Joining Elicit

It's not really a life change, but it's not *not* a life change, either 🌅

Light Coming on the Plains No. I by Georgia O'Keeffe (1917)

Sometimes people say that most jobs never get advertised. Either the positions get filled internally, or they’re given to someone specific before they get a chance to become public, like a reputable specialist that the company headhunted, or someone who was referred by a current employee. Or, better yet, the position is specifically crafted for someone and would never have existed without that person’s idiosyncratic set of skills.

I don’t know what fraction of all jobs is like this. Of the three real grown-up jobs I’ve had, the first two followed the classic model of see job posting → send résumé → get interviewed → get hired. The third, however, did not. It began this week, sort of, and it happened thanks to a mix of all the things in the previous paragraph. After spending nearly two years working for Elicit (and its predecessor organization, Ought) as an external contractor — a gig that itself came into being through a fairly unlikely coincidence of events — I am now joining it as a full-time permanent employee, in a position that evolved over my time there and is primarily going to involve evaluating AI models, but also occasional software engineering and writing. I’m pretty excited about this.

In fact, I’m excited enough that, for the first time, career stuff is the most exciting thing in my life — more than writing this blog! (Which raises questions about what to do with it, but that’s the topic for next week. Spoiler alert, this is my 199th consecutive weekly post.) In practice, my day-to-day work isn’t going to change that much, at least not suddenly, and yet it feels like a pretty momentous transition, philosophically.

At a time like this, it’s in good taste to take a moment to reflect on life and my career, and pay attention to the details of the path being taken.

Light Coming on the Plains No. II by Georgia O'Keeffe (1917)

I’ve been interested in Ought (a name that literally nobody in my family and friends understood when I told them about it) since I learned about it a few years ago. It has existed for a while as a research nonprofit, trying to figure out ways to help with AI alignment. It had strong cultural links to rationalism and effective altruism, although not really direct ones anymore, I feel. (In this it mirrors my own path quite well.) Eventually it stumbled upon the idea of scaling good reasoning. Whatever the future holds, it’s a good bet that we’ll do better as a civilization if we have more “reasoning that reliably arrives at true beliefs and good decisions,” which as of today is unfortunately still quite rare.

And so Ought began working on a tool called Elicit (a name that my family and friends understand somewhat more, but usually mix up with “illicit,” which, well, tends to raise eyebrows). Elicit uses the latest advances in AI, especially large language models, to facilitate the analysis of whatever knowledge hides in scientific papers. Today Elicit is its own company, spun off from Ought, which technically still exists but doesn’t do anything. “Scaling good reasoning” became its official mission, and it started to fulfill it by working on the task of automating systematic reviews. As in, you want a systematic summary of the state of knowledge on a topic? Either you hire researchers and give them 6 months, or you boot up Elicit and you do it in a day. We’re not quite there yet, but we’re working on it. Eventually, if we’re successful — and the signs are encouraging — we’ll branch out to other areas where good reasoning is needed, and there will be true beliefs and good decisions galore.

Scientific papers, and how they’re a terrible way to hold and convey knowledge, have been a special interest of mine for a while. So I feel pretty lucky to be able to contribute to the task of making it less terrible to interact with them. In fact, it’s remarkable just how many of my interests are aligned with working at Elicit — science, AI, writing, coding. I don’t think I’ve ever been this motivated to work somewhere. It’s been nearly two years, like I said, and I feel no compulsion whatsoever to do anything else. (And if it so happened that I had to do something else, working in AI evaluations seems like a pretty promising field to be in.)

This high motivation is a notable change for me! I’ve never had a regular job for more than two years. When I look back at my adult life, it involves a lot of studying, and a fair amount of freelancing; I’ve resisted the classic model of being an employee quite a bit. I have been worried in the past that I get bored of things too easily, and would never spend the years required to become a true exert in anything. But of course that was false: I just hadn’t stumbled yet upon something I truly wanted to do for a while. Tentatively, I think it might be the case with Elicit, and that’s really exciting.

Plus my colleagues are all sweet, I get to work from home, and I’ll travel to California every couple of months. If this job hadn’t existed, we would have had to invent it. Which I guess is exactly what happened.

Light Coming on the Plains No. III by Georgia O'Keeffe (1917)