Don't make me wait

It’s 2025, after signing up for an app or website with your e-mail address, you wait to receive the “verify your account” message in your e-mail inbox. You keep refreshing your inbox, the message arrives about two minutes later. Why is this not instant?

An easy case to make for why it should be instant comes from the world of e-commerce. We know that even milliseconds of delay in your software lower customer retention and satisfaction, thus lowering company revenue. What makes the waiting-for-confirmation-mail problem even weirder is the fact that software companies invest heavily in generating new user sign-ups.

Okay, I want to make a second point, not from the world of e-commerce, but from the world of ancient, old school commerce: banking. Banks understand that the most valuable customer behavior (opening a new account) deserves massive optimization, even when it seems inefficient:

Banks have extremely weird behaviors by the standards of parking engineers; the typical user behavior is to stop in for only a few minutes but the behavior the bank wants to optimize for, new account opening, can take half an hour to several hours. Through what turns out to be a simple result of queuing theory, bank branches end up with a lot of parking that appears mostly underutilized almost all of the time, and this is close to optimal.

Excerpt from the excellent patio11 article.

Okay, I hope to have made my case of why it should be instant. Now the question remains why clearly companies don’t care about making it instant. I remember back in 2005, 20 years ago, being annoyed by this. Maybe it is like daylight saving time? Everyone hates the thing, and clearly we should do better here, but since it’s a once and done thing people don’t complain long enough for the problem to be fixed? Maybe we rely on the “stay signed in” button too much? I don’t know. Anyway, to all people designing an e-mail confirmation service, please speed it up. Don’t make me wait.

Killing your darlings

Our second biggest cost is taxes, and our biggest cost is opportunity cost.

~ Attributed to Larry Page

Larry Page and Noam Shazeer both obsess over opportunity cost. It’s the cost that matters most when evaluating policies, personal or public ones.

Here’s the problem: opportunity cost is invisible. The counterfactual, by definition, doesn’t exist, so you never see it. Kahneman calls this “what you see is all there is”.

A solution to this ocular blind spot in your personal life is deciding against doing the thing you really want to do. You kill your darling. You will be super conscious of what you left on the table, the counterfactual becomes clear, and you actually take all the other options more seriously.

Starting Sabbath

Usually after the 6th day of consecutive work, I feel drained. I hate the feeling of being drained and what it does to me: stay in bed longer, watching YouTube videos. It just becomes easier to tell myself that I deserve this cheap form of rest. Ironically, being drained has a negative effect on my sleep time and quality, as I fall into bad, lazy patterns.

So instead of staying in bed and complaining, let’s do something against the draining downward spiral: I’ll take a page out of the Jewish book and adopt ancient deliberate decompression: Sabbath. Keep it simple: starting with the Friday communal prayer at noon, I’ll intentionally only do things that give me energy:

• Meditating
• Taking walks
• Drawing
• Reading up on new tech
• Programming hobby projects
• Household chores
• Meeting friends
• Cooking
• Stretching/yoga

And for everything else: just don’t do it.

I worked for more than 30 years on these systems and they could never work. I implemented a very fast NLP symbolic parser, for which the team I worked with created grammars for 8 languages, including Japanese. In 2007, with a grammar of 60,000 rules, we could parse at a speed of 3000 words/s (see https://github.com/clauderouxster/XIP for the Open Source version). The parser could extract syntactic dependencies, and could use ontologies. But language is like sand, the more you try to grab, the more you leak. There was a kind of futility in trying to compress languages into rules, nothing actually scaled up. Still, we managed to win competitions as late as 2016 with SemEval sentiment analysis, and in 2017, we also ranked first in a legal document extraction campaign organized by IBM, but to no avail. It was a lot of work, and the conclusion was very simple. We had to push our grammars as far as possible into lexical grammars, which eventually LMM managed to really implement. We discovered very early, that context was all that mattered. We tried to create grammars that would apply to a full paragraph instead of sentences, but then the performance would plummet. The reason why LLM work, is that at each step they compress the whole context into a meaningful vector, which they then used to guide the rest of the generation process. I spent my whole life in the pursuit of a perfect parser with very brillant people, and I really find hard to say that not only did we fail, but that LLM is the response we were looking for.

~ Claudius

Getting used to it

Is underrated.

Don’t like it, but it’s useful enough to do it? Just get used to it.

More on this: You Can Try to Like Stuff by Dynomight

Never record a meeting

Always transcribe it by hand. People act differently when they know they are being recorded. For me, this is the main reason I am against recording. I want to hear people’s genuine and slightly unfiltered takes.

Producing an artifact is one way to make meetings more effective, though. Therefore, just transcribe the meeting by typing it all out on the meeting screen for everyone to see. That’s the sweet spot.

Pre-LLM list of arguments:

• People know exactly what you are writing, thus are more trusting
• People can opt-out of being transcribed
• It keeps you engaged in the meeting as you really have to listen
• You can slow/speed up the convo with an excuse that you need some typing, you can ask clarifying questions -- you are much in control of the meeting

Post-LLM list of arguments:

• Hand transcriptions are better than machine generated ones when the meeting has many people, many topics, a mix of (non-English) languages and subpar audio quality
• It keeps your typing skills sharp, a crucial skill for working with the LLMs
• Humans are more likely to read writing that’s not produced by LLMs, rational or not