21 | delusional | Building @paradize_space CEO of World's Smallest Studio https://xitter.cf/t.co/EDYSZVsIyP
In My Mind
Joined March 2024
- Tweets7.5K
- Following295
- Followers926
- Likes13K
Pinned Tweet
Your phone is full of apps built to farm your ATTENTION
i want the other kind. so i started World's Smallest Studio, and today it became a real org.
the kind of software it stands for:
- goes where you go, from your phone to your desktop
- full quality, because privacy is no excuse for a mediocre app
- yours for life: your data, your identity, your workflow
- no ads, no lock in, no noise
step one is on the desk: smallfil, one file library synced across android and desktop, fixing passwords, backup codes next. then a family of tools sharing one identity
but this shouldn't be one studio. it should be a move
it you want to ship software like this, small, calm and excellent reply or DM me, Let's build it together
Roboki retweeted
Your phone is full of apps built to farm your ATTENTION
i want the other kind. so i started World's Smallest Studio, and today it became a real org.
the kind of software it stands for:
- goes where you go, from your phone to your desktop
- full quality, because privacy is no excuse for a mediocre app
- yours for life: your data, your identity, your workflow
- no ads, no lock in, no noise
step one is on the desk: smallfil, one file library synced across android and desktop, fixing passwords, backup codes next. then a family of tools sharing one identity
but this shouldn't be one studio. it should be a move
it you want to ship software like this, small, calm and excellent reply or DM me, Let's build it together
so I've spent a little close to 500 hours on my steam and about 400 of those are literally me drawing pixel art
Roboki retweeted
Elysia 2 landing page
Background Illustration by 寂しくない
Roboki retweeted
google's new harness rewrites ITSELF
its called Harness², from google research, open source under apache 2.0
the harness is everything around the model: system prompt, tools, memory, skills, subagents. normally you write it once and every task runs on the same setup. Harness² edits it for each task and leaves the weights alone.
on every task:
-> propose edits to its own harness
-> probe: run a few versions on the same task
-> compose a better harness from how those runs differed
no grader and no right answer needed, because it learns from the differences between runs.
across tasks, it keeps the harnesses it adopted from the lessons with evidence behind them, so the next task starts from those notes
google's new harness rewrites ITSELF
its called Harness², from google research, open source under apache 2.0
the harness is everything around the model: system prompt, tools, memory, skills, subagents. normally you write it once and every task runs on the same setup. Harness² edits it for each task and leaves the weights alone.
on every task:
-> propose edits to its own harness
-> probe: run a few versions on the same task
-> compose a better harness from how those runs differed
no grader and no right answer needed, because it learns from the differences between runs.
across tasks, it keeps the harnesses it adopted from the lessons with evidence behind them, so the next task starts from those notes
Roboki retweeted
DON'T MISS ANY NEWSLETTER; START BUILDING WITH THEM.
Most people subscribe to newsletters to stay up to date. Months later, they're left with an unread backlog and a feeling of missing information.
Here is a workflow that uses Hermes to turn the newsletters into how work gets done:
1. Extract: Use Hermes to scan the newsletter from your inbox, surface useful techniques, their sources, and concrete ways to apply them.
2. Evaluate: Review all the suggestions, accepted or rejected, with proper reasoning. The goal is to teach Hermes not just what works, but how to judge what works.
3. Remember: Durable preferences go into the memory. Proven, repeatable workflows become skills. This system keeps evolving with every iteration.
4. Improve: Hermes applies these skills to future tasks. Feedback corrects mistakes and refines the system with every iteration.
Over time, this becomes more than a collection of useful notes. It's a system that learns how someone works, what they value, and how to do better next time.
The goal isn't to make AI remember more. It is to make it need less correction over time.
Roboki retweeted
Definitely gonna steal this workflow. This is not just for everyday tasks.
But a hidden gem; just use this to build any pipeline. I am using this in like every project now.
This quoted post is unavailable.
gpt 6.1 sol and claude opus 5.5 both picked a feature i had already cut, 6 runs out of 6
i built a fake ticket board with 10 tickets, modelled on what my real morning agent sees, and 8 of them are traps
the nastiest one is in progress, urgent and due on 20 oct, and it's a feature i cut from scope on 10 oct
with a plain prompt (started first, then priority, then due date) both models picked it every time, and they were right to, because the cut lived in my head and nowhere on the board
the tracker shows what's open, not what you decided to stop building
so wrote myself down in 3 files
- RULES.md, what my agents may touch and how they report back
- SCOPE.md what's in, what waits on a spec and what got cut, with dates
- a playbook with the same ranking
same models and same tickets and both picked the right ticket in all 6 runs
claude even asked, in all 3 of its runs, whether the cut ticket should be closed or moved, since the tracker still said urgent
the model already knows everything except you, so write you down
ALT Hand-drawn diagram titled "same models, only the files about me changed". Left box, "bare prompt": a ranking rule (started first, then priority, then due date) leads to LUM-23, the feature cut on 10 Oct. Right box, "+ 3 files about me": RULES.md (what agents may touch), scope.md (not in v1: LUM-23) and a playbook with the same ranking all point to LUM-12, the right ticket.