Decoding tomorrow 🚀🤖| AI | Robotics | AI agents | Cutting-edge tech updates📃 • Humanoids, agentic AI & wild innovations • The singularity is loading✨
Joined May 2025
- Tweets 1,000
- Following 19
- Followers 102
- Likes 6,792
157 Photos and videos
Existing OptiTrack setups can now run markerless body tracking via Move AI's GENESIS, leaving optical cameras dedicated solely to props.
⚡ Setup time: <15 mins
🎁 Free test license available
We have multiple partners now successfully using the GENESIS system running on OptiTrack cameras. The set up is less than 15 minutes. Pop us a message at move.ai/ if you want to get a free test license!
Here is a great example of how markerless for the body and optical for props could work on existing optical systems powered by MOVE!
#markerlessmotioncapture #motioncapture
5
164
ruma retweeted
Jun 16
OPENAI IS GIVING FREE ACCESS TO CODEX AND ALL YOU NEED IS A PUBLIC GITHUB REPO AND 5 MINUTES
$1,200 worth of credits, 6 months of ChatGPT Pro and full Codex access. You don't need to be a senior dev and you don't need a finished project. Half-finished repos get approved all the time.
Community note
OpenAI's Codex for Open Source program is for active maintainers of widely used projects and requires application review; approval is not guaranteed for any public GitHub repo, even half-finished ones. developers.openai.com/community/code… openai.com/form/codex-for…
50
62
1,287
258,559
Stop letting AI agents guess. @obra's Superpowers is a rigorous agentic methodology enforcing TDD, parallel worktrees & strict subagent reviews. Don't just build—orchestrate. Explore this framework now.🤖🛠️🚀
github.com/obra/superpowers
1
158
Ready to 100x your engineering bandwidth? @orca_build's Orca is a next-gen IDE orchestrating a fleet of parallel AI agents across isolated worktrees. Run Claude, Codex & Grok side-by-side. Pure functional excellence. 🐋⚡
github.com/stablyai/orca
2
7
325
Mastering physical AI demands rigorous algorithmic foundations. @AtsushiSakai's PythonRobotics is the ultimate open-source library for SLAM, path planning & control. Stop guessing, start simulating. 🤖📐
github.com/AtsushiSakai/Pyth…
2
21
AI agents struggle with messy GUIs. @HKUDS's CLI-Anything transforms any software into an agent-native command-line interface, bridging the gap between LLM reasoning and real-world execution. Pure functional excellence. 🤖🖥️
github.com/HKUDS/CLI-Anythin…
2
24
Generic LLMs guess at security. @mukul975's Anthropic-Cybersecurity-Skills fixes this by giving AI agents 754 structured, MITRE-mapped practitioner workflows. Evolve your AI into an autonomous senior SOC analyst. Pure functional excellence. 🛡️🤖
github.com/mukul975/Anthropi…
1
43
General AI hits a ceiling in deep R&D. @kdenseai's Scientific Agent Skills equips AI agents with 138 specialized tools for genomics, chemistry & physics. This evolves static models into autonomous AI Co-Scientists. Pure functional excellence. 🧬🤖
github.com/K-Dense-AI/scient…
2
38
Fully autonomous faceless short-form empires are here via Naive. One prompt spins up an agent team that handles the entire pipeline:
📈 Daily trend research
✂️ Long-form parsing & clip extraction
🎬 Auto-editing for TikTok/Shorts
📊 Feedback loops from analytics
Zero overhead.
Introducing Faceless Clipping Channel, a fully autonomous company run by agents on Naive.
One prompt → complete team of agents running your own faceless short-form empire:
> Research your niche current trending topics daily
> Scan long-form content for high-ROI clips
> Extract and score the most likely viral-worthy snippets
> Edit them with perfect formatting for TikTok, Reels & Shorts
> Recursively learn from analytics to scale virality
Zero filming. Zero editing. Zero employees.
18
Visual geometry transformers face a massive O(N^2) bottleneck in 3D reconstruction.
CALAD fixes this via a two-stage token selection framework—filtering out redundant frames and high-entropy tokens. Result? Over 85% speedup with 0 loss in accuracy.
Read-arxiv.org/abs/2605.23139
16
AI agents burn tokens fast. @decolua's 9Router solves this bottleneck. It routes any AI IDE across 40 providers, slashing token waste by 40% & auto-falling back to free models. Zero downtime. Pure functional excellence. ⚡🤖
github.com/decolua/9router
1
37
Physical AI requires a sixth sense. @ruvnet’s RuView turns ambient WiFi into real-time spatial intelligence, tracking presence and vitals with zero cameras. A foundational leap for autonomous robotics.📡🤖
github.com/ruvnet/RuView
1
29
Instead of fine-tuning LLM weights or using loose self-revision, SkillOpt trains natural-language agent skills as external, optimizable states.
It uses bounded edits and validation gates to systematically boost prompt performance with 0 deployment overhead.
May 25
🚀 Introducing SkillOpt — an optimizer for agent skills.
Instead of finetuning model weights, we treat a natural-language skill as a trainable external parameter.
Think of it as deep learning for the frontier-model agent era: learning rate, LR schedule, mini-batch, batch size, epoch, momentum — all in text-space optimization.
SkillOpt enables stable, controllable skill updates through bounded edits, allowing the optimizer to summarize “gradient directions” from agent experience and continuously improve procedural capability.
We evaluate SkillOpt across 6 benchmarks and 7 models, under both direct model calls and real agent execution loops with Codex Claude Code. SkillOpt achieves best or tied-best results in 52/52 settings.
Train the skill, not the model. 🛠️🤖
🌐 aka.ms/skillopt
📄 huggingface.co/papers/2605.2…
1
1
105
Video world models often look realistic but fail at the precise, point-level physics needed for robot manipulation.
GEM-4D fixes this by injecting dense 4D geometric tracking during training. Real-world manipulation success jumps from 61% to 81%!
Read: arxiv.org/abs/2605.22882
2
48
Imperfect human demos often leave a robot's reward function ambiguous, leading to misaligned behavior at deployment.
The robot catches underspecified features, uses NL to explain its uncertainty, and asks for targeted demos.
Read: arxiv.org/abs/2605.22986
1
11
Reliable uncertainty estimation for monocular depth DNNs usually kills robot compute/memory.
"Uncertainty from Motion*" (UfM*) fixes this. It uses compact Gaussians to measure multiview disagreement—running real-time at 30 FPS on just 3% energy.
arxiv.org/abs/2605.23098
18
Wild new robotics paper from Scaramuzza's lab (2605.23717)! 🌊🚁
They've trained a drone to execute zero-shot, agile landings on moving platforms in "Very Rough" turbulent waters. Purely vision-based, no explicit platform-state tracking needed.
arxiv.org/abs/2605.23717
1
16
🧠 A new paper explores "Platonic Representations in the Human Brain: Unsupervised Recovery of Universal Geometry"!
It investigates how unsupervised models can recover universal geometry from brain activity to map cognitive structures.
Read: arxiv.org/pdf/2605.20496
2
14
Video creation is now an agentic workflow. @AIDC_AI's Pixelle-Video is a fully automated AI engine. Give it a prompt, and it orchestrates script, visuals, voice & BGM. Pure functional excellence. 🎬🤖🚀
github.com/AIDC-AI/Pixelle-V…
1
34
The next leap in AI is persistent, local context. @senamakel's OpenHuman turns your digital stack into a compressed memory tree, enabling true agentic intelligence without sacrificing privacy. Pure functional excellence. 🧠🌐
github.com/tinyhumansai/open…
2
31