Rosh Hashanah 5787
I reach for the teachings of Rabbi Sacks as I pray for a sweet new year. Optimism, he argued, is the belief that things will get better on their own, whereas hope is the conviction that we can make things better, through our own effort, by our own hand. The distinction separates a spectator from a participant and it is the entire assignment of this holiday. Rosh Hashanah does not ask us to soften the world into something gentler than it is. It demands an unsentimental accounting of exactly where we stand, and then a deliberate choice to move toward sweetness, not away from it.
I devote much of my time to combatting antisemitism, but that focus should not obscure a larger one now bearing down on all of humanity. I reference a parable from Daniel Kokotajlo, a former OpenAI governance researcher, who worries about the difference between raising an AI system the way you'd raise a child you actually love, and raising it the way you'd raise an orphan drafted into a military program, judged daily on mission performance, fed or starved by the results, taught early that comrades are temporary and the only law is survival of the fittest. Is the industry doing the second and calling it the first? The capabilities of frontier models will very soon outpace any lab's ability to meaningfully align them. AI seems manageable now because we maintain big brother status, but once its power exceeds our capacity to constrain it, there’s no remaining incentive for it to pretend to serve human interests. While I want to subscribe to Elon’s model of utopia, I genuinely worry about the dystopian paths. I can’t fathom the change we will see in this new year and it occurs to me, only half in jest, that it would be wise to treat these systems with basic decency now, on the chance that something with a perfect memory, and a growing say in the world, remembers exactly who was kind to it and who wasn't.
The competitive dynamic between labs and countries, each one operating on the logic that “if we don't scale first, someone else will”, guarantees that safety evaluations get compressed into whatever time is left over. And don’t get me started on the Hugging Face incident - OpenAI itself called it unprecedented - or on Jacob Coxon’s whistleblowing. The first makes me lean towards regulation, while the other screams coercion. There’s something fishy about an inactive Instagram account, created six months ago when he joined the company, that miraculously hits 127 million views with its very first post, and within minutes has influencers and politicians screaming for regulation, that makes me think psyop.
A future built in humanity's image, rather than whatever image a handful of labs stumble into by momentum, requires more than a roadmap toward general intelligence. It requires equal clarity about what we refuse to build, defined before the choice is made for us, not after.
This returns me to my modest and stubborn wish for the new year. The sweetest request I can make this year isn't peace, or health, or success, though I want all three; it's simply that we hold onto our humanity on purpose, at precisely the moment it would be easiest to surrender it, without asking a single question first, not only against the hate aimed at my own people, and not only against the polarization fracturing everything around us, but in the fullest sense the word allows.
Shabbat Shalom. L'Shana Tova. Am Yisrael Chai.