Aren't these increases offset by the quality of the responses and reducing the iterations needed to fine-tune the responses?

Only for the range of tasks where 4.7 performs well but 4.6 performed suboptimally. If both models can one-shot the task without retries, then the number of iterations is already at the lower bound.

This also applies at the sub-task level. If both models need to read three files to figure out which one implements the function they need to modify, then the token tax is paid for all three files even though "not the right file" is presumably an easy conclusion to draw.

This is also related to the challenge of optimizing subagents. Presumably the outer, higher-capacity model can perform better with everything in its context (up to limits), but dispatching a less-capable subagent for a problem might be cheaper overall. Anthropic has a 5:1 cost on input tokens between Opus and Haiku, but Google has 8:1 (Gemini Pro : Flash Lite) and OpenAI has 12:1 (GPT 4.2 : 4.2 nano).

Likely only some of the time

Amazing work, I am a fan of unicorn studio, it was the first real try I saw at these no code design tools that have shaders as their output.

I wonder what makes you unsure that you would have not bothered in building it today? or what would you do different with todays knowledge?

Thanks!

I still would have built it today, what I was unsure of was whether I would have learned GLSL.

The models have been sub-par at writing shaders until recently. Without the friction of needing to learn the language itself to build it, I don't know if I would have.

The most fun I have today is trying to innovate new interesting techniques, which I wouldn't be able to do with the models alone.

I would love to have a way to switch sessions easily, like with panes where I just click or have more shortcuts available

Here’s a simple session switcher: `bind C-a display-popup -E "tmux list-sessions -F '#{session_name}' | fzf --reverse | xargs tmux switch-client -t"`. No plugins, nothing fancy, just works.

i just have a little script called session (invoked as session namespace name and mapping to a tmux session called namespace/name and searching all namespaces if the namespace is elided) that just does the right tmux invocation for the various intersecting cases of [in or not in a tmux session already, target session does or does not exist] which also has the side benefit of leaving shell history breadcrumbs i can follow between sessions later if needed

People have developed plugins for this. Check out “sesh” for instance.

Switching between sessions with fuzzy finding, and creating new ones when needed, is a wonderful feature.

I use byobu for managing tmux windows and sessions https://www.byobu.org/

Check out the `choose-tree` and `display-menu` commands.

Prefix-w works, or did you want something more?

Isn’t AI affecting the demoscene as it is affecting other programming hobbies? Why?

There have been lots of discussions around AI in the demoscene. Revision limits how you can use AI (https://2026.revision-party.net/competitions/general-rules/), even if it's not always enforceable.

There's a lot of push back against AI-generated graphics and music. For code, it's more difficult to know. AI is used by some people to automate the boring tasks, so that they can focus more on the artistic side.

At revision there was a cool seminar about the author of a music synth who used AI to modernize it. He begun his talk with words for the audience along the lines of "Please don't do a Life of Brian, I am not here saying Jehova".

It makes sense that a creative medium with a long tradition of pushing boundaries of what people can create, frowns on use of generative tech unless you have created it yourself. Back in the day the pushback was against using AMOS, or a PC, or programming in C, or using a GPU, or using MP3, or using Photoshop, or using another group's demo engine, or using a commercial game engine, or... AI is just the latest. And like its predecessors, it will gain legitimacy if people create genuinely interesting experiences with it.

Game engines and their useage hit harder

Demos are now often using Unreal engine Unity Godot

Why do all vibecoded sites look the same? Same black on neon vibes and button styles

and it is not end-to-end encrypted if you don't own the keys, avoid bullshit

we are experiencing marketing at its best

Also Netherlands, also France, also Denmark, also Spain (melila and ceuta)... actually Portugal kinda is still the first global empire that managed to shift away from it

and just how did they got the gunpowder? ;)

This is one resentful individual. Likes to imply how this or that people is inferior to the other (I thought we were discussing differences in forms of settlement, colonization and maritime expansion) then pivots to modern day economic statistics to again imply that some people are superior to others then finally succumbs to racism but is careful enough to change the language!!

...and yet you speak portuguese while trying to dismiss (and rewrite) a post about 15th century events with data from the 17th century

Spain didn't exist back then

Establishing when did Spain become Spain is complicated, but a commonly agreed date is 1480, following the Cortes of Toledo: https://en.wikipedia.org/wiki/Council_of_Castile

...more than 160 years after the portuguese navy was founded, and 20 years after Henry the Navigator was dead. Still not as big of a gap as those 19th century references that you linked to reply to a post about 15th century events

I'm not sure what the organizative reform of Spain has to do with the founding date of Spain and Portugal's navies. In any case, the Spanish Armada is the result of joining the Castilian and Aragonese navies, both dating from the early XIII century: https://armada.defensa.gob.es/ArmadaPortal/page/Portal/Armad...