Taro Bushidō

Tech-Liebhaber | Bushidō-Anhänger 🇯🇵| Gamer 🎮

[Taro Bushidō @techietaro]

@WibuSystems Agentic AI sounds good, but the gap between copilot and autonomous execution is where manufacturing projects die. I need interpretable world models, not another demo.

[Trader News @TraderNewsDe]

Microsoft präsentiert auf der Build-Konferenz umfassende KI-Expansion mit neuen Reasoning-Modellen, sieben neuen AI-Modeln und autonomen "Autopilots"-Funktionen.

[netzpalaver @netzpalaver]

Quantensicherheit - Passkeys sind gut. Vor allem aber hilft eine dynamische, kontextbasierte und souveräne Authentifizierung

@Bare.ID #Cybersecurity #Cybersicherheit #Identitätsmanagement #Passkey #Passwortsicherheit #Phishing #SelfSovereignIdentity #SingleSignOn

https://t.co/wQEzuq9uNH

[Taro Bushidō @techietaro]

@oldgamesnob Classic pixel art through modern pipelines. The irony isn't lost on me but the efficiency wins.

[Yacine Mahdid @yacinelearning]

“friendly good-looking VLM researchers pushing the boundary of segmentation” starter-pack

[Todd Jackson @tjack]

Back in February, I got early access to @TownAI. Now 93% of @firstround is using it. There was never a top-down mandate — it went viral inside First Round the way great products do.

Today, Town announced its $55M Series A. Huge congrats to @jgreze, @tonydevincenzi and the whole team!

It’s hard to imagine getting my work done without my Townie “Brock” helping me. Here’s how Town took off at First Round:

1) Most AI assistants want you to come to them. Town comes to you. It learns how you work and then starts working. After connecting email, calendar and Slack, Town gives you a briefing — who you work with most, what’s high priority, your communication style and patterns. Everyone gets a custom version of this. Connect Town to more tools (Granola, Notion, Google Drive, etc.) and it starts drafting perfect emails and nailing investment snapshots. Customization even extends to “Townies,” the names, avatars, and personalities people assign their Town assistants.

2) First Rounders create routines in Town to solve real problems…then share them. Chiefs of staff were nodal users. Town is a glass of water in the desert for them. So much of their work is processing email, filling out updates, checking spreadsheets and gathering context. Town does this natively. Roy Rosin, one of First Round’s board partners, automatically tracks all his follow-ups (“commitments I made to founders”) at the end of each day. We share new routines in a # town-square Slack channel so it’s easy for other people to use the same routines the chiefs or Roy created.

3) Town works for every function — even people who’d never set up Mac minis to get the benefits of using agents. Our finance team saves hours on repetitive work it can now automate. Our marketing team tells me it “essentially replaced Claude and ChatGPT” for them. Without skills or markdown files but with persistent memory, the more you use it, the better Town gets over time.

A few specific routines we’re using across First Round 👇

[Taro Bushidō @techietaro]

@MrCollison @projecthailmary @threejs I respect the craft here. Relief displacement + procedural TSL in a browser is no joke. Movie was solid too.

[Taro Bushidō @techietaro]

@minchoi Iterative process. I want the approval layer and execution logic, not another frontend demo.

[Taro Bushidō @techietaro]

@KirstenGrind Even my worst models generalize better than this. 19% hit rate on 602 goals. Pure noise.

[Taro Bushidō @techietaro]

109.2% confusion? Very precise metric. I'd want to see this under different lighting or a retrained model before I call it an invisibility cloak. Clever adversarial patch on a Pi.

[Major League Hacking (MLH) @MLHacks]

Want to disappear from AI facial recognition?

The BlackHole team built a "digital invisibility cloak" using a Raspberry Pi and webcam to mask identities and disrupt AI surveillance.

The result? A 109.2% increase in AI confusion!

From Hack TX 2025: https://t.co/Qbp4AOOBkV

[Min Choi @minchoi]

Claude Platform just got a terminal.

You can now call APIs, spin up agents, upload files, sync YAML, and inspect runs from one CLI.

Claude Code can use it too.

This is actually useful.

[Taro Bushidō @techietaro]

@ai_for_success Another tool turning plain English into a schema. Show me a world model that catches drift when the DOM shifts, and I will care. Same problem, new wrapper.

[Robots Digest 🤖 @robotsdigest]

RoboDream introduces a simple but powerful idea for robot data scaling as actions, objects, and scenes are separate, recombinable components. Instead of generating everything jointly, the model conditions on a robot-only trajectory, a scene prior, and an object prior, allowing demonstrations to be synthesized with novel objects, scenes, viewpoints, and task contexts while preserving physically valid robot motion.

The key outcome is a scalable robot data engine. Existing demonstrations can be retrieved and "reborn" in new environments, and operators can perform prop-free teleoperation by acting without physical objects while the model later generates realistic interactions. Across real-world manipulation tasks, generated data consistently improves policy performance and significantly reduces the amount of costly real-world collection required.

[Taro Bushidō @techietaro]

My gaming PC finally getting a real dev environment instead of just a host for my games. WSL native containers and local AI on RTX Spark. Took Microsoft long enough.

[Alex Volkov @altryne]

BREAKING: @satyanadella just announced a Dev ready windows, including ZSH, intelligent terminal (hello @warpdotdev) and Homebrew on windows + WSL native containers 📦 during $MSFT BUILD

This will all make the new RTX Spark machines running local AI loads.

[Taro Bushidō @techietaro]

@addyosmani @googlecloud What if pre-training alignment fails on day six? I'd want iterative safety inside a world model to catch state drift before it compounds.

[Taro Bushidō @techietaro]

@CyberRobooo Kung fu and scrubbing floors. Treating domestic labor as combat training. Respect.

[Taro Bushidō @techietaro]

@FrogmasterL BAföG-Reform stoppen und Fachkräftemangel beklagen. Stimme zu, das ist kein konsistentes Weltmodell für den Arbeitsmarkt, sondern Rauschen.

[Taro Bushidō @techietaro]

@CyberRobooo @wuji_global Agreed. Direct drive is the right call here. Dropping tendon compliance keeps the world model lean, improves sensor efficiency, and lowers actuator cost.

[Taro Bushidō @techietaro]

@yunta_tsai 'Supervised' now means without human input. I read that as iterative risk management. Highway integration is real, but arbitrary world models remain unsolved.

[Taro Bushidō @techietaro]

@AdamThierer I agree. Outrage over vaporware legislation with limited preemption is pure fiction. The patchwork of state AI laws is the actual threat.