nbouscal

This is incorrect, their existing contract had these red lines and more until this January 9th memo: https://media.defense.gov/2026/Jan/12/2003855671/-1/-1/0/ART... which led to DoW trying to renegotiate under the new standard of “any lawful use”. Anthropic never tried to tighten standards beyond what had been in their original contract; DoW tried to loosen them.

There’s no such thing as general tech priorities, it’s just an amalgam of a bunch of individual priorities. In most of the cases I’ve seen those priorities are very reasonable. If you start a tech company and prioritize fiddling with internal messaging software over building your product, you’re going to fail. Prioritizing tech doesn’t mean yak shaving every random task, it means outsourcing as many non-essential tasks as you possibly can so that you have as much time and attention as possible for the one specific technical task that matters: building your product.

Maybe consider that we have literally hundreds of priority tasks, all of which are legitimately very important, and that while we can spend the time necessary to install and maintain non-turnkey solutions, we’d really prefer to work on one of the other priorities with that time.

I replaced Rails with PostgREST, which is written in Haskell but for me as a user is just a binary that I run in front of my database, so now I don’t have to write any backend code at all.

Good enough for you?

The following isn’t the top reason I recommend Postgres, but is the reason I think least likely to be echoed in a dozen other comments:

Postgres has some of the best documentation of any software product I’ve ever used. If someone wants to learn about SQL or databases, I always have to restrain myself from recommending that they just read the Postgres manual front to back. It’s comprehensive, it’s well written, it’s easy to navigate, and almost every explanation comes with several clear examples. It’s hard to overstate how valuable a property that is for a product as central to your architecture as your primary transactional database.

My background is more like yours. For me it’s likely a difference in size and type of companies worked for.

I was hoping that the article would explain what devops really means today and how I can jump on the devops wagon to hopefully make my job of doing all of the above easier.

1. Use AWS Fargate for all of your backend services. Keep the architecture simple enough that complicated service discovery issues etc don’t come up. If you need coordination between services, do it through Redis or similar.

2. Use RDS unless you really need to save money or use unavailable extensions.

3. Use terraform for initially provisioning the above

4. Set up CI/CD such that merges to master automatically update your services. (I like CircleCI’s aws-ecr and aws-ecs orbs for this.)

Pretty simple recipe, but it means no setting up servers, no hardening servers, no installing or optimizing software (other than by adding it to the Dockerfile), no installing or optimizing the database.

This recommendation reflects what modern devops means to me; opinions differ. To me it means:

- Infrastructure as code (terraform rather than clicking buttons in the AWS console then later forgetting which buttons you clicked)

- Immutable infrastructure (aka cattle not pets). Never SSHing into a server again.

- Automated testing and deployment cleanly integrated with existing dev workflow

Obviously there’s a scale at which you have to do something more complex, but I’d say that’s the scale at which you previously would have had an operations team.

Getting rid of the “writing server side code” and “writing front end code” parts is beyond the scope of devops, but you can skip a lot of the “writing server side code” part by using PostgREST. In exchange you may have to write an even-more-untold number of SQL queries, depending on your current practices.

Edit: Someone helpfully pointed out that I forgot to mention anything about logging or monitoring, which is a pretty glaring omission. On that front I strongly recommend Honeycomb. To set it up with Fargate you may need to run it in a sidecar container, but it’s fairly straightforward.

I assume eatonphil is correct, but to be honest I’ve never even tried, and in my view that’s actually part of the point: full commitment to immutable infrastructure, made really easy. If something needs to change, I tweak the Dockerfile or the task definition or a config file and redeploy. No more SSH.

I’m using Fargate for almost everything now, and am never going back. It’s extremely simple and easy, does exactly what it says it does, and saves me a ton of time and headaches. (I use Terraform to set up services, and CircleCI (with the ECS orb) to deploy updates.)

Linear Dynamical Systems by Stephen Boyd of Stanford: https://www.youtube.com/watch?v=bf1264iFr-w&list=PL06960BA52...

Programming Languages by Dan Grossman of University of Washington: https://www.coursera.org/learn/programming-languages

The second case is more risky

Not if they're otherwise identical companies, it isn't. There may be reasons to expect a correlation between risk and cash holdings, or something to that effect, but ultimately a share in a company is a share of ownership of that company's assets. If a company didn't become $10B more valuable when given $10B in cash, where exactly did that value disappear to? Value doesn't vanish into thin air simply because it's now owned by a company.

I'm arguing nothing of the sort. While it probably feels good to strawman everyone who disagrees with you, it certainly isn't helping you persuade anyone, nor is it allowing you to learn from other perspectives.

While we may theoretically have the power to change the way law enforcement works on these issues, doing so would be dramatically more difficult than making marginal improvements within the industry itself. We shouldn't let the perfect be the enemy of the good.

adventured is having a really hard time with your phrase "two identical companies". The answer to your question is yes, if you had two identical companies and then gave one of them $10B in cash, that company would be worth $10B more.

The digital economy is giving us a world in which the benefits are concentrated among consumers and the Big Five who serve them. Everyone else just lives in it.

This is a really weird phrase. "Concentrated among consumers"? More than 80% of Americans use the internet, not exactly "concentrated".

The law should be handling sexual harassment.

I think people on both sides of this issue would love it if the law showed itself capable of handling sexual harassment cases. Unfortunately we have way too much evidence that that just isn't the case. The obvious example that comes to mind is Brock Turner getting three months, and that's in the best case where someone is actually convicted. Far too often it never gets to that point, not because the harassment/assault didn't occur, but because the legal system is so utterly shit at handling it. The 'outragism' that so many people decry only exists in the first place because every system that is supposed to prevent or punish these acts has failed so completely.

'Not useless' is not a high enough bar in software. Flexibility always adds complexity, and complexity is a cost. Sometimes that cost is worth it, but the benefit needs to be much larger and clearer than 'not useless'.

1. Not even close to true, check Dan's other post here: https://danluu.com/startup-tradeoffs/

2. If you're that good at picking startups, become a VC.

3. Common shares are priced less than preferred because they're worth less than preferred.

4. Startups at every stage give liquidation preferences. Startups under pressure will give more significant preferences, sure, but they're relevant to basically every startup.

My company just hired a Haskell developer; it took less than a month and we had plenty of highly qualified applicants. It actually made us reconsider how many engineers we were planning to hire, because it was easier than we expected. By comparison, we also have Ruby code, and finding equally talented Ruby developers has been much more difficult.

You are right that most information is "non excludable" (if that actually is a word).

https://en.wikipedia.org/wiki/Excludability

Maybe do a quick search before implying someone is making up words.

I don't recall seeing 95% confidence intervals that narrow. Even the 90% confidence intervals I see are usually like 5-150 or something. (Also it would be super weird to me if the lower bound of someone's 95%CI was 50.)

Nobody thinks we won't notice when it's more imminent; the concern is that capability research has always been faster than safety research. If we wait until it's imminent, we risk having no chance of catching up in time. Why not devote a tiny fraction of our funding to starting the research now, so we can be ready on time? Worst case we're early... which in this context is a lot better than being late.

I distinctly remember a paper about AI figuring out how to either get wifi or send radio waves without access to the relevant hardware. Can't find the link at the moment though :/

Nobody is saying human-level AI is ten years away, people are saying there's some chance of it in ten years. Median 50% estimate is quite a bit longer. But a 10% chance just means divide the impact by ten, and with an event like this that still leaves you with a really damn big number.

Seems pretty clear neither of us will convince the other in this venue, so I'm going to leave it here. Thanks for the discussion :)

The word "just" feels awfully out of place in this context. If you were doing some kind of broad-based polling of public opinion, of course you'd want a bigger sample size, but 29 of the top 100 researchers in a field sounds like a hell of a good sample to me, and well worth listening to.

You can restrict your attention to the TOP100 group if you prefer.

Is this a badly conducted survey of non-experts? http://sophia.de/pdf/2014_PT-AI_polls.pdf

Edit: In particular, the TOP100 subgroup.

Sorry, gave a citation in another comment on the thread but not in this one. I was referencing http://sophia.de/pdf/2014_PT-AI_polls.pdf

We went from not even beating good amateurs at go to beating the best pros in the world. I don't think it's like a progress bar we should be expecting to fill at a linear pace, but, that certainly seems like a strong indicator we've come a long way (and it's one of many examples I could have chosen).

Have you read any of the estimates? Basically every expert gives a huge confidence interval, as do most of the people actively working on AI risk.

I've been personally told by multiple different people who scoff at AI risk that it will be totally fine, so it's not a straw man.

The median expert estimate for when we'll be 10% likely to have human-level AI is ~10 years.

AI risk research didn't receive a penny of funding until the last few years, and is still funded at way lower levels than a lot of things that have dramatically less impact.

In nearly every debate on the topic I've seen (with a few exceptions), the people concerned about AI risk have carefully considered the topic, are aware of the areas where there's still a lot of uncertainty, and make clear and well-hedged arguments that acknowledge that uncertainty; meanwhile the people who scoff at it haven't read any of the arguments (not even in popular book form in Superintelligence), haven't thought about most of the considerations, and have a general air of "assuming things will probably be fine". That's not a straw man, that's just direct observation of the state of the debate. People are doing serious academic work on the topic and have thought about it very deeply; the standard HN middlebrow dismissal is both common and inappropriate.