WAIT. This is actually insane. Client discovery call ended at 3:15 on Tuesday. Working prototype was in their inbox at 3:32. They DMed me asking how. The honest answer is I did not do anything except watch Lovable do it. Granola transcribed the call. Lovable read the transcript. The Connectors layer did the rest. Full breakdown of the workflow ↓

Before Claude Fable 5 got banned, I turned all my fine-tuning research and experiments into a product: Finetuner.dev (A CLI for data generation to fine-tune AI models). Finetuner uses Codex 5.5 or Opus 4.8 as the orchestrator and Chinese models (DeepSeek v4 Pro, Kimi K2.7, MiMo v2.5, etc.) to generate dataset rows. This system is unique because these are not Python-script-paraphrased datasets. Each and every row is handcrafted by Chinese models based on the orchestrator model's instructions. Now anyone can generate datasets to fine-tune small language models (1B to 30B models). I achieved 10x lower costs and 5x better dataset quality. Releasing the product in a few days. (I'll open-source the skills.)

You can now run a GPT-5.5-level model on your own computer for free. NVIDIA released Nemotron 3 Ultra as a full open model. And if you still need an API, it gets even cheaper. Atomic's tested it against GPT-5.5 on the same task: 3 HTML5 physics demos from scratch. GPT-5.5: 11k tokens, $0.57. Nemotron 3 Ultra: 11.3k tokens, $0.051. The output of Nemotron was better. The price was 10x lower.

The next AI race won't be about size. It will be about precision. Heidi is already running that race.

Robotics used to depend on rigid rules and controlled conditions. Daniel Rausch explains why AI is helping robots become more flexible, adaptable, and capable in the real world. 🤖 #Robotics #AI #Automation

One of the biggest unlocks for scaling robot learning won’t just come from bigger models or better hardware. It will come from building real infrastructure for high-quality, diverse robot demonstration data at scale, collected across many environments and by many operators. COBALT is doing exactly that: turning everyday smartphones into a global, cloud-native teleoperation system that makes it possible to gather rich robot data without expensive setups or closed labs. This kind of accessible, parallel data collection could become foundational infrastructure for the next phase of robotics. cobalt-teleop.github.io

Check out this Special Issue on Artificial Intelligence and Machine Learning for Smart, High Performance, and Resilient Buildings and Cities. asmedigitalcollection.asme.o… #ai #machinelearning #sustainablecities

Our 4B model beats Gemini 2.5 Flash on adversarial legal reasoning. 100% accuracy vs 30%. Not because it's smarter, because it follows a knowledge graph instead of vibes. Here's the full story 🧵

$5,280 a year on AI subscriptions. Claude Max, ChatGPT Pro, Gemini, Cursor. $440 every month going to the cloud. a mini PC for $1,700 does the same thing on your desk. GMKtec EVO-X2 - a box the size of a hardcover book. inside it the chip that Lisa Su personally brought on stage at CES and signed with her name. 128GB of memory in one pool. runs models that cloud subscriptions behind $200 paywalls throttle during peak hours. here there are no limits - your hardware, your rules. NVIDIA's RTX 5080 at $1,000 lost to this box by 3x on a real AI workload. a discrete GPU beaten by a mini PC the size of a lunchbox. electricity - $9 a month. pays for itself in 10 months. after that $440 every month stays in your pocket. the cloud made sense when nothing on a desk could compete. now it can.

Anthropic CEO Dario Amodei: "it is absolutely wild that people are talking about the same tired political issues, when we are near the end of the AI exponential" here are three things that should stop you cold: one: at Anthropic, models already do "100% of today's coding tasks" two: a country of geniuses in a datacenter, 90% likely within 10 years three: he won't even buy more compute, because AGI is already this close meanwhile the world is busy with memes, elections, celebrity beef the most important shift of your lifetime is happening quietly, and the gap is already opening the ones who start building with these tools now end up on the other side of it that's exactly why I put together a step-by-step guide on building your own AI agent agents are the part of AI moving fastest right now, the full walkthrough is in the article below

A year before Claude Code hit a ~$2.5B run rate, its creator Boris Cherny sat down at Bessemer to explain where it came from It began as an internal experiment not something meant to be a product It runs on the same public Anthropic API and models anyone can use The philosophy: don't bet against the model, and don't trade away capability And the multi-agent workflow grew out of Anthropic engineers wiring up fleets of role-based agents internally ~56 min origin story, free ↓

MCP solved tool calling. A2A solved coordination. What solves transport? venturebeat.com/ai/mcp-solve…

“You can offload a task, or even a job, but you can never offload your learning” Also said brilliantly by @yacineMTB “you can offload thinking but you can’t offload understanding”

. AI could become one of the world's largest new sources of electricity demand The chart highlights how different AI electricity demand could be from traditional data center growth. Conventional ("general purpose") data center demand rises steadily, reaching around 1,300–1,400 TWh per year by the 2050s. AI-related electricity demand starts from a very small base, but grows much faster. By 2030, AI electricity demand is projected to be roughly ten times larger than today. By 2050, AI data centers alone could consume around 3,000 TWh annually, already exceeding the electricity demand of conventional data centers. By 2060, AI-related demand could approach 5,000 TWh per year, nearly four times larger than general-purpose data center demand. North America remains the largest contributor throughout the period, while Europe and Greater China also see significant growth. By the 2050s and 2060s, the rest of the world becomes an increasingly important source of AI-related electricity demand. Source: @wef

Built on the shoulders of giants this weekend, stack credits: @huggingface (Space hackathon) @PollenRobotics (Reachy Mini) @modal (LLM TTS endpoints) @livekit (WebRTC SFU) @Alibaba_Qwen (Qwen3 TTS) @NVIDIAAIDev (Nemotron) @threejs (3D Reachy twin) @suno (Reachy Radio)

Patent researcher explains Anthropic’s first agentic AI patent: "Claude is not just being built to answer prompts. it is being built to use computers like a human." in 27-minutes, he breaks down Anthropic’s first US patent: 234 pages, 174 figures, Google Brain → Adept → Amazon → Anthropic. The patent is about multimodal agents that understand image text interfaces and automate software actions. Worth more than a $499 agentic AI course. Watch today. Then you’ll understand why Claude is quietly moving from chatbot to computer operator ↓

Fantastic findings and thread. A single biological neuron performs an enormous amount of computation, orders of magnitude more than that performed by an artificial neuron in a neural net. Parameter count in an LLM is not the same as number of synapse in a brain, as each of the latter packs in so much more computation.

Moonshot skipped DeepSWE. An independent KernelBench run found regressions. Practitioners building model routers are asking why. venturebeat.com/ai/open-sour…

INSIGHT: The AI supply chain is becoming the new global power map. And it’s starting to matter as much as traditional energy👇 For decades, global power was shaped by who controlled energy. Today, another strategic layer is emerging: AI infrastructure. The stack looks like this: 1) Energy (power generation) 2) Chips 3) Data centers 4) AI models 5) Applications And the key insight is simple: The bottlenecks are no longer just software. They are chips and electricity. That’s why companies like NVIDIA and TSMC sit at the center of the system. And why hyperscalers are racing to secure energy, compute, and data center capacity at unprecedented scale. In this new structure, the biggest winners won’t just be app builders. They’ll be the ones controlling the physical backbone of AI. Because in the AI era, power isn’t just political or economic. It’s computational. Follow and turn on notifications to stay up to date.

THESE 5 SKILLS WILL SOLVE ANY OF YOUR PROBLEMS IN AI businesses don’t need complex automations, they need a solution that will solve their problem These skills will help properly shape a request and a plan for building a project or automating a process Their task is not to complicate the work but to make it as simple and effective as possible > Skill Creator > Superpowers > Context Mode > ClaudeMem > Frontend Design (or a similar UI/design skill) Save this so you don’t forget