Remark in guaranteed safe AI newsletter: 

Niplav writes

So, a proposal: Whenever someone claims that LLMs will d/acc us out of AI takeover by fixing our infrastructure, they will also have to specify who will pay the costs of setting up this project and running it.

I’m almost centrally the guy claiming LLMs will d/acc us out of AI takeover by fixing infrastructure, technically I’m usually hedging more than that but it’s accurate in spirit.

If transformative AI is developed soon, most open source projects (especially old ones relevant to internet infrastructure) are going to be maintained by humans with human response times. That will significantly increase the time for relevant security patches to be reviewed and merged into existing codebases, especially if at the time attackers will submit AI-generated or co-developed subtle exploits using AI systems six to nine months behind the leading capabilities, keeping maintainers especially vigilant.

I usually say we prove the patches correct! But Niplav is correct: it’s a hard social problem, many critical systems maintainers are particularly slop-phobic and won’t want synthetic code checked in. That’s why I try to emphasize that the two trust points are the spec and the checker, and the rest is relinquished to a shoggoth. That’s the vision anyway– we solve this social problem by involving the slop-phobic maintainers in writing the spec and conveying to them how trustworthy the deductive process is.

Niplav’s squiggle model: Median $~1b worth of tokens, plus all the “setting up the project, paying human supervisors and reviewers, costs for testing infrastructure & compute, finding complicated vulnerabilities that arise from the interaction of different programs…” etc costs. I think a lot’s in our action space to reduce those latter costs, but the token cost imposes a firm lower bound.

But this is an EA Forum post, meaning the project is being evaluated as an EA cause area: is it cost effective? To be cost effective, the savings from alleviating some disvalue have to be worth the money you’ll spend. As a programming best practices chauvinist, one of my pastimes is picking on CrowdStrike, so let’s not pass up the opportunity. The 2024 outage is estimated to have cost about $5b across the top 500 companies excluding microsoft. A public goods project may not have been able to avert CrowdStrike, but it’s instructive for getting a flavor of the damage, and this number suggests it could be easily worth spending around Niplav’s estimate. On cost effectiveness though, even I (who works on this “LLMs driving Hot FV Summer” thing full time) am skeptical, only because open source software is pretty hardened already. Curl/libcurl saw 23 CVEs in 2023 and 18 in 2024, which it’d be nice to prevent but really isn’t a catastrophic amount. Other projects are similar. I think a lot about the Tony Hoare quote “It has turned out that the world just does not suffer significantly from the kind of problem that our research was originally intended to solve.” Not every bug is even an exploit.

In AI safety adoption is also a reason. I expect labs to adopt stuff that they pay for, whereas research / open source public goods they'd be like "oh that's cool" then never use it. 

Yall, I have been off and on distracted from my work by intense and unpleasant outrage/disgust at immigration enforcement, ever since Trump's first campaign speech close to ten years ago. I have few visceral moral convictions, and this is the strongest. I wish my strongest conviction was a positive one, where I'm brimming with hope and warmth. But instead, anger is more salient. 

I don't know what to do about it. I dont think spending kajillions of dollars of lawyers so that the victims' lives can be somewhat less inconvenienced passes the ITN test, and i don't have kajillions of dollars. So it's basically like, I know I'm not gonna do anything about it, I just have to sit and try to let it distract me as little as possible. Total bummer, feels very disempowering. 

It'd be great to be able to transmute these feelings into a positive vibe. 

I was just sent this https://www.mzbworks.com/prayer.htm --- really fantastic. TLDR luck/magic is real but only works on one thing. I normally think of luck like the compass in Pirates of the Caribbean (2003) (that points to what you want most), although unlike this essay, I normally think of it where the user can juggle multiple goals and the compass will adjust to that. Here, with the author's notion of prayer, we can really only activate the power of luck on one thing. Perhaps "at a time", perhaps not. 

But this fellow was not an average young man; a little more was expected of him. He was not satisfied with Jesus' answer, and said, "Master, all this have I done from my childhood." In other words; "I've already done that, I think I'm ready for the next step on the way."

So this remarkable Rabbi leveled with him, and gave him the next step; "Then go, sell what thou hast and give to the poor, and follow me."

You can almost see the young man's mind working. "What? Me, who has kept the Commandments all my life? Must I do this too? There's nothing in the Law and the Prophets about this! People will call me crazy! Why can't I keep what I've got and be saved right here?"

So the Bible tells us the young man "went away sorrowing; for he had great possessions."

It doesn't say anywhere that this young man was damned, or that Jesus reproached him, or cursed him. He probably went on keeping the Commandments, and may have lived an excellent life, a happy life, as a pillar of his community.

But he had wanted something more; and he never got that. He never became one of the inner circle.

Just found it charming that jesus was like "oh you should've mentioned you beat easy mode already, i usually don't get around to telling people there are different difficulty levels". But the application to charitable living/giving is obvious. I literally recently said to myself "eh give yourself the beef cheat this month (i'm mostly vegetarian just open to cheating once or twice a month), you donated a kidney" which is of dubious validity, and yet, just might work (in some sense). 

but there's a sting in the tail. I knew a man who decided to concentrate entirely on money-making for a few years so that he would have money to study music and compose. He was an overwhelmingly talented man in half a dozen fields, and he became, in a scant six years, almost a millionaire. But that six years had wrought a tragic change in him, for by the time he had the money he had once vowed to devote to his composing, he had spent five years without once touching his piano, his cello, his flute and bassoon, his books.

He said, when asked about his decision, "Oh, music. That was just a juvenile notion. Now I've got a business to look after." The very act of making money had altered the man himself beyond recognition. Yet he had sincerely loved music, and the world of music is much the poorer by the loss of strange and lovely compositions, now never to be completed or published -- so that use of mental power backfired 

setting aside the fact that I almost literally did this personally, I thought this would resonate with some of yall who've thought about value drift. 

I think "outdated term" is a power move, trying to say you're a "geek" to separate yourself from the "mops" and "sociopaths". She could genuinely think, or be surrounded by people who think, 2nd wave or 3rd wave EA (i.e. us here on the forum in 2025) are lame, and that the real EA was some older thing that had died. 

I roughly feel more comfortable passing the responsibility onto wiser successors. I still like the "positive vs negative longtermism" framework, I think positive longtermism (increasing the value of futures where we survive) risks value lock-in too much. Negative longtermism is a clear cut responsibility with no real downside unless you're presented with a really tortured example about spending currently existing lives to buy future lives or something. 

I wrote a quick take on lesswrong about evals. Funders seem enchanted with them, and I'm curious about why that is. 

https://www.lesswrong.com/posts/kq8CZzcPKQtCzbGxg/quinn-s-shortform?commentId=HzDD3Lvh6C9zdqpMh 

I love these kinds of questions! I attempted a roundup here but it never really caught on 

nitpick: you say open source which implies I can read it and rebuild it on my machine. I can't really "read" the weights in this way, I can run it on my machine but I can't compile it without a berjillion chips. "open weight" is the preferred nomenclature, it fits the situation better. 

(epistemic status: a pedantry battle, but this ship has sailed as I can see other commenters are saying open source rather than open weight).