totally agreed on the "focus" aspect of it
that is why i was so happy about composer models from cursor when they came out but they werent very great
imagine opsu 5.5 level models at those speeds and faster.
yeah very true. I asked Opus 5.5 max if even such a thing is possible and it thought so, albeit after a lot of hurdles, especially memory related ones being overcome.
I think you are right and I think we are close to that speed, with codex ultrafast we are near 500 TPS but only with Astra and that is not financially viable. I would guess we will have wide spread near 1000 TPS with 6.1 sol level model with enough usage to, without worry, do the kind of things you describe in about 6 months max
hard part being the ideation is very true. i struggle with this all the time. even meditate on it for hours. but I think it is an important muscle to practice these days
I can understand how you might feel that way. this was the name of one of the first text to image apps I built 4 years ago with one of world's first auto prompt enhancement with a fine tuned currie model
why not! :)
I first thought this wouldn't have been possible actually but here we are :)
Opus used grok cli to generate the images too. a lot of coordination work
this took a lot of interesting ideas coming together tho
Opus 5.5 is instructed to use grok cli to generate the images, then review them and then extract the elements (nearly 100 of them) then combining them in a single scene in layers + picking an mp3 from a folder and matching the scenes dynamisms to the file's waveform etc. i think it is pretty cool :)
I like it cuz it is my idea, but I am also very impressed wiht opus being able to coordinate all of this
The largest gains appear when a task is digital, testable, and repeatable. The closer work gets to factories, field conditions, certification, and human judgment, the smaller the total-program gain becomes
audit of productivity, business adoption, prices, jobs, infrastructure, households, and the feedback loops that would have to appear before the macro story becomes truly extraordinary.
The word singularity is often used for several different ideas: impressive models, rapid adoption, AI-assisted research, self-improving AI, or a complete economic transformation. Those claims require different evidence.
This repository asked a practical question: can a small outer research loop discover agent organizations that solve exact problems better, and can that loop keep improving them over time?
Each numbered folder contains one reusable SKILL.md seed beside the complete run that produced it: protocol, prompts, benchmark, raw agent outputs, failed attempts, scoring, audits, charts, and limitations.