reflection

July Update

This month’s update is coming a little bit early because I will be traveling next week, quite possibly without a laptop.

Some personal un-fun news to get out of the way first:

  • I have inflamed tendons inside both shoulders since my travel to the US last month, such that I cannot perform many basic tasks without experiencing incapacitating pain, so I feel like a cripple a couple of times every day. My physiotherapist expects at least four months of recovery.
  • I separately developed Tennis elbow due to—wait for it—excessive reading in the last few months. It turns out that my body does not tolerate books as well today as it used to when I was a teenager. Bleh. My physiotherapist expects between 3 and 48 months of recovery (!!). It is still unclear how I will continue to read, since I do not enjoy audiobooks. To compensate, I adapted my Youtube sessions using a base filter to select only 20+ minutes videos, which I listen to in the background of other tasks. More on this below.
  • my mom experienced a “minor stroke”—her words, allegedly forwarded from her doctor—and I will need to travel later in August to assess what’s next for her (and for us).
  • expectedly, recent blood work revealed/confirmed that I have inherited some cholesterol-related genetic issues from my bio parents and time will likely come this year to start the corresponding treatment, to accompany me for the rest of my life. (It will not be my first lifelong medication, but I will still shed a sacrificial tear at every new marker of senility.)

All these things are under control (or fully escape any hope of control) and so there is not much more to add to them. However, there is plenty more actual fun that was had this month.

❦❦❦

Since the house project is on the rear view mirror (or something like that, more on that below) time is coming to start doing something “hard” again. There are a few criteria that need to be met:

  • I need to feel like it is a tough mountain to climb (emphasis on “mountain”/up and “climb”/develop, instead of “desert”/alone and “survive”/maintain);
  • it should teach me something about myself;
  • it needs to create some optionality, including in ways geographically close to home. As much as I like creating optionality across the ocean, I now want it to be more balanced somehow.

Developing a side-project into a viable business on my own and making that a primary project would feel too much like surviving a traversal of the desert, and not enough like climbing a mountain. So I would prefer a primary project to be a shared endeavor with other like-minded folk.

So as a stepping stone, I selected a medium sized hill to climb as a kind of “warming up”: investing into my own “networking,” as in, meeting new folk without aim to turn them into friends or coworkers. Networking is hard (for me) for various reasons which are hardly relevant here. I wrote about them elsewhere. It is still worth working on though.

As part of this process, I registered to various platforms and put some feelers out about possible gatherings in my area, and already found a few people that made me very happy to chat with, and with whom I hope to spend more quality time together. Since I regained access to my own decent venue to host e.g. serious dinners, I feel I also unlocked a few privileged networking dynamics and I fully intend to explore that.

❦❦❦

Relatedly, I discovered a service called “Vibend”. It purports to facilitate networking. In a nutshell, one pays an event fee (EUR 7.50, probably US $10) which gets one “matched” to a group of other customers^W “amazing like-minded people in your city” and receive an invite to hang out at a fixed place and time, usually at a cafe after lunch or a restaurant for dinner. The rest is up to the group (including paying for their food/drinks).

This was an enlightening experience!

On the one hand, I tried it out: the meetup experience is mediocre at best. The matching is poor. The resulting group dynamics are all over the place. There is no moderation. I did not get the impression that participants really wanted to see each other again, and repeat visitors mentioned that they come to meet new folk every time, instead of forging persisting connections. “Make new friends offline” is the tag line and I did not see it happening.

On the other hand, I studied the business itself, and the execution is impressive. The marketing campaign on social media is very good (9/10). The web site is rather good (8/10). The sign-up, followup and feedback processes are streamlined by e-mail and very polished (10/10). Geographical selection is superb (10/10) - Amsterdam, NYC, Paris: locales with a large expat community and migration throughput, where even without repeat customers, there will always be a steady supply of new customers. Given how hands-off the owner of the business is and how little tech is needed, it’s likely 80%+ profit. I’m also pretty sure the product is too cheap, but as long as the median individual experience continues to be mediocre they would attract bad reviews if the price was higher.

Why this got me excited: I recently had an idea for a similar service and Vibend would be serious competition. Competition is good! Competition means there is already proof of the business model. Competition also means a source of inspiration to get started, both on process and pricing. And I understand the 3 locales probably just as well as the current team. And… I already have a list of folk who have told me that they would rather go to my meetups (if/when they exist) than those from Vibend.

This feels like optionality.

❦❦❦

Here is another fun thing that happened. One of the remaining house projects is to improve my yard. The project is too large to do on my own so I need help. This is how I discovered Yardy. I am not yet sure I want to work with them, but it did not stop me from studying their business plan.

Purportedly, they offer “Ikea for yards”: conceptually, there is an online catalog of possible designs with clear prices; you point to what you like, and they make it happen—materials, labor, guarantees, etc.—within a fixed budget envelope. This is a stark departure from the usual Dutch market where the customer needs to think hard of what they want first, explain that in detail to one or more gardeners to obtain a quote, haggle on that quote, haggle on material sourcing, and then deal with unreliable delivery timelines, no guarantee on outcomes, and unpredictable budget overruns.

The customer funnel is also pretty effective: they have a sharp marketing campaign going on (9/10, super convincing), and they onboard leads with an enticing proposal: you get a “free consult”, after which you are free to either use them or only use their recommended design and “bill of materials” to do the work yourself, or with some other business. The asterisk goes after “free” in that the consultation is really €250 and is waived if you let them do the work.

Anecdotally, the value proposal is clear (originality: 8/10, market fit in the Netherlands: IMHO 10/10). I know intuitively there is demand. I will assess the effective demand when I go and visit their office for my “free consult”: if the office feels busy, that means they can at least keep the lights on with just design consultations (€250 per hour is no chump change) and once the customer has paid the cost to travel to their office psychologically (it’s hard to reach!), it must be soooo much easier to up sell them to the full package.

Their library of pre-made designs, incidentally, is decently priced (compared to self-run projects), and if they stick to their library they probably have economies of scale on materials and onboarding / up-skilling for new workers. This is a pretty decent business plan. I hope I get a chance to chat with the owners to learn more; it looks like it is still a small team.

❦❦❦

There’s also two other kinda-fun things that happened in July.

One is that I got selected to participate in a kinda-sorta startup incubator run by my city council. It is a little bit a funny story.

I had learned about it from the local newspaper. It was so incongruous (a “one-year program for entrepreneurs” with all the attributes of an incubator, but with the word “incubator” missing, ran by a city council of a community of 30k people) that I didn’t really take it seriously at first. I could not really motivate myself to apply until the deadline was near. But I still saw this program as a good networking opportunity, so in the last 20-30 minutes before the deadline, I conjured a few random thoughts about a business thesis and true experienced facts, put them down in the form, and this got me invited to give them a pitch.

With a 40% acceptance rate and an unknown line-up of mentors, that pitching session did not feel like something I should put enormous effort into. However, I also do not like to appear like a fool, and I wanted my thing to be at least as good (and ambitious) as all the other candidates. So I did spend an additional 20 minutes to jot down a narrative in my notebook before the pitch session (no slide deck) and put myself in the right mindset, and it just worked.

Note that up to that point and even later, I had no idea whether I was really serious about this business thesis and whether I really wanted to do anything about it.

However, a week or two later, I was chatting with some unrelated company founder and they taught me the government (like, country-wise) was currently financing workshops all over the place to create demand for exactly the service I had thought of providing. That sketched out an eye-boggling total addressable market. My jaw dropped (surprise), then shut back up (thinking), and now I definitely plan to go to these mentoring sessions and explore that topic further.

The other fun bit that happened in July is that I met with a few people who have various types of ambition, and that included some folk with less experience than me but way more ambition. Two things happened: one, I felt stimulated (as in, challenged) by those conversations, and this felt good. The other thing that happened is that they actually bought my time (as in, with money), unprompted, to consult for them about their career growth. That also felt good. Unexpected, not fully surprising, but still good.

❦❦❦

Then there is a steady stream of high-caliber individuals, with plenty of ambition and remarkably good ideas, who are just starting their own project and are starved for quality time with like-minded folk. I was happy to oblige—and I took the opportunity to learn quite a few things in areas adjacent to my current interests. There is a little bit of exciting ambiguity in these conversations: everyone is pitching to everyone, naturally, as in, “please sell your labor to me and not someone else, and let’s fight pleasantly to determine the precise boundary of our labor/money/time relationship.” However, at the same time, the subtext is really “all these things are so hard to do alone, maybe we could take a break from the fighting ring and do some bonding as equals.” These are interesting relationships to explore.

A few candidate projects caught my attention in these conversations. I’m still not too sure where I want to go with them. (More on that below.)

❦❦❦

Unrelatedly, yet another rather fun thing happened. This one is very recent (yesterday) and it made me laugh so hard, so I can’t resist sharing today.

How it started: I thought of a thesis that editors are like gatekeepers for access to written stuff, and book/story publishing channels without editors were much more vulnerable to an infestation of poorly AI-generated text. I decided to test the thesis with a channel I am somewhat knowledgeable about: certain subreddits where users exchange short, very smutty stories for mutual—ahem—entertainment. Could I measure a significant uptick of lower-quality writings in the last few months/year?

It turned out that no, I could not, and this surprised me. It took me very little time (as in, just one attempt on ChatGPT) to realize that all of the current hosted inference engines are heavily censored, to prevent them from generating the type of content that fits in that particular publishing channel. It turns out that their payment processors forbid them from supporting certain activities, and so they censor any output adjacent to these activities.

What gives?

There were two things to learn here. (Besides the fact that payment processors play the despicable role of international morality police these days, but that’s hardly new.)

One is that somewhat surprisingly, the cultural space where one would expect the most corruption of creative production is where it is currently the most authentic. Everything, so far, seems to be written by humans for humans. I assume this is because the creators never get paid and have no expectation to be. (NB: Written short stories / fanfics. Video does get paid—OnlyFans, etc.—and some traditionally published books too.)

The other was an immediate, surprising and supreme desire on my side to try and see if I could be the first person who could successfully muddy these pristine waters with my AI-supported smut. At the very least, it did help me push my disdain for the tech aside for thirty minutes and learn the bare minimum to run an experiment.

This is how I (summarily) learned how to use Llama.cpp, Hugging Face, prompt engineering and a few other things to get LLM inference run on my own computer and produce uncensored smut.

This is the part that was entertainingly hilarious. Due to a combination of low-quality models (I can only load so much on my limited computer), and the sea of absolute garbage that the internet provided to these things to be trained on, the inference served me with the most absurd and surrealistic combinations of scenes and (imagined) visuals depicting two people wanting to do things with each other and then doing them together.

Humor, it is said, exists when the brain expects something and something different and unexpected happens instead. There are many ways that two people can go from barely knowing each other to knowing each other very well, and the LLM showed me quite surprising itineraries indeed. All amusing, some worrying, few plausible, and all very creative.

Do not read me wrong: this interaction was serious editing work. None of the responses were suitable for sharing with other people (at least, methinks, without making them uncomfortable). None of this can be (currently) fully automated and I had full creative influence over the result. To produce just one short story—500 words—I refined my prompt dozens of times, re-generated multiple times on each prompt, and eventually selected segments of multiple inferences to glue together manually before I felt had something decent to post. In all the time it took me, I could have written the story myself.

Except… I could not have done it exactly. You see, I wanted to write something smutty about Mormon Elders on a mission, and I know very little about the Mormon church or Mormon missions, so I lacked all the background knowledge to make a good story. But I knew the training set, and I knew all the knowledge was there. I effectively used the inference engine to tease out latent knowledge that I did not possess myself, like one in old times would pull the fibers from the wool to make a string. And of course it was way cheaper than paying a US author with prior experience with the LDS church to write smut (if such an author even exists!).

The bottom line was a lukewarm success: after 24 hours, the short story is still trending on its forum, multiple readers are asking me for “a followup to my story with my fellow Elder”, one was trying to console me that “my crisis of faith was worth the discovery of an authentic relationship” and it also got a lot of internet virtual cookie points.

I might run the numbers later and see if there’s a cost effective way to generate and sell e-books to this avid (and never satisfied) audience. At the very least, OpenAI and the like self-selected out of that market (for now). This feels like an opportunity. It also seems as if I have accidentally acquired adequate editing skills through many years of—ahem—side projects, and perhaps I could leverage that.

❦❦❦

Meanwhile, July also gave me ample time to think and study.

For one, I continue to want to share my story about how I rebuilt my house, solved a couple of critical architectural problems that probably affect many other folk in my country, and offer my services to help others solve the same problem. I still hope I can muster some free time to do something about that.

Then because I could not read any more, I transitioned to watching (and listening) to a bunch of people talking about their life path and experience. Of course, the large majority of content in that category is garbage: rich folk selling trainings to wannabee rich folk about avenues that are now closed. But there is the occasional gem.

In particular, I enjoyed:

Of course, I also did enjoy a few reads, despite my malfunctioning elbow. In particular:

Panic! at the Job Market - Matt Stancliff
Where Matt analyses the current dynamics of labor/money trades in the tech industry, how they are different from a few years ago and what we should expect next.
Disrupting management - A fix for Tech Culture? - Chelsea Troy
Where Chelsea points out that traditional tech management is ethically corrupt at best, and actively toxic on average, and proposes a different approach. I like her ideas.
Why don’t developers water the plants - John Yorke
John reflects on the pitfall of organizational design: without intentional structure, tasks that are important or necessary but do not map to the core purpose of the individuals already in the team, get ignored.
A discussion of discussions on AI bias - Dan Luu
Where Dan reflects on the fact that LLM inference is still multiplying the most problematic discriminating biases about human populations and there is still no hope on the horizon.
The six dumbest ideas in computer security - Marcus Ranum
This was written in 2005 and is still relevant. Ironically, I read this just a few days before Crowdstrike burned the world down. Some things people never learn.

❦❦❦

So here we are. I find myself with a number of serious projects in front of me:

  • continuing the project from April.
  • networking like “Vibend but better”.
  • something something incubator between now and one year from now.
  • straight-up non-tech career consulting billed hourly.
  • something about advising and directing home remodeling projects.
  • something about investing into (read: acquiring) a super-boring business (e.g. car wash) and getting “real world” experience from that experiment.
  • publishing AI-generated smut for money.
  • one really cool urbanism project that someone whom I met recently successfully pitched to me.
  • another somewhat cool tech infra project that someone else whom I met recently also successfully pitched to me.
  • probably some other cool projects that numerous people I already plan to meet will likely successfully pitch to me.

What happens next?

❦❦❦

One thing that needs to happen real fast is a funnel factory. I need to be able to try out things using my own pipeline for inbound interest on these various projects. I want to be able to build multiple web sites (one per topic), each with a contact form, some pricing gadget, some marketing campaigns, and a way to collect contact details, then collect customer cases, then convert whatever needs to be converted, with metrics etc.

But making just one of these things is already so much work! I really can’t stand the thought of doing all this manually.

Just last week, I was asking Claude, ChatGPT and Copilot to help me with integrating Stripe and a bog-standard 3-column pricing model into a Flask application, and I was simultaneously enthusiastic about how little boilerplate I needed to write myself (these tools are truly helpful for that) and also absolutely irritated by how much grind there was remaining to tie all the loose ends together—mail integration, metrics, visual design, copy editing, etc.

Can we do better?

Well, it turns out that…

❦❦❦

The “algorithms” on Youtube and Instagram really know me well. They led me to Ship Fast, a tool from a good guy over somewhere in Europe, which gives you a template to do all of the above multiple times, but for a fixed price.

Gold! Gold! So many people are producing virtual gold all the time and it is all so inspiring. Shoutout to Marc Lou.

❦❦❦

References:

Publication Date: