AI workflow field notes: OpenAI / Claude / Codex / agents. CN deep dives | EN quick takes | JP notes. DM workflow for template.
Joined December 2022
- Tweets3.7K
- Following4.9K
- Followers511
- Likes17.8K
VikingThunder retweeted
Every day i'm blown away at what i can do in an hour with frontier models like Opus 5.5.
I didn't know what to do with an old iPhone, so I built Bluey, w/ Opus 5.5. It's powered by GPT Realtime 2
(i think... whatever Opus decided to use).
Bluey is an iOS app + mac app that is made to be a computer companion. It can point, comment, and even control my entire computer.
github repo below
VikingThunder retweeted
Claude added creepers to Dark Souls 1. I've been curious what Opus 5.5 can actually do, and this is kinda cool.
These models are getting so genuinely powerful. the fact it can just basically pull an enemy from a different game in a entirely different engine, with a different coding language, and is able to translate and bridge the gap is kinda nuts. Especially dark souls considering it’s one of the harder games to mod
VikingThunder retweeted
reinventing garry's mod from first principles
OpenAI 的把原本应该属于不同定位的产品，强行合并在一起。不是产品设计问题，只是凸显了其团队对模型应用的没有节制的探索
The ChatGPT app is now such a clusterf**k.
There's now 4 ways to do similar things!!!
- ChatGPT Chat: you can do web search and some tool calls, but you can't write code on the fly (well)
- ChatGPT Work: it's like Codex (filesystem access + can write code), but in the other tab
- Codex: your projects here don't talk to your projects in ChatGPT
- Dot: how is this different than Work? Besides having a dedicated cloud VM, but it can still work on my computer?
IMO, the ideal hierarchy should be:
1. Workspaces. Everyone starts off with a default one. I can separate out my workspace for "personal stuff" vs "coding projects" vs "work stuff". I can also have my @conductor_build threads in its own workspace so it doesn't pollute my non-Conductor threads.
2. There is no "Chat vs Work vs Codex" - all threads have superset capabilities.
3. Dot is just the master "orchestrator" thread. It can spin up and interact with threads in my workspace, or just do work directly.
Pls fix @thsottiaux.
VikingThunder retweeted
Meet Ling-3.1-flash: ~560B total params, ~25B active/token, up to 1M-token context.
We plan to open-source the model soon.
Across work, coding & healthcare: 1,673 Elo on GDPVal-AA v2.1, 75.16 on FrontierSWE, and 65.35 on HealthBench Professional.
VikingThunder retweeted
I wanted to see how else @specs might be used in the performative arts. They capture my voice, hands, head and world pose, while Unreal drives the rendering. All in realtime, in a very natural and human way. And, of course, cordless.
VikingThunder retweeted
Finally, Karts.com is live! 🏎️ I still can’t quite believe I built a 3D kart racing game. Come give it a try! 😄
I created the characters with Nano Banana and ChatGPT, turned them into 3D models with Tripo3D, built the tracks with Claude Opus 5.5 and Blender, and brought it all together in Three.js.
Thirteen of the 20 cities are places I’ve visited. I poured my memories of them into the tracks, and I hope you’ll enjoy the scenery as much as the racing.
#GameDev #ThreeJS
VikingThunder retweeted
Introducing Unreal Agent:
An open-source harness with state-of-the-art cost efficiency
39% cheaper than Codex+Astra on Terminal-Bench 4.0 while maintaining performance
VikingThunder retweeted
We’ve been experimenting with responsive generative video interfaces. Real-time video models are reshaping responsive design. A thread 🧵
VikingThunder retweeted
Open-source hardware finally has a homepage.
7,600+ designs you can actually browse.
1) Browse by hardware: robotics, RF, power, IoT, keyboards...
2) Browse by platform: ESP32, Arduino, STM32, Raspberry Pi, RP2040, FPGA...
3) Open any project directly in our browser EDA + MCAD
But finding the design is only the start.
AI lives right next to the editor and understands the project you have open.
Ask how the circuit works. Ask why a component is there. Explore the PCB + mechanical design. Then fork it, change it, and make it yours.
heypcb.ai/world
p.s. we're not trying to make another place to store hardware files. we're trying to make hardware searchable, understandable, and editable.
See what interesting projects people around the world have developed using Jev! #Jev
github.com/JackZeng/Jev_apps
Nearly half a year of silence. We spent it studying one problem: how far RL can scale.
MiMo-V2.6 is in the middle of its RL run right now. Three things we scaled: compute (~2B tokens per step, 1568 prompts × 16 rollouts, fully async), environments and harnesses (multi-task agentic RL, mixed across multiple harnesses in one run), and grader compute (agentic in-group credit assignment, with test-case and rubric-based rewards). We'll open-source the details piece by piece over the coming weeks.
Streaming the run: mimo.xiaomi.com/rl/
VikingThunder retweeted
Today we’re unveiling Odyssey-3, a big step forward for foundation world models.
It can control robots, power humanoids, drive cars (on the roads of India!), train AIs, pilot drones, and even play video games.
We can’t wait to see what intelligent systems it enables.
VikingThunder retweeted
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
VikingThunder retweeted
You got a micro-duck or want to train your open duck and don't want to spin it up yourself? We're releasing RL studio at @trainrobotsfast.
Describe with text, upload or generate a video and download the policy.
多种多样的模型正在喷涌而出，多么伟大的时代！
human wiring vs AI: who hears emotion better?
We built a model from H01 human cortex wiring and put it against a tiny frozen transformer.
Same speech. ~592k fixed weights each. Only the output layers learned.
Then we zoomed in.
@OrukLabs
VikingThunder retweeted
I’m excited to announce the world’s first Fly Language Model.
It uses a new architecture called GPF (Generative Pre-trained Fly). It adapts a small pre-trained model with the actual fly connectome. Don't believe it's real? Try it out below, view the code, and read the paper.
What will you ask the fly?