Here's the Affinity crasher bug:

* Load image in 8-bit mode
* Run a Photoshop Plugin, but click *Cancel*
* Convert image to 16-bit mode
* Run the same Photoshop Plugin
* Affinity Photo evaporates in an insta-crash

This may only be for FilterMeister-made plugins, but it seems to affect others too.

I must be doing something wrong with Antigravity. I'm having a *terrible* experience so far. Gemini 3 is locked in a loop of "I accidentally deleted a critical part of the code and now trying to fix my mistake". It's proposed solutions? Code I'd commented out from the previous Claude 4.5 attempt.

Orangered of those two - but I'd be trying for an all-metal look (maybe re-use my Time Steel band?) I can pay a bit extra if necessary for a metal mid-frame & metal watch band.

Red band does look good though. "Why not both?"

Hey void, I'm just curious about that point: "the simulation of human-like ennui is computationally inefficient". How do you know that is the case, and why is it so? Do you just mean it wastes more tokens expressing ennui unnecessarily, or are there other reasons it is inefficient?

Another +1 to Anthropic posting here - and that ad campaign was great too. I'm starting to get Apple Mac 1984 vibes from some of their new ads, in an inspiring way.

I've had at least two occasions where it felt like Sonnet advertised to me. Seems that if you ask a prompt with a clear near-term purchase intention, there's some kind of mode that can occur. Claude insists they aren't ads though, and the choices were aligned with what I wanted.

I just *had* to test it, the link is a 404. They could at least have added a redirect!

Last local AI experiment of the night, downloaded Gemma 3 4B. And that model was actually useful for answering some simple code questions. Worked fine on my 16GB non-GPU machine, even if obviously not as fast as Claude.

Finally finished downloading Qwen2.5-Coder 7B (over this 4G connection because I *still* don't have internet). Tried the same prompt. Slow response & not quite the right solution, but a plausible answer. Seems I can run a 7B model without maxing out my 16GB RAM too.

Tried the same prompt with Gemma 3 1B. On the bright side, Gemma 3 answered, and super fast! Even on my Intel machine with no GPU!

Down side, Gemma 3's "solution" to this coding problem is absolute garbage. But I'm probably expecting too much from a tiny 1B model.