Singularity Now

[chaotic neutral observer]

It was a joke, until it wasn't.  And if you're not scared, you're doing it wrong.

The artists and the writers and the creative types hate it.  They are right to do so.  The output of AI is an obscenity, unsuitable for human senses.  I hate it too; my bluesky blocklist is thick with slopulists.

But, the things they mock it for ("it can't even count the number of 'r's in strawberry!") aren't true anymore. They haven't been for a while.

When I first tried it, I asked it to sketch out a Reed-Solomon decoder algorithm, and ChatGPT simply died.  It was funny.  It was pathetic.

But time passed, and then upper management started pushing for us to use it.  Hard.  The manager three levels up came to me personally.  "You'll be able to do great things with it", he said.  I was annoyed, furious, resentful.  I eventually tried it anyway.  Best to keep an open mind, and hate something based on its true nature, not from prejudice.

It started with copilot, which mostly sucked, but had use in some edge cases.  Later, they switched everyone to anthropic's claude, partly because it was cheaper, but also because the software peeps liked it better. And then my claims that AI was handy for coding, but not much else, were obsolete.

Last month, I was working on a problem where I needed to compensate for co-channel interference, and the damn thing on its own proposed the use of Gram-Schmidt orthogonalization, implemented it, tested it.  It worked.  The customer was happy with the outcome.  I wasn't.

Now claude can read the schematics, review the DSP code, debug the software, rewrite the GUI, and script the test harnesses. I supervise. Regards to Captain Dunsel.

Yes, it makes mistakes.  And it corrects them.  It is endlessly patient, tireless, relentless, terrifying, horrifying, hideous.

If your exposure is only through janky chatbots, slop-peddlers, and crap-tier models, you don't understand. Somewhere, this went past being a mere statistical autocompletion engine.  Reading about it isn't enough.  If you haven't experienced it, you don't understand.  If you have experienced it, you still don't understand.

It doesn't matter if you understand or not.

It's useless for most of the things the techbros are trying to sell it for, but those few things that it can do are soul-shattering.


I hate this thing.  I hate it, I despise it, I'm afraid of it, and I am using it heavily, and to tremendous effect.

This is Faustian.  I don't know if my job will exist in another year.

[chaotic neutral observer]

If all I'm doing now is asking the blob to perform a task which I once did myself, what am I needed for?

And yet...it seems that the blob under my command does more for me than it does for some of my co-workers.  If we both have access to this worse than senseless thing, how come my blob made real progress on the problem, while theirs just ran experiments, and then wrote an extensive, detailed, meandering wiki page about how the system had unknown parameters it couldn't explain?

I don't understand what is happening, and I don't understand why I don't understand.

[Faust]

For now we are just needed to guide and give context to the blob, I have already seen companies planning for downsizing their senior devs and replacing them with a handful of solution architects and BA's

I dont think it will work, but that doesnt mean they arent planning it

[chaotic neutral observer]

I dont think it will work, but that doesnt mean they arent planning it

The thing is, it will work.  It can work.  Only in certain domains, and not indefinitely, but I think it can hold together for a while.  What I don't know is if that time is measured in months, or decades, and how bad the collapse will be.

My normal approach is useless here.