover and realize their importance. Make sure we save those.
Seeds are really important and easy to lose. Svalbard already flooded.
Information security is really hard and there aren't that many people working on it
I personally don't think formal verification is effective. Most attacks are just social engineering, so making that harder using eg two factor auth is the most important thing.
NVIDIA hack was a bunch of teens. Hacking is pretty easy if you try hard, but there's not much economic incentive, and it's pretty easy to get caught too.
If you like the stories here, go straight to the source instead: Darknet Diaries.
Most people aren't very ambitious, so if your movement wants to do the most good, trying things that'll likely fail is comparatively effective.
Suffering is not necessary for doing hard or important things!
When you're trying to get rich, once you're rich, life gets easier. When you're trying to maximize long term impact, as your impact and leverage grow, life gets harder. Still have to take time to recharge. Self care, therapy, and antidepressants are great.
Patient philanthropy doesn’t make sense when you expect a lot more capital to come in and need to get practice deploying it
Conflates supervised learning and reinforcement learning. Oversimplifies reinforcement learning.
Otherwise feels like a shorter, updated version of Superintelligence
Doesn't consider very short timelines, in which the compute overhang from training will be much less than long timelines.
Seems unlikely to me that AIs would coordinate against us, but then again the whole piece is somewhat premised on a Pascalian mugging
All that said, I think organizations that don't care about safety could proceed down a path similar to the one he describes with unbounded optimization pressure and negative rewards. Anthropic will have to figure it out before those orgs do, and then show everyone it's strictly better to use our methods.
The New Fire 3/5
I skipped the first half of the book, too basic for me. Second half was interesting. A lot of it was targeted at policymakers. What does the US need to do to stay competitive with China?
I like the hopeful attitude.
3 competing groups:
Cassandras - safety-oriented people predicting disaster
Warriors - people who want to apply the tech to offense
Evangelists - tech-types who think there isn't much risk and we should build capabilities as quickly as possible
Good coverage of the 2013 Cyber Grand Challenge, though the stuff they called AI back then was very different from current methods.
Coverage of GPT-3 and large language models is a little dated given our recent alignment work
Completely agree we should open up our borders and brain drain the rest of the world, but doesn't seem like there's much appetite for this in Washington 😢
Undone S2 3/5
Love the animation. Story was ok. Seems like they could've done more with the medium and the talent.
Thanks for reading! Subscribe for free to receive new posts.