mikeocool

If your end point does something like:

* read from the database

* make a request to an API (or really any kind of long running non-database thing)

* write to the database

You're going to end up with a transaction that is open way longer than it needs to be, particularly if you're upstream API is misbehaving, which will potentially end up causing a lot more grief.

I love SQLite, but my clients expect minimal data loss and downtime when one of my servers goes down.

How are people solving that issue with SQLite? Everytime I’ve investigated it, it seems like the state of the art is not very battle tested WAL shipping solutions.

If I’m setting something like that up, suddenly running Postgres with its battle tested replication starts to look not so much more complicated in comparison.

Clean!

If I'm understanding this correctly, you're going to need a LOT of user ratings before the rating content is worthwhile to most people.

Maybe focus the site on capturing ratings to start instead of sharing ratings for products that don't have many ratings (visual studio appears to be highest rated product with 2 ratings).

If you just showed me two related products and had me click on my favorite, I'd probably do that 10 times for no good reason.

Kinda seems like OpenAI didn’t actually have that idea or the ability to execute it, if they had to go to apple’s supplier and lie to them to get them do it.

Yeah, the models have all been really good at generating greenfield apps for a really long time (in the scope of LLM time).

I suppose it’s interesting to see how they make better greenfield apps. But I am much more interested in how they solve hard problems in existing gnarly codebases.

I'm sure they could technically do that if they have the right wording in their terms. But most consumer services are not actually going to give you a paid service without billing info, because it'd create such a headache for them to collect.

Even if they sell the debt to collections, the collections service is probably only going to pay them fractions of pennies on the dollar, since "bill for random service someone signed up for the internet" is not likely to have a high recovery rate.

Yeah I find the Mac mini trend is kind of baffling.

It seems like it's driven either by 1) people hearing Macs are good for AI, buying one, and using Claude for inference, not realizing that you interact with the anthropic API from an internet connected hair dryer. Or 2) people want their agents to have blue bubbles.

I find it hard to believe that enough normal people are doing on device inference is driving Mac Mini's out of stock. And even if they were the Mac mini is not actually a very good platform for it.

My Claude.md is a maybe 30 lines. I only adid something if it repeatedly does something that annoys me.

It still tends to do those things, but maybe 40% less.

I think you’re probably right about people’s feelings on the matter.

Though in my experience dogs that are ok with fireworks are the edge case.

Her instagram account (yourowkelsey) is probably the best thing my algorithm has served up in a while (not a high bar, but hers is in fact really good).

She posted through the trip and answered a lot questions about logistics and the experience.

I think I get your point, but few PhD candidates pay tuition in the US and most actually receive a living stipend. In exchange they act as teaching and/or research assistants.

I drive a 2026 Toyota, how do I enable this new good native infotainment system of which you speak?

The one in my car sucks, and to use the most basic features (navigation, music) that cost nothing on CarPlay (beyond my phone bill) cost $15-25/month from Toyota.

dynamically share resources

This has been a big pain point me with various VM solutions I’ve tried. Having to allocate say 8GB to a sandbox, and a) having that RAM eaten up when I’m not using it and b) only having 8GB when I am using kinda sucks.

Yes, I could stop the sandboxes when I’m not using them, but that also kinda sucks.

I love Postgres. I buy using it for many many things.

I really don’t understand why everyone insists that you should use it as a work/message queue.

There are lots of purpose-built bullet proof queuing systems that are simple to setup and administer (or just use SQS).

Your queue is likely to have very different access patterns than the rest of your data, and sticking it in Postgres means you’re probably going to end up setting up partitions or optimizing auto-vacuum on that table way earlier than you probably need to mess around with this things in your scaling.

If your queue has more than a few hundred jobs a day (or you anticipate that like anytime soon), just use a queue.

Also worth considering (and probably waiting and seeing) if the current AI progress is a net positive for society on the grand scale.

Not saying they didn't have an impact -- just that it's worth debating whether the impact is truly a net positive for society on a grand scale, particularly if we're going to give them a pass on less than stellar behavior.

Zuckerberg likely has similar tendencies, should he be given a pass for making social media such a large part of our society?

I’m not sure it’s necessarily a given that what Jobs, Gates, and Carmack created are net-positives when talking on a civilizational scale.

Though Carmack certainly gets credit for keeping me entertained for a good portion of my youth.

And gates probably gets some civilizational credit for what he’s done with his wealth after he made it.

I apparently use Claude differently the most people who talk about using Claude on the internet.

I’ll typically have a bunch of short sessions over the course of a day. Anytime I start a task that isn’t going to very directly benefit from the existing context I start fresh.

I don’t find a lot of benefit in explaining the project overall to Claude — I’ve deleted a lot of that explanation from my Claude.md because it didn’t seem to impact much.

I typically start a task by pointing it to 1-2 files and giving it some explanation of what I want done, and it figures it out.

Basically never hit context window limits or compactions, and can’t remember the last time I hit a 5 hour or a weekly limit.

I don’t think you go lucky — renting from small landlords who live in/near building is the way to avoid this. In my experience, say the word “leak” they’ll be there in 10 minutes. And good tenants have value to them.

At two years with a corporate landlord, they’ve done the math and see they can cheaply renovate your apartment, jack up the rent, and make more money than they can by renewing you.

Small landlords also haven’t built software to collude with each other to raise prices.

Whether or not scrapping publically available websites is unethical is probably up for debate. In some cases at least, courts have found it to be legal, even when the site is throwing up technical barriers or issues cease and desists.

What is likely unethical is the fact that they offer residential proxies. The residential providers of those proxies are frequently not aware they’ve been opted in to provide such a service.

I dunno, I think in the past year “AI” has gone from meaningless buzzword to having a negative connotation amongst the non-tech population.

“That’s so AI” is legitimate slang and it does not mean “that’s so cool and automated!”

Seriously. Whenever I read the thinking output I get mad and turn down effort to medium or low.

Just output the code and we’ll work through it!

I feel similarly about having codex review claude’s plans. I don’t think I’ve ever seen it catch a major issue. It just points out things that would have inevitably been addressed during implementation anyway.

Not using beta ingress was probably a non-starter for a lot of people, since it was the only option for 4 years.

Then there was an upgrade process that required a fair amount coordination between when you changed your manifests, when you upgraded your cluster and when you upgraded your ingress controller.

PodSecurityPolicies also gained a lot of traction and didn’t really have an alternative before it was deprecated.

Also, custom operators don’t all subscribe to the don’t break non-beta resources in the same way core does.

Even with all its complexity, k8s doesn’t solve every problem — good luck running an FTP server or anything that needs to dynamically allocate a large range ports on k8s.

I would absolutely trade flexibility for complexity. Particularly for edge cases like hard coded ports.

People think they want infinitely scalable solutions. But I think what they fail realize is that by the time they actually hit a scale where it matter they’re going to have to change so much about their infra that prematurely using the scalable solution didn’t really but them much except a bunch of headaches when they didn’t have scale.

I’d take a cronjob running cert-bot and some monitoring to ensure a domain’s cert isn’t about to expire over cert-manager any day.

IIRC cert-manager has about three layers of custom resources to conv through when figuring out why a cert isn’t renewing.

There was once a time when we could deploy software without spinning up 3 etcd databases, multiple controller processes constantly running event loops, and a virtual networking layer, before you even get off the ground.

Perhaps those days are behind us.

I don’t have an answer I’m in love with today, I basically just want less moving parts.

As for EKS, having to monitor and manually scale the built in DNS service or else my queries are just going to stop resolving is not the type of thing I expect to have to manage on a managed service. I see they have finally released autoscaling for CoreDNS, though it took them 6 years.

Yeah I’ve always meant to check out nomad and never had an opportunity.

Though as I recall, it makes heavy use of consul, which I have used in anger, and makes me a little weary (though that experience is likely very out of date).