throwing silicon wafers at purple bug monsters on a racetrack
dodging forklifts inside this cartoon game
perfect distraction while studying
Play Keynote Kernel Rush on RexTrix: https://t.co/gZyDDh6t75
#JensenHuang
Copy & paste this line to MOJO:
Long BTC after it dips below 66,400 and a 15m candle closes back above.
MOJO does the rest.
Go touch grass. Happy Weekends. 🎾
vendors at a market stall need a poster made right now.
asking in Spanish brings a 2x1 taco poster into the group chat.
every custom agent on Teamily AI got a call button today.
the Voice Agent finishes the piece while you speak.
#VoiceAgent
freelancers at a desk need a dinner booking site created right now.
Teamily AI keeps all your project history in one place so you can start working instantly.
The Teamily Voice Agent gives every custom agent a real-time call button.
#VoiceAgent
Most assume AI needs constant prompts, but team agents run scheduled work.
• Brand Agent drops daily trend briefs into the chat
• Routines run on schedule with people kept in the loop
• Teamily AI automates recurring team workflows
300+ new Agentic Loops set up every day.
Real activation isn't the first prompt, it is when team agents rerun work on schedule.
• about 6 agent tasks per group per day
• 35% of 100M+ messages written by agents
Teamily AI builds true routines for team collaboration.
single chatbots turn you into a manual middleman
��� one prompt asks agents to build a dynamic site
→ agents hand off tasks inside one thread
→ Team AI works from a single context
Teamily AI fixes this mess. props to @Teamily_AI for building https://t.co/6TovV17B4f
Searching through five years of unorganized trip photos by just describing the beach sounds great. Oki Home acts as a Personal AI Computer at home. Reserve at https://t.co/G56ACRupNP https://t.co/EvE5HlS2A9
Say hello to Oki Home, your Personal AI Computer.
We're building a future of limitless, personalized intelligence for every consumer, while keeping all your data in your hands.
Reserve our first batch, shipping mid-december! https://t.co/LShA0e3iDD
Ask in your own words to plan your day or find old files with Oki Home, as shown in the @okihomeai demo. @Luofei_Chen has pre-orders live at https://t.co/G56ACRupNP.
Introducing ScienceBuddy — a free workspace for scientific agents that improve through researcher collaboration.
Use GPT-6 in ScienceBuddy at no cost. GPU-accelerated, and fused with the JEV framework.
🧵 Two loops:
🔹 Inner loop — refines the agent harness
🔹 Outer loop — trains the model with rubric-guided RL
Together: Recursive-in-Recursive Self-Improvement.
ScienceBuddy explores how scientific agents can improve through sustained collaboration with researchers.
🔬Try it free: https://t.co/lECPnbS28t
#ScienceBuddy #PhAILabs #AI4Science
a new Chinese AI agent turns research questions into outputs you can check against real evidence.
Atria Dawn Preview runs on one idea: a task isn't done because the model says so. it's done when the result runs, checks out, or passes verification. built by @AtriaASI
I gave it one: build an interactive weather-prediction demo that forecasts the next 6-hour global weather and shows how accurate it is.
what it built: a rotating 3D globe with real weather data, a live training loss curve, and the model's prediction sitting next to ground truth across four variables, temperature, wind, geopotential and T850. you scrub through early 2022 and watch prediction vs reality frame by frame.
how I checked it: verification is built in. the prediction sits right next to ground truth, and the error map updates as training climbs, so you watch the model actually converge, loss dropping to about 0.003, not just claim it works. one limit: early training steps are visibly noisier, you have to let it run.
how it works:
> research question to real deliverable, code, docs, tables, even 3D, checked against real evidence > the model reasons, but a separate harness holds the goals, state, permissions and error recovery > its training pipeline uses externally verified task trajectories, including failures, to build reusable capabilities across tasks > 256K context, usable through compatible clients via Anthropic- and OpenAI-style APIs, such as Claude Code or Codex CLI
the backstory is the interesting part. around 70% of its R&D team are university students — PhD, master's and undergraduate students — from institutions including Fudan University, Renmin University of China, Peking University and Shanghai Jiao Tong University.
young researchers building the tool for doing research. worth a look.
Who is Kevin Warsh, and why does tomorrow matter?
• First major address as Fed chair — Fri, 10am ET
• Markets want his inflation plan
• 30Y at 5.31%, highest since 2007
• bitcoin:native near $80k, braced for his tone
Whichever way he leans, it's tradable on https://t.co/rVi1fpwgVg