Notices by Jason Gorman (jasongorman@mastodon.cloud)

Research finds an interesting correlation between confidence in AI output and belief in the paranormal.

Another correlation I'm seeing with belief in the paranormal is the use of rhetorical tricks to flip the burden of evidence on to the person asking for it.

It's like if the US government denied an export license for Lucky Charms because General Mills claimed they could reverse time.

Developers: "The problem is that the faster we go, the more we get things wrong and have to fix them or do them over again, which actually makes us slower overall"

Big Tech: "I see, yes. What if... hear us out... what if you could get things wrong much, much faster?"

So you read that <your favourite programming thought leader> is vibe coding now, and you've taken that as proof that "AI" produces great code.

I'd probably take it as evidence that <your favourite programming thought leader> is susceptible to social proof like the rest of us.

"Code is now disposable, you can regenerate systems in minutes", he says as I watch him going round and round in circles trying to get Claude to fix a bug.

"Hi. I'm an AI engineer".

You engineer AI?

"No, I use AI for engineering."

Hi. I'm a compiler engineer.

I get it. The boss is asking for an estimate. And you're thinking "But there are so many variables".

Pro tip: the answer is 12.

12 what?

Well, that's their problem.

After more than a year reading engineering leaders talking about how "AI" is gonna 2x, 5x, 10x their dev team productivity, I'm seeing them pivot to "It doesn't have to be perfect anymore!".

My guess is that 2x, 5x, 10x failed to materialise, and "good enough" has been identified as the culprit.

Feels like we should have some Christmas music. It doesn't get much more Christmassy than this.

https://www.youtube.com/watch?v=qORYO0atB6g

"Coding is no longer the bottleneck".

So what you're saying is that NOTHING HAS CHANGED.

Watch software developers. I mean *really* watch them.

When they say they spent an hour "writing code", they've actually spent most of that hour *reading* code and *thinking* about it.

Wanna boost your developer productivity?

* Write code that's easier to understand

* Get better at understanding code

More code ain't gonna help here.

An "AI" coding assistant that shows you the changes it's proposing, but won't let you apply them until you've passed a quiz about them.

"It's really good at summarising code I don't understand".

But, weirdly, it's pretty bad at summarising code I *do* understand.

Recorded history be like...

Neighbour 1: "My grass needs cutting"

Neighbour 2: "I have a goat who will eat that grass. I see you have apples. Let's do a deal."

Dude In the Castle: "What's in this for me?"

Neighbours 1 & 2: "Er. Nothing"

Castle Dude: "My insane friend will burn your houses down"

"At the end of the day, Jason, LLMs are just like a compiler. Do you write everything in machine code?"

I suspect I would if the compiler got it wrong 30% of the time.

Seeing posts from managers complaining their devs have given up on "A.I." coding assistants. "They must be using it wrong. Do they need training?"

At some point, we may have to admit there's no way to get a tune out of a ham sandwich. And it's not because we haven't had ham sandwich training.

@skinnylatte I remember being really taken aback by a girlfriend from NY when she talked about homeless people. In every other respect, she was liberal. But on that subject, really very right wing. And this was 35 years ago.

"Hoorah! We don't need those pesky software developers anymore". Really?

Pause to consider how you'll fare in competition against people who actually know what they're doing.

You might also pause to consider that *when* you need a developer's help, they might not be so keen on cleaning up your vibe-coded mess as a career move.

And, since employers haven't been hiring and training new developers for 3 years, that skillset could be about to get much more expensive.

Why do I believe LLMs are about as reliable as they're going to get? Physics, my dear chums. Physics.

"For the case of α=0.1, reducing the error by a factor of 10 requires the allocation of ten billion more compute resources. And, if we focus on the power consumption itself, the exponent is 0.05, meaning an astronomical 10^20 more computing power for just one order better in accuracy."

https://arxiv.org/html/2507.19703v1