I had the opposite experience when Gemini 3.1 first came out. It didn't show up as a model option in my fully updated Gemini CLI, and I subsequently figured out I had to install this Cursor-lookalike thing called Antigravity to try it. I'd like to stick with my existing editors, thanks.
On a related note, the AUR package previously named antigravity has been renamed antigravity-ide[1] after some lively discussion, and the new thing lives at antigravity2-bin.[2]
I switched to Niri at the end of last year after over a decade on i3.[1] Having horizontal scroll unbounded by my monitor size and workspace count unbound by the number of shortcut keys I have configured has been very freeing, and the graphical stuff is nice too.
My only remaining pain point is that its X compatibility layer, xwayland-satellite, does not yet support drag and drop between X and Wayland programs.[2]
I'm in a similar boat, but I've found where I once habitually put things in the same workspace every time and was able to trivially recall them, I now end up all over the place.
Also I've been missing scratch deeply.
I'm sure it's solvable with some diligence and config changes, but I haven't invested the time yet.
I'd agree with sph about having one workspace per activity. I've never had a rigid workflow with lots of permanent named workspaces, but I have a workspace-naming script that lets me label my numbered workspaces after they've been set up.
Other things that help include a fuzzel-based open window searcher and, to be honest, restrained use of Niri's flagship scrolling feature. Most of my workspaces most of the time are the same size as my screen, with the scroll used very sparingly for usually temporary overflow.
I guess it also helps that I never used the i3 scratchpad so I don't miss it.
I mentioned elsewhere that scrolling WMs shine when you use a workspace per activity. You should never "have stuff all over the place", you should be working on a single one until you context switch.
>LLMs are not doing searches, they are doing statistical computation of likely results.
This was true of ChatGPT in 2022, but any modern platform that advertises a "deep research" feature provides its LLMs with tools to actually do a web search, pull the results it finds into context and cite them in the generated text.
In my last conversation with a Google support person, I was sent a clearly LLM-generated recommendation to switch to a competitor's product. Either they're not doing this, or the support person wasn't using Gemini.
It's standard practice for customer support people to chase away unprofitable customers (in the US; no idea how Google works). Human or LLM, they may simply not want your business.
ChatGPT's image generator has been able to do this since last year. That NBP still can't is baffling. They should at least train it to respond to requests for transparency with a solid colour pink background.
This. Gpt-image-1/1.5 are the only ones that have this built in - though I'd love to have an insider view if its natively considering the alpha channel or just feeding it through a rembg-style post processor.
Interesting read! A lot of AoC challenges involve navigating 2D grids, which can map quite nicely onto the text adventure model of connected rooms with compass direction exits (a grid of straightforward little passages, all alike). This insight led me to attempt Day 6 from last year's Advent of Code in Inform 7[1], though I ultimately admitted defeat on the second half. I've always found Inform 7's Mathematics Textbook English syntax quite charming, though perhaps I would have a different perspective if I'd ever attempted to build anything substantial with it.
Last year was my first participation and did everything in javascript in the browser. It’s high level enough to not lose your time in details, you have a graphical output if needed (canvas), text output, threading, parsing, …
I'm trying to complete each challenge in a different language. I've written up the first five days on my blog[1]. Have completed a few more than just these, but I don't know that I'll get all of them done by Christmas.
I wrote a short blog post[1] about this pattern a few years ago. I still don't understand why so many apps use this workflow, especially ones that previously didn't. Just let me deal with the project creation stuff on the first write instead of requiring all this up-front commitment!
reply