⬢ chaos dancing star @paradigmainc
Icecrown Citadel
Joined February 2022
- Tweets8.7K
- Following449
- Followers5.9K
- Likes42.9K
several more quality of life improvements out today
Finally, a round of quality-of-life improvements to Flywheel:
- Pin a graph at the center of your universe.
- Navigate Free View and Timeline with the Navigator.
- Search within one graph, or across your whole universe.
- Filter a node’s lineage and show related nodes on the map.
- Send feedback from inside the app.
- Rebind keyboard shortcuts (improved).
- Choose whether scrolling pans or zooms the canvas.
- Detect and set up your local coding tools from Settings.
tensorqt in SF retweeted
Finally, a round of quality-of-life improvements to Flywheel:
- Pin a graph at the center of your universe.
- Navigate Free View and Timeline with the Navigator.
- Search within one graph, or across your whole universe.
- Filter a node’s lineage and show related nodes on the map.
- Send feedback from inside the app.
- Rebind keyboard shortcuts (improved).
- Choose whether scrolling pans or zooms the canvas.
- Detect and set up your local coding tools from Settings.
Today: we're bringing back tag clusters, with a brand new look.
Sometimes you want a visual queue of what's what.
Tag nodes with "cluster: your name" and Free View shades them as one area, at any zoom, effectively creating visual groupings for your Flywheel nodes.
Clicking a tag cluster in Flywheel also opens it in the side panel: how many nodes it has, where they sit on the canvas, with sort and filters.
A bestiary of chaotic attractors, each carrying its own dihedral Dₙ symmetry.
Every disk holds a different attractor of the generalized Field–Golubitsky map:
z ↦ {a₀ + a₁|z|² + a₂ Re(zⁿ) + a₃|z| Re[(z/|z|)ⁿᵖ]} z + a₄ z̄ⁿ⁻¹
#builtWithMarimo
tensorqt in SF retweeted
i am officially naming this genre of company “cringelabs”, characterized by opaque attempts at claiming sovereignity by dubious fine-tuning and boomer bamboozling
tensorqt in SF retweeted
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
I use these all the time
Today: we're bringing back tag clusters, with a brand new look.
Sometimes you want a visual queue of what's what.
Tag nodes with "cluster: your name" and Free View shades them as one area, at any zoom, effectively creating visual groupings for your Flywheel nodes.
Clicking a tag cluster in Flywheel also opens it in the side panel: how many nodes it has, where they sit on the canvas, with sort and filters.
tensorqt in SF retweeted
Today: we're bringing back tag clusters, with a brand new look.
Sometimes you want a visual queue of what's what.
Tag nodes with "cluster: your name" and Free View shades them as one area, at any zoom, effectively creating visual groupings for your Flywheel nodes.
Clicking a tag cluster in Flywheel also opens it in the side panel: how many nodes it has, where they sit on the canvas, with sort and filters.
tensorqt in SF retweeted
As we scale the number of agents on a problem, Flywheel’s research nodes/graph grows by an order of magnitude.
We needed a new way to interact with our own research, something that takes us straight to the critical results among 10k+ nodes produced overnight. So we built Envoys
tensorqt in SF retweeted
Custom views, just by asking.
Ask Envoys for a lens on your research and it tags, filters and reorders your graph around it, from just the results behind one question to a single line of work, in order. No digging through everything to find the part that matters.
tensorqt in SF retweeted
Plan what’s next beside the evidence.
Envoys turns your results into the next experiments, linked to what they build on. You watch them land on your graph as it works: new nodes marked, a counter of what’s waiting. Agents stage the work; only you commit it.
tensorqt in SF retweeted
Next up: Envoys.
Your own Claude Code and Codex, working on your research graph without leaving Flywheel: ask it anything, add to it as you go.
They run on your own setup: your settings, skills and subscription. That makes Envoys free.
Today, we’re giving the timeline view some long-overdue love, with some polish and a new look.
In the new Flywheel Timeline, the most recently updated graphs come first.
In the local app each card also shows its folder and how many uncommitted changes it has, so you can see where work is waiting to be committed.
super interesting direction, co-signed by @paradigmainc
