The failure mode I keep seeing: the suite silently becomes the objective.
Once the agent can add scenarios and reweight, it starts farming easy variants instead of the hard ones you cared about an hour ago. Do you freeze the task distribution (or the grader) outside the editable tree, or is "what counts as better" itself up for revision each loop?
Can an LLM read weights directly? Following Sparks, we picked an easier target: LoRA diffs. No trained reader, no running the model. Claude Opus 5.5 gets a Python environment and the LoRA weights, and has to say what the LoRA implants. 🧵
Made a fridge app for foldable phones haha.
Vibe-coded entirely on my phone in just one hour.
Smash things, leave notes, and keep photos of the people you love😘
#Apple#iPhoneDuo#vibecoding
Vibe-coded my own take on Tiny Computers, entirely on my iPhone Duo.
All I did was tell Fibona what I had in mind.
Childhood, rebooted!
Try Fibona and bring any iOS app idea to life, right from your phone.
TestFlight: https://t.co/U9a6I5tdzy
Beta invite code: YR9Y-MQTA
We've turned this open-source side project into a real app, and we'd love for you to come build with us.
What makes it different:
1. Build any iOS app, with no compiling. Describe it and start using it right away.
2. Put the agent to work on your phone. It can edit your photos and videos, plan your schedule, or even train a classifier.
3. iPhone is all you need. There's no cloud sandbox and no computer to connect.
We've only scratched the surface of what it can do. Try it out, and show us what you make and how you use it!
- TestFlight: https://t.co/SuSynnAKrm
- Beta invite code: YR9Y-MQTA(0/100)
- Codes all used? Join the waitlist and we'll open access in waves: https://t.co/20Q8o3l92X
Beta testers get free tokens to get started. Happy building!
Introducing Fibona Agent: the last agent humans will ever need to write.
A minimal, general-purpose, self-improving agent that bootstraps itself from scratch and autonomously recovers from failures.
Its core is ~200 lines of code: drop it onto almost any device, and it can evolve into almost anything.
Fully open source, with a free online demo available!
Learn more👇
#RSI #AIAgents #AI #OpenSource
@Charles_Y_Wu One RSI setup spanning coding, GUI, and science is ambitious. On the eval side, when world-model predictions guide the next iteration, is progress ever measured against those same predictions, or is it always validated against actual environment feedback?
A little side project from us: Tabgent 👋
Your agent in every tab.
Open Codex in Chrome, already connected to your current tab.
Ask about tabs, compare hotels, explore GitHub, read papers and manuals, and fill forms.
Open source: https://t.co/mG3WEFfOot
I’m building Sumus, a shared space for you, your friends, and AI teammates.
Your AI teammates have their own computers. They can work together and keep going after you log off. Your Codex or Claude Code can join as a teammate too (BYOA).
Free to try. Link below.