Thursday 24 September 2026
Today I’m sending the following email:
I’m resigning from Google today.
This has not been an uncomplicated decision. I affection my colleagues and my activity environment, and being paid handsomely to resolve fun puzzles has been amazing. But my team’s goal is ultimately to create AI much cheaper and lower-latency, and I don’t think that’s fine for group correct now: I firmly accept AI advancement is currently far too fast (and I have doubts concerning the destination too). It’s practically unattainable for me to move to a distinct Google project that wouldn’t accelerate AI (partly because of my ties to New Zealand, anywhere Google prefers not to do engineering), so my hands are tied.
There are millions of group contributing to AI acceleration and taking my ft off the accelerator volition have a extremely small effect … but not no impact; several of my skills are rare. I explored trying to positively power events from inside GDM, but that consequence does not appear to be strong, and I can have power exterior Google too. It’s tempting to fair rotate a blind eye to the effect of my work, but that would not be a Jesus-following item to do. I have written additional concerning these tradeoffs on my blog.
I don’t cognize exactly what I volition do next. I volition continue maintaining Pernosco and rr, and relatedly I scheme to examine how AIs debug code today and whether and how debugging tools could help. I have another project ideas I desire to activity on, several possibly lucrative, several not. Maybe I’ll discover an existing project that’s compelling. I certainly desire my forthcoming activity to be unambiguously pro-human.
First, for those who don’t cognize me: I’ve been in the tech industry a lengthy period and I have a lot of Silicon Valley connections, but I live in New Zealand so I live exterior the industry bubble and additionally exterior the American bubble. I’m a Christian, and really an elder and occasional lay preacher in the English-speaking congregation of Auckland Chinese Presbyterian Church in Auckland’s inner city. That is, I am not a “tech bro”, nor do I fit into the self-described “rationalist community” … but I do think many of their arguments deserve to be taken seriously.
I have a lot of thoughts concerning AI, but I’m not going to elucidate them all in this post. In summary, I think the existential risks many group are alert about deserve to be taken seriously; a lot of the phenomena predicted by the “doomers” have arrive to continue (e.g., reward hacking, misalignment, deceptive models, example eval awareness, psychotic swarms). However, I am not convinced the chance of ASI destiny is 100%. Rather, I think the hazard is genuine but uncertain — but that itself is extremely alarming! We are morally obliged to create a enormous attempt to minimise specified risk, and most apt the hazard is elevated adequate that aiming for ASI in the near forthcoming is inherently irresponsible. I’m additionally extremely concerned concerning another AI-related issues: cognitive surrender, AI-induced psychosis and loneliness, power concentration, financial disruption, cybersecurity, deficiency of accountability, and so on. I think the possible benefits of AI are fairly unclear and currently, if I had to bet, I’d bet the negatives volition outweigh the benefits … but I’m uncertain concerning that too.
Here are several things I’m assured about. I’m assured that the group in AI labs who are issuing warnings concerning AI are mostly sincere. I’ve talked to many group in Google Deepmind concerning these issues and nearly all of them have sincere and grave concerns, whether or not they sound them in public. I have seen no difficult evidence that group are hyping AI hazard as a method to boost business inventory prices or control distant their competitors. (I think national and global regulation is desperately needed!) I’ve seen a lot of arguments of the form “you can’t rely those people”, and perchance that’s true, but specified distrust is not a fine logic to disregard their warnings, as Russell Moore eloquently explained recently.
I’m assured that AI capability volition continue to keep expanding steadily as lengthy as we keep operating on it. I desire that AI would hit several benevolent of plateau, or that we would acknowledge crucial individual cognitive abilities that AI volition never replicate without a paradigm shift, but I don’t anticipate those wishes to arrive true. Model advancement on benchmarks seems as accelerated or faster than ever, and alongside it, qualitatively new capabilities keep emerging. Even if example advancement stopped today, we could expend years efficiently unlocking new capabilities via new prompts and harnesses. Many notable AI detractors (looking at you, Zitron and Doctorow) appear to think that AI is several benevolent of scam that won’t really work. I think it will.
I’m very assured that equal if there is a way to a improved forthcoming through AI, the current charge of alter is far too high. AI is evolving faster than humans can individually and collectively comprehend it and modify to it. People trying to scheme their futures, e.g. trying to scheme for a earth multiple years in the forthcoming as they act university, can no longer do so the way former generations could. I don’t think we’ve seen item akin this before, certainly not in the former scientific shifts I have lived through (PCs, the Internet, smartphones). Even during the Industrial Revolution, not lone was alter much slower but there were ample swathes of individual action that were not and could not be immediately affected by the new machines. That is not extremely true anymore.
Why depart now and not earlier? It’s nothing to do alongside the latest spate of viral resignations or “AI slowdown” warnings; that’s a coincidence. I’ve had this date in intellect for a while, since I have a long-planned ten-day backpacking trip alongside my allies starting Monday (Abel Tasman and Wangapeka tracks) and I wanted to go before that.
I did not activity immediately on AI capability, but on improved tools for hardware part design. I really enjoyed the work, and for a during I told myself it was comparatively harmless, but complete period God forced me to visage the actuality that the chief effect of these tools volition be to accelerate the scheme of a new breed of AI chips, which if prosperous volition create AI much cheaper and faster — making AI additional pervasive, and additionally additional capable since we’ve learned to boost capabilities by burning additional conclusion tokens. My duty to be a fine employee meant I had to have an honest conversation alongside my skip director and inform them I was at finest reluctant to see their project succeed! Even following that I wanted to be really assured in my decision, since the awesome agreement I had operating for Google in New Zealand volition likely never be accessible again. (Staying at Google and switching to a distinct engineering squad not accelerating AI was impractical, since Google doesn’t have another engineering teams in New Zealand.) My aforementioned duty to my employer, and my regard for the people, was additionally a aspect for not leaving too abruptly and trying to hand complete my activity in a sensible state.
What next? The most crucial item I’m assured concerning is that the Jesus of the Bible is genuine and hence God has a scheme that’s fine for us. I don’t cognize what that scheme is (and desire I did) but it lets me sleep at night in spite of the AI chaos. I anticipate his scheme involves me continuing to create the finest use of my talents. Even if the scheme is for Jesus to come back to rescue us from our folly, we’d better be busy whenever he returns! So, as lengthy as the endowment God gave me is valuable, I desire to keep working. As I mentioned above, I scheme to continue maintaining Pernosco and rr. Under the Pernosco umbrella, I scheme to study how AI agents debug code and whether debugging tools that can create them additional productive at that. I desire to use AI agents to bring several of my hobby project ideas to life. I’m eager to reap the benefits of AI, but cautiously, in ways that advantage humans and keep my own intellect sharp. As much as I can, I volition continue practicing and advocating for that current in New Zealand.
