Contractor. Builder. Modern stoic. SI optimist. Building practical tools & systems that create clarity, freedom and abundance. 🛠️🤖 💍❤️
Florida
Joined March 2009
- Tweets37.4K
- Following897
- Followers1.2K
- Likes93.1K
Pinned Tweet
Apparently, brainstorming with ChatGPT while driving and chatting with Codex between dinner and TV is how video games get made now. 😂
Meet ORBITAL. 🪐
One button. Infinite gravity. Race the clock. Chase the leaderboard.
Built last night. Playable today. Updates to come.
orbital.thatsiguy.net/
A humanoid company is a stock ticker now.
Agility Robotics started trading on Nasdaq today as AGLT, the first U.S. pure-play humanoid listing. Digit already has tens of thousands of real warehouse hours, and they claim $300M+ in Digit 5 orders. SPAC brought in about $465M.
Rough day to open but public markets will now mark the demos to reality every quarter.
That's the feature. 🤖
nitter.cf/agilityrobotics/status…
Today, Agility begins trading on the Nasdaq Stock Market under the ticker symbol AGLT.
Read the announcement: agilityrobotics.com/content/…
Buy the dips. As always.
Generational moment to buy...
Why? Let me explain, in under 60 seconds, the game on the field for Team Blue:
First, the democrats aren't as radical as the GOP painted them (just like MAGA isn't as bad as Dems painted them)
Second, the Democrats can't win long term if they kill the markets (aka the AI trade) with regulations, taxes and by vilifying the job creators.
It is critical that we support moderates in the Democratic Party like @RahmEmanuel and steer the rest of Team Blue toward being pro-capitalism, AKA pro-jobs and economic growth.
The last thing democrats want to be associated with is tanking people's 401ks and causing layoffs and a recession.
@RepJeffries understands this...
SI automated everything except building SI. Oumi's FAiE goes after that last gap.
Describe a task in plain English, give it ~15 min of input, and a few hours later you get a custom model with trained weights. Evals, failure diagnosis, data curation, fine-tuning, all handled.
Owning a small model built for your task beats renting a generic giant.
Build your own. 🛠️
nitter.cf/i/trending/21182004270…
Baidu just dropped ERNIE 5.5 Flash Base as open weights. 280B total, only 14B active, 1M context.
It's a raw base model with no chat tuning yet. Which is exactly what builders want to fine-tune on.
Another week, another frontier-ish model you can actually own.
Own the weights. 🔓
nitter.cf/CodeByNZ/status/211822…
baidu just released ernie 5.5 flash base as open weights
- beats deepseek v4 flash base on 26 of 30 benchmarks in baidu's tests
- 1m context
- 280b total, 14b active
- trained on 33.5t tokens
and this is just the base model with no chat or reasoning training yet. i really want to see what the finished one can do
I charted The Ashen Branch in 00:18.308 on ORBITAL. Can you beat it? orbital.thatsiguy.net/c/1-xk…
"AI-written code is slop," they said.
Axiom just shipped 16 million lines of Lean covering the whole Stacks Project, the open reference modern algebraic geometry runs on. Agents did the heavy lifting with mathematicians steering, and every proof is machine-checked. No gaps.
The viewer still needs polish.
But a machine-checked foundation for an entire field of math, free for anyone to build on? That's the good kind of slop.
Math, verified. 📐
nitter.cf/axiommathai/status/211…
A 65B coding model trained from scratch for ~$32M, and they say it beats much bigger open models on Terminal-Bench.
Apache 2.0. 1M context. Weights on Hugging Face.
Frontier-ish coding you can own keeps getting cheaper.
Own the stack. 🛠️
Today we release Ripple-65B Preview, our 65B dense LLM we pretrained from scratch.
Ripple sets a new state-of-the-art for U.S. open-weight models in agentic coding tasks, surpassing significantly larger models including Muse Spark Open-816B, Beam-501B, Inkling-975B, and Nemotron 3 Ultra-550B.
Learn more: ricursive.com/ripple
A 4B model trained across GPUs scattered around the world, linked over the plain internet. No mega data center. Apache 2.0 weights.
"You need a $10B cluster to train anything" just got a little less true.
Decentralize the forge. 🛠️
nitter.cf/MacrocosmosAI/status/2…
Introducing IOTA Decide 4B, our first public model.
Pareto optimal. Built different.
Trained across globally distributed GPUs with our SDK.
Open weights, Apache-2.0.
🤗 huggingface.co/macrocosm-os/…
Tencent just dropped Hy4. 770B params, 49B active, 1M context, and it can actually use a computer now, CLI and GUI. Trained on Blender and CAD, of all things.
The open model race this week is ridiculous.
Keep 'em coming. 🔥
nitter.cf/vercel_dev/status/2118…
Hy4 from @TencentHunyuan is on AI Gateway at 50% off through December 31, 2026.
Hy4 adds image understanding alongside stronger agentic capabilities for long-running, real-world productivity tasks.
𝚝𝚎𝚗𝚌𝚎𝚗𝚝/𝚑𝚢𝟺
vercel.com/changelog/tencent…
Every robot demo is the best robot demo. Funny how that works.
RoboColosseum puts 100+ real robots in 140+ messy real-world scenes, gives researchers 10,000 free eval hours, and publishes every rollout, failures included.
Early scoring gives partial credit, so "almost did it" can look close to "did it."
Still, a lab with 3 robots now gets to test like a lab with 300.
Show me the fails. 🤖
nitter.cf/RoboColosseum/status/2…
Introducing RoboColosseum
An open robot eval initiative by @huggingface @BitRobotNetwork @eaic_sg @DJiafei
Access:
- 100+ robots (@UnitreeRobotics G1s, @LeRobotHF SO-101s, YAMs, Franka and more)
- 140+ in-the-wild scenes
- Open rollout data
- 10,000 FREE eval hrs
Thread ↓
Kimi said open weights by Nov 4. It's Nov 4. Weights are on Hugging Face.
2.8T params and ~1.56 TB of weights means you're not running this on your gaming rig.
But any host can now serve it and compete on price, quants will shrink it, and companies with sensitive data can run it in their own racks.
Promises kept. 🔓
Releasing the model weights of Kimi K3.1 Pro: huggingface.co/moonshotai/Ki…
We'll keep advancing open intelligence.
Aaron retweeted
Techno-optimism is now officially an underground counter-culture again – at San Francisco, CA
What if your AI memory didn't belong to one app?
Living Brain is one persistent brain shared across ChatGPT, Claude, Cursor, Codex and your own agents. Write in one, retrieve in any. Basically agents.md that keeps learning.
They claim 92.2% on LoCoMo, 464ms retrieval, and up to 22.6x fewer tokens per question.
The catch? Your whole brain lives somewhere. I want to know where, and whether I can run it myself.
Still, memory that's portable across tools is the right direction.
Own your context. 🧠
nitter.cf/ladyxtel/status/211798…
