Yup Claude Fable does it all. And building software has become an expensive proposition now. Now we wait for the cycle to become affordable and ubiquitous again. Given the trillion dollar valuations that day seems far off. Hope I'm proven wrong.
AI writes code. Then it forgets.
FURNACE learns your codebase + context and keeps that knowledge.
Switch devices? Knowledge travels with you.
New session? It remembers.
Different AI? Same intelligence.
Code that actually belongs in your project 🧵
https://t.co/CLEOdH7prQ
Your Al editor doesn't know your codebase, FURNACE fixes that.
Connect your design docs. Search with intent, not keywords. Generate code that actually fits your architecture. Full feature guide -
https://t.co/MBiVhItNFP
Introducing FURNACE - a MCP server that turns your Al editor from autocomplete into a full development partner.
https://t.co/qeKTCt9lHD
Knowledge graph. Tiered search. AI agents. Syncs with Linear, Jira, Notion.
There’s only one thing I’m jealous of.
Yesterday, I picked up a friend at his house. It was chaos — toys scattered everywhere, two toddlers running wild, remodeling in progress.
I told my friend’s wife:
“I know life feels hard right now, but you’ll look back on today as the best of times.”
If you have kids at home, you’re living the golden years right now. Savor every moment.
This drone becomes a flying manipulator! 🥏
Researchers at the The University of Tokyo developed this aerial robot.
Built with four pairs of ducted fans linked by actuated joints, Dragon can reshape itself mid-flight.
This allows it to grasp objects and perform tasks typically reserved for ground-based manipulators. Each segment has dual rotors, and its navigation stack calculates the most efficient shape for each object.
Total payload? More than 3 kilograms.
P.S. To increase Dragon's battery life, they consider allowing it to walk on the ground.
Holy shit... Tencent researchers just killed fine-tuning AND reinforcement learning in one shot 😳
They call it Training-Free GRPO (Group Relative Policy Optimization).
Instead of updating weights, the model literally learns from 'its own experiences' like an evolving memory that refines how it thinks without ever touching parameters.
Here’s what’s wild:
- No fine-tuning. No gradients.
- Uses only 100 examples.
- Outperforms $10,000+ RL setups.
- Total cost? $18.
It introspects its own rollouts, extracts what worked, and stores that as “semantic advantage” a natural language form of reinforcement.
LLMs are basically teaching themselves 'how' to think, not just 'what' to output.
This could make traditional RL and fine-tuning obsolete.
We’re entering the “training-free” era of AI optimization.
Holy shit... Google just killed the cloud AI monopoly 🤯
They dropped 'Coral NPU' an open-source chip architecture that runs LLMs on WEARABLES with milliwatt power consumption.
512 GOPS while sipping battery like a smartwatch notification.
This isn't another incremental improvement. It's architectural surgery on how edge AI works.
While everyone's burning billions on cloud compute, Google reversed the entire chip design philosophy:
→ ML matrix engine FIRST, not bolted on
→ RISC-V based, fully open-source
→ Hardware-enforced privacy (no cloud = no leaks)
→ Works with TensorFlow, JAX, PyTorch out of the box
The kicker? It's the first low-power NPU designed for transformers on wearables.
Your AR glasses will run Gemini locally. Your earbuds will translate conversations in real-time. Your smartwatch will understand context without touching the cloud.
Synaptics already shipping production chips with this architecture.
The age of cloud-dependent AI just got an expiration date.