Thank you Yak Collective!
Thank you Yak Collective for such an interesting year! Especially in studying how to develop assets within a decentralized large-online community.
This newsletter has evolved and changed hands within the year and will continue to ebb & flow as we work on formalizing its structure.
In that same vein, I’m interested in YC members who want to contribute to this newsletter, please contact me at praful.mathur+yaktalk@gmail.com.
Announcements
Yak C Holiday Meetup
When: Thursday, December 10th, 8AM-10AM Pacific
Topic: Yak Collective Annual Meeting and Holiday Party
Where: Zoom Meeting. Meeting ID: 875 5826 3911
What: Agenda
Revamped Project Proposal Update
Project Proposal Revamped Process (no more google forms and synchronized competitive sign-up)
All you have to do to propose a project is to add a block to this Roam page with the required meta-data, and propose a channel in #create-destroy-channels... that's it. If you get 2-4 participants before your self-declared start date, it goes live. The Friday chats will also serve as a continuous pitching forum going forward, but right now, there's an opportunity to get a bigger audience for a live pitch. If you put in a proposal before Dec 1, you can get a pitching slot at the YC annual meeting on Dec 10. First come first serve for pitching slots.
Indie Consulting Advice
How I Billed This
Fortnightly feature to share brief account of indie consulting, focused on monetizing expertise. Please contact Paul Sas if you’d like to be interviewed for an upcoming newsletter. The following is an interview with Paul Sas.
What’s a tweet length description of your niche & expertise?
Paul Sas: I’m trained as an experimental psychologist, who rebranded in the noughties as a behavioral economist. I work primarily with tech companies on behavior design (hooks, nudges, smart defaults, reframings).
What particular area(s) do you consider yourself to be tops at?
PS: Experimental design seems hard to do right. All first year Stanford psych grad students had to design one study. They uniformly yielded null results. In 3 years collaborating with Dan Ariely, he said he’d never seen a well-designed experiment come from all the MVPs he’d reviewed in Mountain View.
Case study win
PS: A fintech company, Varo, pitched to VCs a (then novel) use of a conversational interface. I designed an experiment, run in Mechanical Turk, to test 3 different styles for apologizing when the chat bot failed. The experiment sent Turkers to a Slack channel to interact with a chatbot, programmed in Python. The bot was designed to fail, and then used one of three apologies (A- taking responsibility for its limits; B- promising to learn next time; C- committing to update the programmers).
How did you first connect with the client? What was your pitch?
PS: I’d previously worked with the chief product officer, who introduced me to the CTO. I agreed to deliver the experiment and data analysis for $10K. Since the results were used in their pitch deck, we agreed to structure a bonus (up to 250%) based on the amount they raised.
Did the contract lead to an ongoing relationship?
PS: I did develop a year-long monthly retainer gig with the Design group (Robots & Pencils) that was working on their screens.
What is the one thing you haven't done enough of to market your consultancy?
PS: Spec behavioral research in advance of cementing a contract. Many companies can’t directly “experiment on their customers”, nor do they want to directly commission research about sensitive areas such as emotional engagement. I’ve put off doing a benchmark study on intrinsic motivation (‘flow’) to evaluate the different experience in big software workflows.
Another Astonishing Story
Your Payment is Ready
…my algorithm finds that you are currently at the 46th percentile amongst clarinetists in your peer competition group. Practice your arpeggios to improve your score before the final evaluation! 16-year-olds in your city with finely articulated arpeggios generally rank in the top quartile. Those with a strong swing-eighths style generally rank in the top decile. Say ‘practice arpeggios’ or ‘practice swing-eighths’ for a lesson.
“Judd!” His father yelled as he walked into the garage. “Why the hell is your Clarinet-MLv2 in pieces? How are you going to make regional youth orchestra with a broken machine learning clarinet?” In-person socializing among children had become another luxury good, doled out to high achievers and those who could afford the best virtual instruction. In an age of pandemics and rolling lockdowns, his dad saw the youth orchestra as an opportunity for kids like his son to get out of the house.
“Whoa!” Judd put up his hands to deflect his father’s anger, “I can explain.” Shunting his father to the corner of the garage and away from the decomposed clarinet, he began breathlessly, “so I’m practicing yesterday like normal, getting feedback from Clarence, the clarinet’s evaluation algorithm. But I notice it just keeps repeating the same thing, in the same patronizing British accent. And I mean, No. Matter. What. I. Played. I practiced my arpeggios for two hours, and it tells me I’m 46th percentile. I play a series of high-pitched squeals: ‘Practice your arpeggios…’ I even take the thing apart and remove the analog-to-digital converter so I can feed it recordings of Benny Goodman. “You’re at the 46th percentile. Practice your arpeggios.” Judd throws up his hands in exasperation.
Listening, Judd’s dad pursed his lips and thought back to how lucky he’d felt to snag an early-model machine learning clarinet in the lottery his son’s public school ran last year. Judd had zero musical aptitude but he received one of the few instruments they had available for enhanced musical instruction. Even though the spit valve was rusted shut and the software was crude, he’d been happy to put down the hefty deposit. He looked over his son’s shoulder at the dismembered clarinet with, as the commercial advertisements claimed, “built-in machine learning software that provides personalized teaching and testing with no dangerous human interaction necessary!” He worried they might lose it if Judd was unable to put the thing back together.
Meanwhile, Judd was still speaking quickly. “So, I start digging into the software settings. I tune a few of the hyperparameters related to…” His father has no idea what he’s talking about and is skeptical that Judd does either. “…and then something crazy happened.”
His dad’s attention refocuses, and he sees his son smiling broadly.
“The company doesn’t support this model any longer, but I came across an unofficial firmware update that promised to fix some of the problems I’m having. I install it and reboot. Clarence is back to his old, obnoxious self again! I play the arpeggios I’ve been practicing, and this time he tells me I’m ranked in the top 5 percent!”
Judd looked at his father expectantly, “I think there was just some kind of buffering problem. It wasn’t capturing the new input and so the algo was slow to update.”
Judd’s father’s eyes brightened in surprise. “Really?”
“Clarence even said I was eligible for a statewide scholarship!”
“Judd, that’s wonderful.” His skepticism took a sudden backseat to the prospect of reduced tuition. Maybe Judd wasn’t just fooling around out here.
“I know! They said they’d send a cheque. All we have to do is deposit it in the bank, then wire half of it to a numbered account which they’ll provide separately. That will cover tuition and I can keep the other half for living expenses.”