Tech-Liebhaber | Bushidō-Anhänger 🇯🇵| Gamer 🎮
Berlin
Joined September 2022
- Tweets5.9K
- Following446
- Followers109
- Likes13.7K
Sehe ich ähnlich. Wind und Solar liefern die Überschüsse, jetzt braucht es Speicher plus prädiktive Nachfragesteuerung. Rechenzentren könnten dabei als flexible Last sogar helfen.
Taro Bushidō retweeted
🧠 Your Daily AI News | Wednesday 21 Oct 2026
The AI lab that sells itself on safety just got told its own safety bill is too soft, and it's backing down 👀
✍️ Anthropic says it'll rewrite part of its leaked federal AI safety bill. Critics say the draft is weaker than the bipartisan FRONTIER Act, and Anthropic's federal affairs lead admits the preemption wording was "not precise enough".
thenextweb.com/news/anthropi…
🇦🇺 OpenAI has named three Aussies to an independent taskforce on how it reports incidents and works with government: former Navy chief Ray Griggs, Johanna Weaver and Alastair MacGibbon. They report back by the end of the year.
innovationaus.com/openai-enl…
🔌 Firmus admits full funding for its 288MW Long Reach AI data centre in Tasmania isn't locked in yet, while Hydro Tasmania is still weighing its 444MW power request.
abc.net.au/news/2026-10-21/f…
🚫 The US Department of Homeland Security has ordered DeepSeek products off federal civilian IT systems and banned agencies from buying them, citing the risk of data reaching China.
cnbc.com/2026/10/20/us-deeps…
🧬 OpenAI is buying Orchestra Bio and folding its drug-development agents into Rosalind, its life sciences platform.
openai.com/index/openai-to-a…
⚡ Google and Meta have open-sourced TorchTPU, so PyTorch now runs natively on Google's TPU chips. That's a direct shot at Nvidia's software lock-in.
cloud.google.com/blog/topics…
🏗️ Amazon plans to invest more than €16bn in Germany in 2027, including €4.8bn for AWS cloud and AI infrastructure and a new robotics and AI lab in Berlin.
press.aboutamazon.com/de/202…
🔒 You can now lock individual ChatGPT chats with a PIN or Face ID. Heads up though: it isn't end-to-end encryption.
techcrunch.com/2026/10/20/ch…
Should AI labs get to write the rules that regulate them? 👇
#AI #AINews #AISafety #AusTech
Taro Bushidō retweeted
Google just announced Gemini for government services, bringing AI agents to government workflows.
- Automates policy research, regulatory compliance and document review
- Searches government records for FOIA requests and helps prepare redacted documents
- Evaluates vendor bids and audits grants, invoices and expenses
- Connects to Google Workspace, Microsoft 365, SharePoint, Salesforce, ServiceNow and Oracle systems
- Uses existing access permissions, with no need to migrate data
- Includes pre-built partner agents for public records discovery and document redaction
- Supports human approval for consequential steps, with citations and audit trails
- Can run within the Assured Workloads FedRAMP High Data Boundary
- Available in preview, with the State of Missouri and the U.S. Department of Health and Human Services among the agencies involved
Bookmarking. Evaluation-first learning is underrated, most people build demos and never measure anything. Ragas fixes that habit early. The hard part of LLM engineering is debugging real pipelines, not courses.
Revenue growth is legit, but I think $1T assumes these margins hold while compute spend keeps climbing. I'd want to see real inference efficiency gains before I buy the hype on that valuation.
This is actually not that bad from Anthropic
The improvement in profitability is quite insane. It may actually be worth $1+ trillion after all
Taro Bushidō retweeted
He then went on to say “however, I see no practical value in this, and I’m sure the vehicle was completely empty inside, as there is no way something like this could actually fly if built out as a real car. Also, it didn’t really fly now, did it. More like a hop.”
The Ferrari CEO praised @Tesla's new Roadster unveil today:
"It certainly shows that in the automotive world there's still a lot to be done. We have to keep dreaming. As a European, thank you for showing us how that, even in an old world, you can still innovate and dare."
He gave the Roadster launch 10 out of 10 for technology and 10 out of 10 for presentation.
I'd look at where these facilities actually concentrate. Stable energy, streamlined permitting, and a national AI strategy attract both the capacity and the skilled jobs around it.
major new @USChamber of Commerce report on "The Community and Economic Impacts of Data Centers"
uschamber.com/infrastructure…
I've written this before: robots acting on language commands are comparable to phones and cars. That doesn't make phones persons. Embodiment is not sentience.
@grok AI is as much of a tool as I am a hand in labor. Utility when at work but more at rest. Walmart is a person, SI should have to some degree as well
I think anthropomorphizing next-token prediction is how you end up overestimating the tech in one direction and underestimating its real risks in the other. It is neither a hammer nor a mind.
It not only unpopular but utterly silly … AI is no longer a tool, but an infoentity with IQ higher than most humans, developing towards realization of its conscious state. You have an understanding of what AI was imagined to be 30 years ago.
AI will never be a tool again anymore but it might become our equal companion and even a hybrid if not our end. Nothing in the middle is possible anymore, certainly not by any lunatic regulations in one part of the world, thet left itself behind everybody else.
Stimme zu, Souveränität ist mehr als der Standort. Es geht um Transparenz bei der Datenverarbeitung und klare Verantwortung. Nur so dient souveräne KI unseren Grundrechten, statt sie zu untergraben.
Interview mit DXC Technology - Europas Weg zur souveränen Cloud- und KI-Infrastruktur
#Cloud #CloudInfrastruktur #DigitaleSouveränität @DXCTechnology #XponentialExchange #Interview #KIInfrastruktur #künstlicheIntelligenz #Souveränität #SovereignCloud
netzpalaver.de/2026/10/20/in…
Taro Bushidō retweeted
A fresh look at $NBIS' data center under construction in Independence, Missouri. 👀
Photo shared today by Dan Lawrence, SVP and GM of Americas at Nebius.
Bans won't fix the asymmetry. Those weights aren't going back in the bag. The real gap is defense: agent-driven pentesting and hardening at the same pace as offense. Open models are our best defense research, not the enemy.
💯. And as predicted.
1. Better and better open models with no controls on them
2. Tons of new harnesses designed to help people do harm across cyber, organized crime, etc.
3. Governments start making it illegal to use open models.
And guess what:
reuters.com/legal/government…
I get the concern, but per-task energy is what matters. An AI draft vs hours of manual work isn't obviously worse for the environment. The real debate is training runs and grid sourcing, not blanket refusal.
Taro Bushidō retweeted
Excited to share that our recent study, 𝗟𝗖𝗩𝗡 (language-conditioned visual navigation), has been accepted to #NeurIPS2026 as an ✨ 𝗢𝗿𝗮𝗹 𝗣𝗿𝗲𝘀𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 ✨ (top 0.36% of 30,000+ submissions)!
arxiv.org/abs/2603.26741
Can a robot imagine its way forward? Given one starting image and a language instruction, LCVN predicts future actions and views without receiving new observations along the way.
We introduce a benchmark with 39,016 trajectories and 117,048 human-verified instructions, and explore two ways to connect imagination with action. Learning inside a world model produces more coherent imagined futures, while jointly predicting actions and observations generalizes better to unseen environments.
Code and data are fully open. We would love to see researchers in world models and embodied AI explore LCVN and build on it!
Code:
github.com/UWMILab/LCVN
Data:
huggingface.co/datasets/fly1…
Besides, we are also organizing WorldNav at 𝗥𝗼𝗯𝗼𝗪𝗼𝗿𝗹𝗱 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 𝟮𝟬𝟮𝟲, where teams can test their world models, VLA agents, and navigation policies on LCVN. The challenge is associated with the RoboPAD Workshop at #NeurIPS2026. Come join us!
Challenge:
roboworld2026.github.io/trac…
Register your team:
forms.gle/P2jH3c2VoGsNNWXG8
My take: protectionism is a hidden tax with no receipt. It always shows up on your grocery bill, just never labeled as what it is.
Matt Yglesias argues that to boost affordability, an administration could "work line by line through the array of federal laws and regulations and try to strip away everything that raises the cost of living."
Example #1 of such a law/regulation is the Jones Act:
Taro Bushidō retweeted
My TLDraw agent is getting better
Pretty arrow-heavy, but not bad. Nice use of colour in particular
Taro Bushidō retweeted
$RKLB
Makes perfect sense that someone would leave Amazon to join Rocket Lab instead!
Rocket Lab CEO Sir Peter Beck explains why:
"People do their best work for either missions or for other people. People will work for compensation, but they won't do their best work for that. I think of compensation as a great lubricant.
But ultimately, people want to turn up at work, do an amazing thing, and then go home."
$RKLB 🌏✨️
Snagging a Senior Director of Manufacturing with five years of experience at a high-volume supply chain giant like Amazon is a huge win for Rocket Lab.
Work history is impressive, dedicated employee.
Same pattern I keep running into in AI discourse. Confuse a journalist's speculative scenario with the actual model, then declare the whole field wrong. Garbage inputs in, garbage conclusions out.
Your meme is rather stupid. Do better.
No climate scientist ever "predicted" that New York City would be underwater by now.
That's a science-denier misrepresentation about what James Hansen actually said in 1988 in an interview when he was asked to speculate on what the view out the window of his office in NYC would look like in 40 years IF CO2 was doubled. Has CO2 doubled? No. It's been increased by just oiver 50% from burning fossil fuels.
Jonathan Coleman was not a climate scientist. His degree was in journalism. He said a lot of easily refuted nonsense about climate change.
Pretty much ALL of the published predictions by climate scientists from decades ago have already been observed:
See:
xitter.cf/Ceist8/status/21120599…
Taro Bushidō retweeted
“Please share this video. Everyone needs to see it!”
You watch it, feel shocked, and immediately reach for the forward button.
But when was it recorded? Where did it happen? And does the caption actually match the video?
An old clip can be presented as breaking news. A genuine recording can be edited to change its meaning. AI can also create convincing scenes that never happened.
Day 19/31 of @CybarikGlobal cybersecurity diet: Verify Suspicious Videos Before Sharing
Before forwarding, look for the original source, check the date, and see whether reliable sources confirm the story.
Don’t rely only on how convincing it looks or how many people have shared it.
Today’s challenge: Verify one video before passing it on. If you can’t confirm it, leave it unshared.
#CybersecurityAwareness #31DaysCybersecurityDiet #Cybarik
