roywiggins

> “rename this function and update callers”

I'm old enough to remember when IDEs could do this without needing a couple gigabytes of matrices to do it

(LLMs are great for anything even slightly more complicated ofc)

Pangram agrees it's 100% LLM written.

pangram flags this whole post as 100% AI generated fwiw (and it reads that way to me)

They could have defended the small claims cases against them in that case?

It happens, eg:

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

Their first study allegedly resulted in "44% were below the cut-off for mild ASD"- ie, practically cured, from a diagnostic point of view.

imho part of the communication problem is that a 6 year reduction actually is large, but it doesn't sound large. Smoking takes about 10 years off your life, and it's deadly in a very mechanical and understood way.

No, that appears to be the implication of this study, which frankly seems like such a large effect that I'm pretty skeptical! I'd say "where's the control group" except the claimed effect is so large that you kind of don't need one, if it's real:

> Prior to the study, 83% of participants had "severe" autism. Two years later, only 17% were rated as severe, 39% as mild or moderate, and incredibly, 44% were below the cut-off for mild ASD.

Emphasis mine. If you are below the cutoff for mild ASD you wouldn't be diagnosed at all.

“The past is a foreign country; they do things differently there.”

Not soon, now. The new reCAPTCHA on desktop shows you a QR code for you to scan with your Google-approved phone to prove you have one.

They can just sell to Aquaman.

> That is textbook enshittification.

Technically, no, not textbook enshittification. Enshittification was originally meant to refer to companies squeezing two-sided markets, not products just getting kinda worse.

not nearly as many words as Ed Zitron at least

It's all fine until OpenClaw decides to start prompt injecting the judge

It could be a stylistic choice, except it's rapidly become an extremely popular one for some reason. It's also the default Claude style. So, take what you will from that. Either someone is writing exactly like Claude on purpose, or they just asked Claude to write something, but either way I'm entirely oversaturated on it. At this point I don't think "Claude", I just start skimming and then close the tab.

It's the em-dashes from a green account.

Meta spent 100 billion dollars on VR, what's a Zuckerbot or two?

I said that people who already know don't think about it. That's not something you can solve by educating them more. When I'm sharing a photo, I am going to think about what I can see in the photo as a data risk, not the invisible stuff that I might intellectually have heard about. It's just not going to come to mind.

People who know about phishing get got by phishing attacks, too. How well has however many years of "cyber awareness training" gone?

You don't have to be irrational to not know things.

Even people who know it, don't think about it and don't connect it with the potential consequences of uploading a picture to a website. And why would they? It's not visible, there's no warning, it's just not something that's going to be top of mind.

yeah I don't hate LLM docs if they're labeled as such. but if someone wants me to use their code or read their README.md they are going to have to make it sound like a human cared about writing it, and right now Claude can't do that

If they're that hard to prompt maybe it's easier just to write the blog posts yourself.

They could write a post demonstrating that you can do that and surface the same bugs in the same codebases.

It would be way more informative than this one, which didn't do that.

"small models can do this if you scaffold them right" might be true, but it wasn't actually demonstrated in the post.

They could have linked their replication in this blog post, which we did all see, if they have one.

Right, but they didn't actually test that, did they?

Maybe that's true, but they didn't actually show that that's true, since they didn't try scaffolding smaller models in a similar way at all.

I think my favorite part is that, because it only applies to "frontier models", if a smaller model is blamed for such harm, it seemingly doesn't immunize the creators at all. That makes very little sense unless you specifically want to make it illegal to not be OpenAI (et al).

Similarly, if a frontier model kills merely 99 people, they aren't covered by this. So go big or go home I guess?

ye olde camera filter