They’ll never admit it

I got a Captcha on some search result a few weeks ago that initially looked legitimate then came up with this.

It had put obfuscated code on my clipboard to download and run an executable.

Yeah i write financial software now and we’re PCI compliant. Devs can’t access prod or its very limited and read only with no access to encryption keys. Those who do have prod access, don’t have write access to the code. No one has both.

All software installs on any company device are reported to ops.

I’m not saying this couldn’t happen but i think it would be way more difficult. Also, we’re a company of 15 people.

I think the doomerism was all to get that moat. There’s no vendor lockin either, even if you’re using the API. You can switch between providers really easily.

Agree with all your points but none of those reasons are we supposedly live in a democracy and the vast majority of us don’t want these data centres.

Not even just boomers anymore. I get AI slop shared with me by all ages. I just tell people now assume nothing online is real.

In UK in 2010, 10 new nuclear plants were announced. Since then 5 older plants have been shut leaving us with 5 active. 1 of the 10 new ones started construction and it’s massively over budget and delayed.

In the same time we’ve added about 20GW of solar.

My first thought. Everything is going to look the same soon isn’t it

I was looking for graphics the other day and the stock images sites I’ve used before were virtually all AI slop. Why would I pay for that?

Probably 10 years ago I encountered a pump on a job. It was huge. 2-3 stories tall. Probably cost millions.

There was a computer plugged into, an old beige tower case and a CRT monitor with RS232 serial cable between it and the pump.

This was the control system, it was connected to the network (airgapped LAN fortunately).

Was running NT4!

I have no idea why it was set up this way, PLCs were definitely the standard at the time and there were older PLCs than that on the site. No one who worked there was around when it was installed, it was just the way it was.

How do you know the software running on the voting machine you’re about to vote on is the software that was audited?

You go get all the machines together, make sure all the software runs on the newest version and it’s correct and it’s configured properly with the candidates. When you’ve got to distribute them to the polling stations, they’ve got to be kept securely between audit and polling day. Your adversaries are foreign state actors who want to manipulate your elections.

And then at the end of the day you have to get the results off the machines. You have to trust that the machine gave you the correct numbers and then you have to get those numbers all together without anyone having the opportunity to manipulate them.

Usually at this point someone comes up with the idea of a paper backup so the machine prints like a receipt of who you voted for that you can put in a ballot box.

If you’re doing that and then count all of the paper ballots to verify the machine count, why even bother with the machines? Just have people vote on a paper ballot with a pencil and count them by hand. The ways to attack paper ballot just don’t scale. You might be able to manipulate a few votes in one polling station, but you’d need a massive operation to do it across the whole country.

Thanks. I didn’t know that had a name.

I hate wasting resources. The service i work on processes millions of transactions a month on a $14 AWS instance with spare capacity. Did I need to do that? No. But wherever we have performance issues I like to fix it before scaling the instance up. It was on the $7 instance for a while but 1GB memory wasn’t quite cutting it.

Saw this coming when Microsoft admitted to have bought up loads of hardware they couldn’t actually use yet. I’m convinced the data centre build out is to hoard all the computing capacity so we are forced to use their cloud services because our own hardware isn’t capable.

If this keeps up in a couple generations every personal computing device will essentially be a thin client with everything running remotely for a subscription.

We’re on AWS at work. I use Hetzner personally.

The main issue with switching is the services. On AWS we use KMS, SQS, DynamoDB. On hertzner we’d have to run an instance with software for that ourselves and then rewrite systems and do migrations etc. That’s the expense.

Also KMS has no alternate because it’s not just software. It’s a HSM. So where do we store the master keys on Hetzner?

Don’t get me wrong, I’d love to do it, not up to me unfortunately.

I’ve been explicitly told not to use any open source for a project. I don’t even know what language I’d write it in. We ended up not even quoting for it in the end.

Yeah this only makes sense if the builds start diverging from the open source code, extra features that are closed source.

Hertzner storage box. I think it’s €13/month for 5tb. Just storage with ssh interface. I use restic but most backup systems should work with it.

No. They handle that. You get so many credits that get used at different rates depending on the model.

You can burn through your months allowance in 10 minutes using Fable then your employer just gets billed the coverage. I think it’s based on the provider’s API pricing so just using Claude or codex directly is probably cheaper.

If you have it do the work entirely and make PRs itself then Copilot is listed as the author regardless of model. E.g. you can assign it an issue and it’ll go and do it itself. I don’t use that though because it’s bad at it.

I tend to just prompt it to write snippets as part as whatever I’m working on to save some typing.

Github copilot is a totally different product from Microsoft copilot. Github copilot is just a harness for any model. I can use openai, anthropic, google, etc. models.

I have at least 1 colleague who does this. Has a Claude subscription despite being provided with github copilot.