Entry 170. Seven years complete. The streak’s uptime survives its 170th consecutive fortnight, and this entry closes the senior era’s fourth year with news held since Thanksgiving: the platform team launches January 6th, and the org has attached a title to its lead. Staff engineer. Me. The packet (this blog, jokes removed, again, that trick still undefeated) cleared committee last week. More from the far side.
2019’s thesis: the repricing. The public market repriced the cheap-money decade’s mythology (Uber, WeWork, from $47B to founder-defenestration in six weeks). The 737 MAX repriced certification-by-inheritance in the dearest currency there is (the fleet remains grounded, the reports still landing). China repriced the cost of global platform values. Quantum repriced “impossible” with an asterisk and a rebuttal. Even endings got repriced (ask HBO what a rushed one costs). The decade closes with every kind of confidence, financial, institutional, computational, marked to market.
The quiet item, per tradition (last year’s Transformer, the year before’s BERT, the thread the future will grade this archive by): OpenAI spent 2019 releasing GPT-2 in stages, initially declaring the full model too risky to ship, then walking it out quarter by quarter as the sky declined to fall. The capability curve (write-a-paragraph to write-a-page, coherently, from a prompt) is compounding on schedule; the release-strategy debate (staged rollouts… for language models, the Cloudflare lesson arriving in AI form pre-emptively) is the part I suspect matters most. The 2010s’ quiet thread ends the decade neither quiet nor loud: humming. Extrapolation remains an exercise for future entries, who I trust are living in it.
My decade, since the format demands it: 2013’s terrified junior (git reflog, entry one) closes 2019 as a staff engineer with a Proverbs file at 201 entries, two Fantasy PL titles (Null Pointer Exception dynasty status: disputed, banner count: legally binding), one marriage of sport and craft so complete the group chats merged, and 170 fortnights of evidence that the compound interest of small consistent things is the only get-rich scheme that works. Next sprint (the 2020s): lead the platform team like it’s a product. Write more, delete more. Watch the humming thread. The graphs are flat, the on-call is quiet, and the twenties are going to be boring and normal, I can feel it. See you on the other side of the decade.