Ben Mann Monthly Aug 2021
Time management, tri training, chip shortages, what is "human"
Purpose
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
Idiana & Greg's wedding
Because they cancelled their wedding twice in a row (COVID 2020, forest fires 2021), they decided to have a lower-key impromptu wedding on a weekend in SF. Friday night dinner at a restaurant, Saturday picnic in Dolores Park, and Sunday ceremony in a small park, reception at a friend's house, dinner, cake and dance party. Portable PA was essential. While I had a great time, especially meeting new friends, for my own wedding I think three days of activities feels like too much. Excellent DJ'ing on an iPhone. I haven't danced so hard since before the pandemic and was covered in sweat by the end.
Time management strategies
I've tried many different time management strategies, but the best I've found by far is a public
#daily-updateschannel in Slack or wherever you talk to your team. At the beginning of your day, make your usual TODO list, ordered from soonest to latest. You're not allowed to copy-paste. If you noticed something sticks around for more than a few days, consider decomposing it into more bite-sized tasks or finding a way to not do it like delegating, finding an alternative, etc. If your day looks extra full, mark some of the tasks as[stretch]. If you plan to do the task with someone else, mark it as eg[w diana]. As you burn through your list, edit your post with your progress:👉 Working on this task now
🌓 Partially complete, but stopped working on it
❌ Decided not to do this task ever
🕐 Didn't get to it today
✅ Done
The list keeps the urgent and the important balanced, creates transparency for you and your coworkers, and let's you focus on burning down a list as fast as possible without having to wonder what you should work on next throughout the day. At the end of the week, you can look back through your updates to provide your manager or reports with an easily consumable snapshot of how you've spent your time. At the end of the quarter you can calibrate on whether you spent your time in service of your OKRs.
In organizations where I practice
#daily-updates, it tends to be contagious. People see how much I'm getting done, pitch in to help on tasks where they have expertise, and ask clarifying questions when they're unfamiliar. When many people participate, the sense of shared team momentum is palpable.
Emotion rating {-3, 3}: mean .5, std .5
Highs: quarter triathlon, dance party, negative echocardiogram, work sprint coming to fruition, office opened, hammock/read/nap day in San Mateo
Lows: hiring is hard, Max flew back to Finland, occasional bad air quality from fires, random sleepy days
Life updates
⚡️ Laser #2
🏄♂️ Surfed at Linda Mar
🚴♂️ Butterlap, San Bruno Mountain
🏃♂️ Quarter ironman
📚 Started reading The Hidden Girl and Other Stories
💼 Hiring
Content
5 point Likert ratings for “I would recommend this content to a friend”, sorted
Despite being a little dated by the development of strong language models, manages to cover a wide range of important and relevant topics.
Primary topic: the Turing test, in which a human judge must decide from a short text-only conversation with an agent whether it is a "human confederate" or a chatbot. The author attempts to become the best "human confederate" by studying the foibles of chatbots, and the strenghts of humans, and succeeds. This has broad and important relevance to human to human conversations, which perhaps increasingly seem like they could be replaced by automatons. "How've you been? Good. You? Oh, same same. Great weather eh? Yep." Ugh. Can't we do better? Do we want to do better?
Favorite quote: "He who cannot draw on three thousand years is living from hand to mouth." -- Goethe. In other words, don't just ask your friends what's new, ask them about their deep past. Don't get bored by the news, read some history or literature.
Does AI being able to do a task make it less human? Most people seem to think so, but Christian points out that historically this has been a moving goalpost (eg chess), and it increasingly leaves us with nothing to hold onto. I'd argue that in the same way teaching our kids doesn't take away from us as parents, AI capabilities shouldn't diminish our humanity.
Reminded me of asking my therapist what the difference was between having a conversation between peers vs a therapy session. She posited that it's about sharing about yourself in a balanced way. The book covered chatbots that failed in both ways: taking too much or too little stake in the conversation. It should be constantly readjusting dance.
A big difference between chatbots and humans: chatbots wait for their turn, whereas in a normal human conversation, people talk over each other, use body language and disfluencies to indicate turn dynamics, and generally are "barging in" all the time. How might we incorporate these dynamics in our AI systems?
Chris Olah on what the hell is going on inside neural networks 4/5
Chris gives an overview of the current state of the art in interpretability research, our dreams for the future, and the most promising frontiers.
We have a very good sense of what's going on in image recognition models. Now we need to apply that to language models.
It's important to work with large models because they will have the most relevant capabilities to explore, and often have cleaner representations internally.
Data visualization experts are hard to find.
Every software engineer should watch this impressive demo of OpenAI's AI translating natural language into live updating code.
The examples were obviously chosen to demonstrate the power of the system in that each code snippet could be expressed in just a few lines. They noted that more in depth requests often resulted in failures or mistakes. Even so, I find this extremely compelling for multiplying my efficacy as a programmer. I'm constantly forgetting the details of various APIs (how the F do I make a colormap in Matplotlib again?). The Codex model has that down pat, so I can just tell it the structure of my data and it'll nail it.
This is only the beginning. I'm sure they'll collect all the human feedback and get a strong flywheel going. Especially beginner programmers can use this as a learning tool, and soon non-programmers will be able to use it too.
As I train for triathlon, many people recommended Total Immersion. It turns out no professional swimmer follows Total Immersion, and instead they alternate between 4 styles: horse, boat, arrow, and windmill.
I also watched a deep dive into arrow, which is good for distance swimming. It already feels like a completely different game than I used to play. I finally understand the sync between feet and hands, and feel the rhythm of the stroke more intuitively.
Swimming with a wetsuit on improved my form dramatically thanks to keeping my legs and upper body close to the surface of the water through buoyancy. I felt like I was in a body-shaped boat, with paddles for arms.
Why the global chip shortage is making it so hard to buy a ps5 4/5
Supply chains are tricky. If you need 1000 parts and only 1 is missing without a backup, you can't deliver your car or whatever you're trying to build.
Many manufacturers have started hoarding parts so they won't be caught offguard, but this exacerbates the shortages for everyone else.
It's pretty difficult to switch out some parts because there's usually not much profit to be made, so supply can't ramp up quickly.
There's a story that most the important silicon comes from Asia, but actually a lot is now being produced elsewhere, with more to come. It's just that the most advanced process nodes are concentrated in a very few places, and these are used in the cutting edge devices like phones, laptops, etc. The older stuff like power window controllers are produced elsewhere with older technology.
A Taxi Driver 3/5
Same lead as oscar winner Parasite, traces an unwilling taxi driver's involvement in publicizing the Gwangju Uprising
After the film, Diana fact-checked. Turns out the real taxi driver had been working with the journalist for a long time, and knew what he was getting himself into from the start. The "self centered asshole learns to be a good human" story was fabricated.
