> For me personally, I don't really know. You can't just do the same thing because the economy is constantly evolving, but I can't see where it's going.
You’re not alone, I’ve absolutely seen the same behavior occasionally with Opus in OpenCode where it takes actions it shouldn’t be able to in plan mode.
Considering it happens across both opencode and other apps like Claude and Codex as well as across models it seems like something inherent to the models themselves and not necessarily a bug in the apps wrapping them. But maybe there’s more opencode et. al could be doing to prevent it.
The harnesses are the part of the stack responsible for tools, so it would be a bug there, not the model. The model itself isn’t doing anything but generating tokens. The harness gives it a blob of text telling it which tools exist, and the model may choose to tell the harness to call one.
“Plan” vs “execute” modes seem more like suggestions the models _mostly_ follow. I have absolutely had models (Codex and Sonnet/Opus) perform actions in plan mode they should never have been able to take like editing files or starting to work on a plan that was just created.
LLMs write like that because people write like that. That they used the rhetorical device three times in a row in a single paragraph makes me think it’s less likely to be LLM and just how that person writes.
“Borderline unusable” is such a hyperbolic way to describe a fully functional design that doesn’t happen to be responsive. Hacker News must be borderline unusable for you as well then, no?
> Hacker News must be borderline unusable for you as well then, no?
On my phone? Yes, absolutely, impossible to hit the links correctly even if I zoom in. Both old reddit and HN is "Fully functional" on desktop, agree, but far cry from "fully functional" on my arguably tiny iPhone.
Is that a ios browser difference? I browse hn all the time on my android phone and I didn't think my screen was unusually big. Maybe they implement some different scaling?
I almost solely use HN on my iPhone browser. It works very well and the scaling is well implemented, although it is a little too easy to accidentally fat finger and vote/flag something without realizing it. I actually find the desktop site (on my laptop) to be a bit hard to use due to its narrowness and small font size, but I'm not sure how universal that is.
reply