August Heatwave
It's been five months since I last wrote. Five months. In my head this seems significant, like that much silence would cultivate into something. Some people (likely some of you) can write every day, capturing even the smallest moments into something worth reading. Me? Just striving for basic literacy and legibility paired with a deep gratitude for those of you who choose to come along for the ride…
And with that the news:
I'm working way too much, trying to massage machines into doing more work for people. I love the project but not the hours. Officially I'm on the "Agentic Experiences" team at Meta which involves trying to apply large language models to doing actual work for people. After 6 months of this I'm starting to feel like I'm getting the hang of it. It's been a season of nights and weekends which bums me out because I'm missing the best parts of the beautiful summer weather.
Shelby is holding down the fort on the home front. Still juggling building her business with raising Tommy most of the hours and covering drop offs and pick ups for Gaël. We're tying the knot in September and she’s taken the lead on planning that.
Gaël is doing the summer camp thing and generally just enjoying the calm before the storm of school. He's off to a new one once again in September, this time it will be much more challenging so we're figuring out what and how to support him.
Tommy has at times full use of any two limbs and the other two are always along for the ride. Everything gets immediately grabbed and put in the mouth. We are saying a few words, but mostly just making fun new and unexpected sounds every day.
It's hot enough out here to enjoy swimming every day, so that's what I'm going to try to do,
Cowabunga dude,
DJ
PS:
One of the rules on this newsletter is I try to de-emphasize work because I think the important things in life happen outside of those walls of the office but since I’ve been working so many hours, I wanted to share more.
What I actually do: I started at Meta last November to build web tools to manage privacy compliance. If for some reason Meta builds a product that doesn't comply with some local regulation in any of the countries that it operates (which is most of them) the government can fine them. And while you might think of Meta as willing to skirt regulation this ends up being an actual low hanging source of income for those governments and potentially a large fine for the company. The incentive alignment is to comply and comply early and well.
This leads to a very large and complex effort into enforcing privacy compliance within the company. That’s where I started.
In February I was moved onto an "applied AI" team, which is focused specifically at looking at our heaviest manual processes and automating them. On a more concrete level, I don't build the AI models everyone is talking about, but I figure out how to use them, normally in a programatic way. When people say AI will replace workers, it's my job to figure out how to make that happen in real terms.
This usually means taking these non-deterministic statistical models and creating the right scaffolding around them to turn abstract problems into engineering ones. To even get started here, you must get a sense of the problems and capabilities of the models. This means the last six months have been a mix of experimentation, but also reading and testing, anything and or anyone that has something works must be immediately incorporated into our work. The whole process feels a bit like a Russian nesting doll. The most inner doll being this understanding/learning process of the models, what they have been trained for, then move outwards creating a programming framework till you get to the most outer layer which is the actual problems you are trying to solve. If you are trying to solve a A->Z problem (let's say the entire privacy compliance problem for some large system) you find you might be able to reliably do A->C, E->J, N->Q, (classifying assets, reviews and otherwise) and then just repeat this process over and over again for whatever the problem might be.
This has thoroughly scratched an itch for me and it provides me the opportunity to work on the abstract elements of problem solving. While it still has a lot of the traditional elements of software engineering, the whole non-deterministic element as well as the playing with words provides this abstraction that I think makes it slightly more…artistic?
On this topic I could likely go on and on, at work I do exactly that. On how I believe this automation could be a good thing, that normally the introduction of efficiency into a system leads to economic growth which normally leads to rise in the standard of living. Or how these tools can either lead to laziness or if used effectively have an amplifying effect both on your ability to think as well as output. All of these topics have huge risk/reward ratios and could and have been unpacked far better than I could or really should in a personal newsletter. This more or less distills down to I am working more than I'd like, but at least I'm having fun doing it.