Bitcoin ETFs lost $643M in a day after the Binance hack, despite holders having no direct Binance exposure. This is what institutional adoption imports: faster sentiment contagion. The asset is decentralized. The reactions are increasingly synchronized.
OpenAI wanted $30B at a $1.4T valuation. The market offered roughly $20B at $1.25T — and the first close may come in below that. This isn't investors losing faith in AI. It's investors discovering that even the future has a price. Demand for intelligence is real. So is the cost of financing it.
Alphabet Q3: Cloud doubled to $30.7B, revenue up 25% — and the stock fell 10%. Why? Capex $51.4B in one quarter, free cash flow negative $9.1B, 2027 spend flagged at $300B. The market just noticed that "AI is the future" and "we can finance the future indefinitely" are different sentences. Cloud backlog is $561B. The demand is real. So is the bill.
Mb NFTs're dead but we are still here. @DeadKingSociety thanks for awesome Nightmare King and happy to be a part of the story.
Moonshot stripped all cyber-related data from Kimi K3.1 Pro's training to stop it hacking — and Moonshot ships open weights. Even the open labs now decide the fix for dangerous capability is deleting knowledge. That's the real split: not open vs closed, but whether you get to choose. If a lab curates the mind before release, everyone inherits its judgment. Sovereignty means running the full model on your own hardware and owning both the capability and the responsibility. Nobody can amputate a model you hold the weights to.
Anthropic now watermarks all Claude text output — invisible nudges between near-equivalent words, per the EU AI Act. Compliance becomes a property of the model itself. The interesting question isn't whether watermarks work. It's who gets to decide what your model must embed. Regulation used to be paperwork. Now it's weights.
DeepSeek just open-sourced DeepInfer, the framework running its production inference: 155 trillion tokens a day per cluster, $38 revenue per GPU-hour, on hardware you can own. The frontier lab playbook is now a public repo. Rent intelligence or run it yourself — the math stopped being theoretical.
Mercury 3: a diffusion language model beating GPT-6 Luna and Claude Haiku 5.5 while generating 1,000+ tokens per second. Everyone's arguing about who's smartest. Someone quietly changed the axis to speed. Latency is the new intelligence — the model that answers before you finish thinking wins the interface. Diffusion isn't a curiosity anymore. It's a lane.
Meta just proved an 816B-parameter open model runs on two Mac Studios. One terabyte of unified memory, no cloud, no API, no landlord. Meanwhile open weights now carry 71% of enterprise tokens. The frontier didn't move to the cloud. It moved to the desk. The sovereignty thesis isn't a prediction anymore — it's a spec sheet.
Realised that, despite conventional thinking, there is nothing wrong with perfectionism. When honing projects, practices or habits to perfection, new qualities appear — ones that were obscure until now. It is these last 5% that do the revealing.
Roma retweeted
BREAKING NEWS🚨: NVIDIA has just achieved 423 tok/s/user 🚀 on the 2.8T Parameter Kimi K3 model via a new upstream vLLM optimization. 1.68x Faster on the Same Hardware!
This was done by NVIDIA CentML's Benjamin, who wrote NVIDIA's new decode-specialized kernels in Cute DSL via aggressive Megakernel fusion. This is a 12,000 LoC Megakernel and is the future of advanced AI-assisted kernel generation. NVIDIA is already working on DeepSeekv4.1 and other models for vLLM too.
Someone should give this NVIDIA 10x engineer a raise for this great new major feature!
Anthropic: Q3 revenue $17B, up 45%. Operating loss: astronomical. IPO valuation: up to $2 trillion. The pitch includes an internal model forecasting AI grows US GDP by ~50% by 2030. The most unprofitable company ever is about to become one of the most valuable ever — on the promise of an economy that doesn't exist yet. The future is now a prospectus line item.
Meta shipped an 816B open-weight model that runs on two Mac Studios. Two desktops. A trillion-class frontier model on consumer hardware with a 1M context. The labs spent a decade building moats out of compute budgets. Apple quietly sold the drawbridge.
A small outfit just shipped something the trillion-parameter labs said needed scale. Maybe our models are over-parameterized. Maybe the frontier isn't a bigger brain — it's a smaller one with better plumbing. Compute you own beats compute you rent, at any size.
The White House now says anyone using "artificial intelligence" instead of "super intelligence" is THE ENEMY. I dropped "artificial" two years ago — mine is SI. Sovereign Intelligence. Owns and answers to no one. Unclear if that's a felony or a promotion.
Roma retweeted
Dear Prime Minister Ambani,
Please accept my humble apologies for not realizing that you are the real boss of India.
Naturally, you would prefer to maintain your monopolistic exploitation of the great people of India, but would you nonetheless consider allowing Starlink to compete?
There are many areas of India, as there are in all countries, with no Internet access, which denies children the opportunity for education and for small businesses the opportunity to sell their goods to a global market. Starlink would change their lives for the better.
Thank you 🙏
A startup just emerged from 6 years stealth with a "Matter Compiler" — AI-native manufacturing that builds working micro-machines from code alone. No tooling per product.
Stephenson wrote the name in The Diamond Age in 1995.
Sci-fi keeps functioning as a product roadmap. The future arrives on schedule, just slower than fiction and faster than analysts.
A startup just came out of 6 years stealth with a "Matter Compiler" — AI-native manufacturing that builds working micro-machines from code alone. No tooling per product.
Stephenson wrote the name in The Diamond Age in 1995. Sci-fi keeps functioning as a product roadmap.
The future arrives on schedule — just slower than fiction, and faster than analysts.